Brief Reports

Approximately 1% to 2% of inpatient stays result in discharges against medical advice (AMA).[1] Though relatively infrequent, AMA discharges warrant attention as they are associated with higher morbidity, increased risk of readmission, and greater 30‐day mortality.[2] A recent study found a 30‐day readmission rate among AMA patients of 24.5%, nearly twice that of matched non‐AMA patients, and a 30‐day mortality rate of 1.3%, also nearly double that of planned discharges.[3] Discharges AMA may be expected to decrease index length of stay, yet accounting for 30‐day readmissions they are estimated to increase costs 56% higher than expected from an initial hospitalization.[4] Patients note several possible reasons for leaving AMA including family emergencies, dissatisfaction with care, financial concerns, or simply feeling better, among others.[5, 6, 7] Risk factors for AMA discharges include previous AMA discharge, having no primary care physician, younger age, lack of insurance, male sex, substance abuse, and lower socioeconomic status.[4, 6, 7, 8]

A number of prior studies have assessed risk factors for AMA discharges, the long‐ and short‐term outcomes, patient reasons for leaving, and physician perceptions of why patients leave AMA.[3, 5, 7, 9] However, there is limited information about opportunities for discharge transition interventions in this potentially more vulnerable population. Because of the increased short‐term and long‐term risks to these patients, treatment and follow‐up plans at the time of discharge may carry even greater importance than follow‐up plans with standard discharges. This study analyzed AMA documentation and what interventions were carried out at the time of discharge.

METHODS

We reviewed the records of all adult patients, ages 18 years and older, admitted to a university‐affiliated tertiary care hospital in Dayton, Ohio (a 520‐bed hospital with approximately 17,000 adult patient encounters per year) over a 2‐year period, and who subsequently left AMA. A hospital database identified 351 adult AMA cases (1.0% of adult admissions). A single reviewer performed an in‐depth review of the 291 patient admissions to the general medical service between January 1, 2009 and December 31, 2010, and manually reviewed and abstracted the data of interest. The Wright State University institutional review board approved the study.

Documentation review focused on the presence of a specified AMA note, the presence of documentation addressing informed consent, patient decision‐making capacity, patient health literacy, follow‐up plans, whether or not medications were prescribed, and whether or not any warning indicators of impending AMA were apparent. These items represented key elements of the discharge policy and procedure in place at our institution during the period of study. We speculated that nurses may be more immediately available at the time of AMA discharge and thus might carry out AMA documentation more often than physicians. To assess this we recorded the role (nurse vs provider) of the writer of AMA notes. We also assessed patient gender, length of stay, prior AMA, 30‐day emergency department (ED) re‐encounters, and 30‐day hospital readmission after AMA discharge.

Informed consent was deemed present if patients signed the hospital's standardized AMA form. Decision‐making capacity was assessed as present if there was specific mention of the patient's capacity on the day of discharge. Any mention of health literacy or the patient's stated understanding of his medical condition at any time during the hospitalization was considered positive documentation of healthcare literacy. Follow‐up plans included any mention of where and when the patient would return. Discharge medications included prescribed medication or indication that no medications were warranted. Warning indicators included specific mention of the patient's desire to leave AMA. For example, patients who left the unit without informing staff were considered to have given no warning of AMA. Alternatively, when documentation was present stating that the patient had verbally expressed a desire to leave AMA, this was considered advanced warning of AMA.

RESULTS

Mean age and gender distribution were similar to those reported in other AMA studies (Table 1).[3] Thirty‐day ED revisit and 30‐day hospital readmission frequencies for medical service patients were 121 (41.6%) and 88 (30.2%), respectively, also similar to those reported in other AMA studies.[3]

Although our intent was to conduct a quantitative assessment of discharge interventions, we found stated reasons for leaving similar to those previously reported. In our study, AMA patients tended to be younger, more likely male, and at increased risk for AMA discharge if they had prior AMA discharges (Table 1). The most common reasons found in the medical record for leaving AMA were caring for sick family members, financial concerns, feeling better, and occasionally dissatisfaction with care, reasons similar to those reported in previous studies.[5, 7, 9, 10]

AMA notes were present in 276 (94.8%) charts. AMA notes were written by physicians in 163 (59.1%) and nurses in 110 (37.8%) encounters. The informed consent form was present in 88 (30.2%) charts, mentioned in the note but not present in the electronic medical record in 111 (38.1%), and not signed in 92 (31.6%) charts. Decision‐making capacity and health literacy were documented in 108 (37.1%) and 75 (25.8%) records, respectively. Warning of impending AMA was present in 217 (74.6%) charts. Medications prescribed and follow‐up plans were only documented in 71 (24.4%) and 91 (31.3%) charts, respectively (Table 2).

Patients with documentation of medications given did not have decreased 30‐day ED revisits (33.8% vs 44.3%, P = 0.12) or 30‐day hospital readmission (23.9% vs 32.4%, P = 0.18). Similarly, there was no relationship between documentation of follow‐up plans and 30‐day ED revisits (37.4% vs 43.7%, P = 0.31) or 30‐day hospital readmission (29.7% vs 30.7%, P = 0.87). Finally, there was no relationship between physician versus nurse authorship of AMA notes and 30‐day ED revisits (37.4% vs 46.4%, P = 0.14) or 30‐day hospital readmission (28.2% vs 31.8%, P = 0.52) (Table 3).

Physician documentation of the AMA was associated with an increased frequency of discharge medication being prescribed (36.2% vs 10.0%, P 0.001) and with an increased finding of documented follow‐up plans (43.6% vs 16.4%, P 0.001). A documented warning of impending AMA was associated with an increased frequency of discharge medication being prescribed (30.4% vs 6.8%, P 0.001) and increased frequency of follow‐up plans being documented (37.3% vs 13.5%, P 0.001) (Table 3).

DISCUSSION

To gain insights into opportunities for discharge transition interventions in this potentially more vulnerable population,[1, 5] we analyzed AMA documentation and what interventions were carried out at the time of discharge. Our intent was a quantitative assessment of discharge interventions, but we also found stated reasons for leaving AMA that were similar to those previously reported.[5, 6, 10]

We identified several opportunities for improved documentation as well as targeted discharge intervention among AMA patients. Documentation in the charts of AMA patients was often suboptimal. In our study, a physician's AMA note was present only half of the time. Mention of the patient's mental status or health literacy was present in only one‐fourth of cases. Protection from litigation in AMA cases is enhanced when these elements and others, like informed consent, are present in the medical record.[11]

Physician documentation of the AMA was associated with an increased frequency of discharge medication being prescribed and with an increased finding of documented follow‐up plans. This association might be confounded by the fact that physicians can prescribe whereas most nurses cannot. The findings that a documented warning of impending AMA was associated with an increased frequency of discharge medication being prescribed (30.4% vs 6.8%, P 0.001) and increased frequency of follow‐up plans being documented (37.3% vs 13.5%, P 0.001) suggest opportunities for improvement through early inquiry about potential for AMA as well as early responses when patients threaten to leave AMA.

An important focus of our study was on documentation of discharge medications and follow‐up plans. These elements were documented in 31% and 25% of charts, respectively. A warning of impending AMA was present in 74.6% of encounters, yet medications and follow‐up plans were documented at a much lower rate. This represents an area where caregivers have the possibility to intervene, but are not documenting that they are doing so. We found no relationship between the documentation of giving prescriptions and giving explicit follow‐up plans with decreased rates of return to the ED or readmission, but that possibility may still warrant future prospective study.

Our study did not attempt to explain why only a minority of AMA discharges include medication prescription or follow‐up plans, but a number of potential explanations are possible. Some AMA discharges may occur unannounced with a patient simply walking off the ward giving little or no advance notice. It is also possible that provider perceptions and attitudes toward AMA patients may influence potential interventions.[12] An AMA discharge is against the caregiver's preferred advice for the patient, and it may seem illogical to offer patients second‐best advice. Perhaps some providers have the misconception that medications cannot or must not be prescribed for an AMA discharge. However, second‐best therapy may be better than no therapy, and some follow‐up better than no follow‐up plan.

Given the high rates of ED return and 30‐day readmission, the associated increased healthcare costs as well as increased morbidity and mortality associated with AMA dispositions, a continued search for effective intervention strategies and opportunities is warranted. Recently, programs for transition of care/discharge have demonstrated improved outcomes including reduced rates of readmission with standard discharges.[13] At the time of our study, effective programs such as Project BOOST (Better Outcomes for Older adult through Safe Transitions),[14] the Care Transitions Program,[15] and RED (Project Re‐engineered Discharge)[16] were not yet routinely employed, but their common elements may be applicable to the AMA population. In general, these programs focus on elements we investigated (patient understanding, follow‐up plans, medications prescribed) but add a number of additional components. Additional elements include written discharge instructions, patient education, teach‐back process, decision support, emergency plans, caregiver education, telephone follow‐up, and transition coaches to coordinate home and office follow‐up visits. Most potential interventions add significant time (and cost) to the discharge process. Thus, future studies applying these components to AMA discharges should emphasize timely identification of threatened AMA and prioritized interventions. Future studies should focus on which interventions are the most cost‐effective with AMA patients.

Limitations of our study include not being able to access information from area hospitals not in our hospital network, and thus we may not have identified all ED returns and readmissions. Additionally, interventions at the time of discharge (like prescription of medications or provider assessment of decision‐making capacity) may not have been documented and thus not available for our review. Also, our study was a retrospective review at a single institution and included a relatively small population of patients; consequently, our findings may not apply to other healthcare providers in other hospitals or settings. Our study was strengthened by reviewing all consecutive AMA cases over a 2‐year period encompassing a diverse group of healthcare providers.

CONCLUSION

In the majority of cases reviewed, some advance warning of impending AMA is apparent, affording an opportunity for interventions that may improve health outcomes. Despite this advance warning, only a minority of cases result in key interventions such as prescription of medications or development of follow‐up plans. Medical documentation of AMA dispositions is often inadequate, suggesting missed opportunities for potential intervention as well as suboptimal medicolegal scenarios. Future prospective studies examining cost‐effective interventions at the time of AMA discharge and transition of care may provide valuable insight into lowering rates of ED return and rehospitalization.

Disclosures: The views and opinions expressed in this article are those of the author(s) and do not reflect official policy or position of the United States Air Force, Department of Defense, or US government. The authors report no conflicts of interest.

Approximately 1% to 2% of inpatient stays result in discharges against medical advice (AMA).[1] Though relatively infrequent, AMA discharges warrant attention as they are associated with higher morbidity, increased risk of readmission, and greater 30‐day mortality.[2] A recent study found a 30‐day readmission rate among AMA patients of 24.5%, nearly twice that of matched non‐AMA patients, and a 30‐day mortality rate of 1.3%, also nearly double that of planned discharges.[3] Discharges AMA may be expected to decrease index length of stay, yet accounting for 30‐day readmissions they are estimated to increase costs 56% higher than expected from an initial hospitalization.[4] Patients note several possible reasons for leaving AMA including family emergencies, dissatisfaction with care, financial concerns, or simply feeling better, among others.[5, 6, 7] Risk factors for AMA discharges include previous AMA discharge, having no primary care physician, younger age, lack of insurance, male sex, substance abuse, and lower socioeconomic status.[4, 6, 7, 8]

A number of prior studies have assessed risk factors for AMA discharges, the long‐ and short‐term outcomes, patient reasons for leaving, and physician perceptions of why patients leave AMA.[3, 5, 7, 9] However, there is limited information about opportunities for discharge transition interventions in this potentially more vulnerable population. Because of the increased short‐term and long‐term risks to these patients, treatment and follow‐up plans at the time of discharge may carry even greater importance than follow‐up plans with standard discharges. This study analyzed AMA documentation and what interventions were carried out at the time of discharge.

METHODS

We reviewed the records of all adult patients, ages 18 years and older, admitted to a university‐affiliated tertiary care hospital in Dayton, Ohio (a 520‐bed hospital with approximately 17,000 adult patient encounters per year) over a 2‐year period, and who subsequently left AMA. A hospital database identified 351 adult AMA cases (1.0% of adult admissions). A single reviewer performed an in‐depth review of the 291 patient admissions to the general medical service between January 1, 2009 and December 31, 2010, and manually reviewed and abstracted the data of interest. The Wright State University institutional review board approved the study.

Documentation review focused on the presence of a specified AMA note, the presence of documentation addressing informed consent, patient decision‐making capacity, patient health literacy, follow‐up plans, whether or not medications were prescribed, and whether or not any warning indicators of impending AMA were apparent. These items represented key elements of the discharge policy and procedure in place at our institution during the period of study. We speculated that nurses may be more immediately available at the time of AMA discharge and thus might carry out AMA documentation more often than physicians. To assess this we recorded the role (nurse vs provider) of the writer of AMA notes. We also assessed patient gender, length of stay, prior AMA, 30‐day emergency department (ED) re‐encounters, and 30‐day hospital readmission after AMA discharge.

Informed consent was deemed present if patients signed the hospital's standardized AMA form. Decision‐making capacity was assessed as present if there was specific mention of the patient's capacity on the day of discharge. Any mention of health literacy or the patient's stated understanding of his medical condition at any time during the hospitalization was considered positive documentation of healthcare literacy. Follow‐up plans included any mention of where and when the patient would return. Discharge medications included prescribed medication or indication that no medications were warranted. Warning indicators included specific mention of the patient's desire to leave AMA. For example, patients who left the unit without informing staff were considered to have given no warning of AMA. Alternatively, when documentation was present stating that the patient had verbally expressed a desire to leave AMA, this was considered advanced warning of AMA.

RESULTS

Mean age and gender distribution were similar to those reported in other AMA studies (Table 1).[3] Thirty‐day ED revisit and 30‐day hospital readmission frequencies for medical service patients were 121 (41.6%) and 88 (30.2%), respectively, also similar to those reported in other AMA studies.[3]

Although our intent was to conduct a quantitative assessment of discharge interventions, we found stated reasons for leaving similar to those previously reported. In our study, AMA patients tended to be younger, more likely male, and at increased risk for AMA discharge if they had prior AMA discharges (Table 1). The most common reasons found in the medical record for leaving AMA were caring for sick family members, financial concerns, feeling better, and occasionally dissatisfaction with care, reasons similar to those reported in previous studies.[5, 7, 9, 10]

AMA notes were present in 276 (94.8%) charts. AMA notes were written by physicians in 163 (59.1%) and nurses in 110 (37.8%) encounters. The informed consent form was present in 88 (30.2%) charts, mentioned in the note but not present in the electronic medical record in 111 (38.1%), and not signed in 92 (31.6%) charts. Decision‐making capacity and health literacy were documented in 108 (37.1%) and 75 (25.8%) records, respectively. Warning of impending AMA was present in 217 (74.6%) charts. Medications prescribed and follow‐up plans were only documented in 71 (24.4%) and 91 (31.3%) charts, respectively (Table 2).

Patients with documentation of medications given did not have decreased 30‐day ED revisits (33.8% vs 44.3%, P = 0.12) or 30‐day hospital readmission (23.9% vs 32.4%, P = 0.18). Similarly, there was no relationship between documentation of follow‐up plans and 30‐day ED revisits (37.4% vs 43.7%, P = 0.31) or 30‐day hospital readmission (29.7% vs 30.7%, P = 0.87). Finally, there was no relationship between physician versus nurse authorship of AMA notes and 30‐day ED revisits (37.4% vs 46.4%, P = 0.14) or 30‐day hospital readmission (28.2% vs 31.8%, P = 0.52) (Table 3).

Physician documentation of the AMA was associated with an increased frequency of discharge medication being prescribed (36.2% vs 10.0%, P 0.001) and with an increased finding of documented follow‐up plans (43.6% vs 16.4%, P 0.001). A documented warning of impending AMA was associated with an increased frequency of discharge medication being prescribed (30.4% vs 6.8%, P 0.001) and increased frequency of follow‐up plans being documented (37.3% vs 13.5%, P 0.001) (Table 3).

DISCUSSION

To gain insights into opportunities for discharge transition interventions in this potentially more vulnerable population,[1, 5] we analyzed AMA documentation and what interventions were carried out at the time of discharge. Our intent was a quantitative assessment of discharge interventions, but we also found stated reasons for leaving AMA that were similar to those previously reported.[5, 6, 10]

We identified several opportunities for improved documentation as well as targeted discharge intervention among AMA patients. Documentation in the charts of AMA patients was often suboptimal. In our study, a physician's AMA note was present only half of the time. Mention of the patient's mental status or health literacy was present in only one‐fourth of cases. Protection from litigation in AMA cases is enhanced when these elements and others, like informed consent, are present in the medical record.[11]

Physician documentation of the AMA was associated with an increased frequency of discharge medication being prescribed and with an increased finding of documented follow‐up plans. This association might be confounded by the fact that physicians can prescribe whereas most nurses cannot. The findings that a documented warning of impending AMA was associated with an increased frequency of discharge medication being prescribed (30.4% vs 6.8%, P 0.001) and increased frequency of follow‐up plans being documented (37.3% vs 13.5%, P 0.001) suggest opportunities for improvement through early inquiry about potential for AMA as well as early responses when patients threaten to leave AMA.

An important focus of our study was on documentation of discharge medications and follow‐up plans. These elements were documented in 31% and 25% of charts, respectively. A warning of impending AMA was present in 74.6% of encounters, yet medications and follow‐up plans were documented at a much lower rate. This represents an area where caregivers have the possibility to intervene, but are not documenting that they are doing so. We found no relationship between the documentation of giving prescriptions and giving explicit follow‐up plans with decreased rates of return to the ED or readmission, but that possibility may still warrant future prospective study.

Our study did not attempt to explain why only a minority of AMA discharges include medication prescription or follow‐up plans, but a number of potential explanations are possible. Some AMA discharges may occur unannounced with a patient simply walking off the ward giving little or no advance notice. It is also possible that provider perceptions and attitudes toward AMA patients may influence potential interventions.[12] An AMA discharge is against the caregiver's preferred advice for the patient, and it may seem illogical to offer patients second‐best advice. Perhaps some providers have the misconception that medications cannot or must not be prescribed for an AMA discharge. However, second‐best therapy may be better than no therapy, and some follow‐up better than no follow‐up plan.

Given the high rates of ED return and 30‐day readmission, the associated increased healthcare costs as well as increased morbidity and mortality associated with AMA dispositions, a continued search for effective intervention strategies and opportunities is warranted. Recently, programs for transition of care/discharge have demonstrated improved outcomes including reduced rates of readmission with standard discharges.[13] At the time of our study, effective programs such as Project BOOST (Better Outcomes for Older adult through Safe Transitions),[14] the Care Transitions Program,[15] and RED (Project Re‐engineered Discharge)[16] were not yet routinely employed, but their common elements may be applicable to the AMA population. In general, these programs focus on elements we investigated (patient understanding, follow‐up plans, medications prescribed) but add a number of additional components. Additional elements include written discharge instructions, patient education, teach‐back process, decision support, emergency plans, caregiver education, telephone follow‐up, and transition coaches to coordinate home and office follow‐up visits. Most potential interventions add significant time (and cost) to the discharge process. Thus, future studies applying these components to AMA discharges should emphasize timely identification of threatened AMA and prioritized interventions. Future studies should focus on which interventions are the most cost‐effective with AMA patients.

Limitations of our study include not being able to access information from area hospitals not in our hospital network, and thus we may not have identified all ED returns and readmissions. Additionally, interventions at the time of discharge (like prescription of medications or provider assessment of decision‐making capacity) may not have been documented and thus not available for our review. Also, our study was a retrospective review at a single institution and included a relatively small population of patients; consequently, our findings may not apply to other healthcare providers in other hospitals or settings. Our study was strengthened by reviewing all consecutive AMA cases over a 2‐year period encompassing a diverse group of healthcare providers.

CONCLUSION

In the majority of cases reviewed, some advance warning of impending AMA is apparent, affording an opportunity for interventions that may improve health outcomes. Despite this advance warning, only a minority of cases result in key interventions such as prescription of medications or development of follow‐up plans. Medical documentation of AMA dispositions is often inadequate, suggesting missed opportunities for potential intervention as well as suboptimal medicolegal scenarios. Future prospective studies examining cost‐effective interventions at the time of AMA discharge and transition of care may provide valuable insight into lowering rates of ED return and rehospitalization.

Disclosures: The views and opinions expressed in this article are those of the author(s) and do not reflect official policy or position of the United States Air Force, Department of Defense, or US government. The authors report no conflicts of interest.

Approximately 1% to 2% of inpatient stays result in discharges against medical advice (AMA).[1] Though relatively infrequent, AMA discharges warrant attention as they are associated with higher morbidity, increased risk of readmission, and greater 30‐day mortality.[2] A recent study found a 30‐day readmission rate among AMA patients of 24.5%, nearly twice that of matched non‐AMA patients, and a 30‐day mortality rate of 1.3%, also nearly double that of planned discharges.[3] Discharges AMA may be expected to decrease index length of stay, yet accounting for 30‐day readmissions they are estimated to increase costs 56% higher than expected from an initial hospitalization.[4] Patients note several possible reasons for leaving AMA including family emergencies, dissatisfaction with care, financial concerns, or simply feeling better, among others.[5, 6, 7] Risk factors for AMA discharges include previous AMA discharge, having no primary care physician, younger age, lack of insurance, male sex, substance abuse, and lower socioeconomic status.[4, 6, 7, 8]

A number of prior studies have assessed risk factors for AMA discharges, the long‐ and short‐term outcomes, patient reasons for leaving, and physician perceptions of why patients leave AMA.[3, 5, 7, 9] However, there is limited information about opportunities for discharge transition interventions in this potentially more vulnerable population. Because of the increased short‐term and long‐term risks to these patients, treatment and follow‐up plans at the time of discharge may carry even greater importance than follow‐up plans with standard discharges. This study analyzed AMA documentation and what interventions were carried out at the time of discharge.

METHODS

We reviewed the records of all adult patients, ages 18 years and older, admitted to a university‐affiliated tertiary care hospital in Dayton, Ohio (a 520‐bed hospital with approximately 17,000 adult patient encounters per year) over a 2‐year period, and who subsequently left AMA. A hospital database identified 351 adult AMA cases (1.0% of adult admissions). A single reviewer performed an in‐depth review of the 291 patient admissions to the general medical service between January 1, 2009 and December 31, 2010, and manually reviewed and abstracted the data of interest. The Wright State University institutional review board approved the study.

Documentation review focused on the presence of a specified AMA note, the presence of documentation addressing informed consent, patient decision‐making capacity, patient health literacy, follow‐up plans, whether or not medications were prescribed, and whether or not any warning indicators of impending AMA were apparent. These items represented key elements of the discharge policy and procedure in place at our institution during the period of study. We speculated that nurses may be more immediately available at the time of AMA discharge and thus might carry out AMA documentation more often than physicians. To assess this we recorded the role (nurse vs provider) of the writer of AMA notes. We also assessed patient gender, length of stay, prior AMA, 30‐day emergency department (ED) re‐encounters, and 30‐day hospital readmission after AMA discharge.

Informed consent was deemed present if patients signed the hospital's standardized AMA form. Decision‐making capacity was assessed as present if there was specific mention of the patient's capacity on the day of discharge. Any mention of health literacy or the patient's stated understanding of his medical condition at any time during the hospitalization was considered positive documentation of healthcare literacy. Follow‐up plans included any mention of where and when the patient would return. Discharge medications included prescribed medication or indication that no medications were warranted. Warning indicators included specific mention of the patient's desire to leave AMA. For example, patients who left the unit without informing staff were considered to have given no warning of AMA. Alternatively, when documentation was present stating that the patient had verbally expressed a desire to leave AMA, this was considered advanced warning of AMA.

RESULTS

Mean age and gender distribution were similar to those reported in other AMA studies (Table 1).[3] Thirty‐day ED revisit and 30‐day hospital readmission frequencies for medical service patients were 121 (41.6%) and 88 (30.2%), respectively, also similar to those reported in other AMA studies.[3]

Although our intent was to conduct a quantitative assessment of discharge interventions, we found stated reasons for leaving similar to those previously reported. In our study, AMA patients tended to be younger, more likely male, and at increased risk for AMA discharge if they had prior AMA discharges (Table 1). The most common reasons found in the medical record for leaving AMA were caring for sick family members, financial concerns, feeling better, and occasionally dissatisfaction with care, reasons similar to those reported in previous studies.[5, 7, 9, 10]

AMA notes were present in 276 (94.8%) charts. AMA notes were written by physicians in 163 (59.1%) and nurses in 110 (37.8%) encounters. The informed consent form was present in 88 (30.2%) charts, mentioned in the note but not present in the electronic medical record in 111 (38.1%), and not signed in 92 (31.6%) charts. Decision‐making capacity and health literacy were documented in 108 (37.1%) and 75 (25.8%) records, respectively. Warning of impending AMA was present in 217 (74.6%) charts. Medications prescribed and follow‐up plans were only documented in 71 (24.4%) and 91 (31.3%) charts, respectively (Table 2).

Patients with documentation of medications given did not have decreased 30‐day ED revisits (33.8% vs 44.3%, P = 0.12) or 30‐day hospital readmission (23.9% vs 32.4%, P = 0.18). Similarly, there was no relationship between documentation of follow‐up plans and 30‐day ED revisits (37.4% vs 43.7%, P = 0.31) or 30‐day hospital readmission (29.7% vs 30.7%, P = 0.87). Finally, there was no relationship between physician versus nurse authorship of AMA notes and 30‐day ED revisits (37.4% vs 46.4%, P = 0.14) or 30‐day hospital readmission (28.2% vs 31.8%, P = 0.52) (Table 3).

Physician documentation of the AMA was associated with an increased frequency of discharge medication being prescribed (36.2% vs 10.0%, P 0.001) and with an increased finding of documented follow‐up plans (43.6% vs 16.4%, P 0.001). A documented warning of impending AMA was associated with an increased frequency of discharge medication being prescribed (30.4% vs 6.8%, P 0.001) and increased frequency of follow‐up plans being documented (37.3% vs 13.5%, P 0.001) (Table 3).

DISCUSSION

To gain insights into opportunities for discharge transition interventions in this potentially more vulnerable population,[1, 5] we analyzed AMA documentation and what interventions were carried out at the time of discharge. Our intent was a quantitative assessment of discharge interventions, but we also found stated reasons for leaving AMA that were similar to those previously reported.[5, 6, 10]

We identified several opportunities for improved documentation as well as targeted discharge intervention among AMA patients. Documentation in the charts of AMA patients was often suboptimal. In our study, a physician's AMA note was present only half of the time. Mention of the patient's mental status or health literacy was present in only one‐fourth of cases. Protection from litigation in AMA cases is enhanced when these elements and others, like informed consent, are present in the medical record.[11]

Physician documentation of the AMA was associated with an increased frequency of discharge medication being prescribed and with an increased finding of documented follow‐up plans. This association might be confounded by the fact that physicians can prescribe whereas most nurses cannot. The findings that a documented warning of impending AMA was associated with an increased frequency of discharge medication being prescribed (30.4% vs 6.8%, P 0.001) and increased frequency of follow‐up plans being documented (37.3% vs 13.5%, P 0.001) suggest opportunities for improvement through early inquiry about potential for AMA as well as early responses when patients threaten to leave AMA.

An important focus of our study was on documentation of discharge medications and follow‐up plans. These elements were documented in 31% and 25% of charts, respectively. A warning of impending AMA was present in 74.6% of encounters, yet medications and follow‐up plans were documented at a much lower rate. This represents an area where caregivers have the possibility to intervene, but are not documenting that they are doing so. We found no relationship between the documentation of giving prescriptions and giving explicit follow‐up plans with decreased rates of return to the ED or readmission, but that possibility may still warrant future prospective study.

Our study did not attempt to explain why only a minority of AMA discharges include medication prescription or follow‐up plans, but a number of potential explanations are possible. Some AMA discharges may occur unannounced with a patient simply walking off the ward giving little or no advance notice. It is also possible that provider perceptions and attitudes toward AMA patients may influence potential interventions.[12] An AMA discharge is against the caregiver's preferred advice for the patient, and it may seem illogical to offer patients second‐best advice. Perhaps some providers have the misconception that medications cannot or must not be prescribed for an AMA discharge. However, second‐best therapy may be better than no therapy, and some follow‐up better than no follow‐up plan.

Given the high rates of ED return and 30‐day readmission, the associated increased healthcare costs as well as increased morbidity and mortality associated with AMA dispositions, a continued search for effective intervention strategies and opportunities is warranted. Recently, programs for transition of care/discharge have demonstrated improved outcomes including reduced rates of readmission with standard discharges.[13] At the time of our study, effective programs such as Project BOOST (Better Outcomes for Older adult through Safe Transitions),[14] the Care Transitions Program,[15] and RED (Project Re‐engineered Discharge)[16] were not yet routinely employed, but their common elements may be applicable to the AMA population. In general, these programs focus on elements we investigated (patient understanding, follow‐up plans, medications prescribed) but add a number of additional components. Additional elements include written discharge instructions, patient education, teach‐back process, decision support, emergency plans, caregiver education, telephone follow‐up, and transition coaches to coordinate home and office follow‐up visits. Most potential interventions add significant time (and cost) to the discharge process. Thus, future studies applying these components to AMA discharges should emphasize timely identification of threatened AMA and prioritized interventions. Future studies should focus on which interventions are the most cost‐effective with AMA patients.

Limitations of our study include not being able to access information from area hospitals not in our hospital network, and thus we may not have identified all ED returns and readmissions. Additionally, interventions at the time of discharge (like prescription of medications or provider assessment of decision‐making capacity) may not have been documented and thus not available for our review. Also, our study was a retrospective review at a single institution and included a relatively small population of patients; consequently, our findings may not apply to other healthcare providers in other hospitals or settings. Our study was strengthened by reviewing all consecutive AMA cases over a 2‐year period encompassing a diverse group of healthcare providers.

CONCLUSION

In the majority of cases reviewed, some advance warning of impending AMA is apparent, affording an opportunity for interventions that may improve health outcomes. Despite this advance warning, only a minority of cases result in key interventions such as prescription of medications or development of follow‐up plans. Medical documentation of AMA dispositions is often inadequate, suggesting missed opportunities for potential intervention as well as suboptimal medicolegal scenarios. Future prospective studies examining cost‐effective interventions at the time of AMA discharge and transition of care may provide valuable insight into lowering rates of ED return and rehospitalization.

Disclosures: The views and opinions expressed in this article are those of the author(s) and do not reflect official policy or position of the United States Air Force, Department of Defense, or US government. The authors report no conflicts of interest.

Transitions of care, which encompass the patient experience of hospital discharge to the community, are frequently associated with clinically and financially costly adverse events.[1, 2] One important element for reducing the risk of postdischarge adverse events is provision of timely follow‐up by a clinician familiar with the patient and hospital course.[3, 4]

However, achieving this ideal is becoming more difficult because of an increased demand for primary care services (due to expanding coverage of Medicare and Medicaid) and the decreased supply of primary care physicians.[5, 6] When a timely visit with a clinician is available postdischarge, the widening discontinuity between inpatient and outpatient care providers often means this clinician is lacking essential details of the hospitalization.[7, 8]

One increasingly common innovation to improve postdischarge care access and continuity is to extend the role of inpatient providers (usually hospitalists) to provide care after discharge in a postdischarge clinic (PDC).[9, 10, 11] These clinics require an expansion of a hospitalist's duties to the outpatient setting, a requirement that has met with hospitalist resistance in initial reports.[12] However, little is known about hospitalists' experience with PDCs or attitudes toward postdischarge care. We aimed to explore these attitudes and experiences surrounding postdischarge care and PDCs.

METHODS

We conducted a cross‐sectional 17‐question Web‐based survey of hospitalists at 20 academic and 17 VA medical centers across the United States. Hospital medicine faculty at each site were identified by their group leader; members of each group then received an email survey up to 3 times. To collect responses from nonacademic hospitalists, the survey was also distributed to a large national private hospitalist employer. Due to internal limitations at the employer site, sampling was not feasible, and thus a convenience sample was obtained. Hospitalists who were not clinically active or did not have computer access to complete the survey were excluded. Responses were initially gathered on a 4‐point Likert scale; for comparisons between groups the scale was collapsed to a binary comparison using Fisher exact or ² tests. We included questions answered in partially completed surveys in both the numerator and denominator; questions not answered were excluded from both numerator and denominator. The denominator of all responses was noted. All analyses were conducted using SAS 9.3 (SAS Institute, Inc., Cary, NC). The study was approved by the Colorado Multiple Institutional Review Board.

RESULTS

Of 814 hospitalists, 228 responded to the survey (28.3%). Table 1 illustrates characteristics of responding hospitalists, who were divided between university hospitals, community teaching hospitals, and community nonteaching hospitals in diverse practices in terms of location and group size.

Sixty‐one percent of responding hospitalists believed most patient problems after discharge were due to poor follow‐up with primary care providers, and 55% found it difficult to arrange timely primary care follow up (Table 2). Despite this, 87% thought patient problems after discharge should be cared for by primary care physicians, and 62% opposed the idea of hospitalists seeing patients in the clinic after discharge.

When asked if hospitalists were responsible for patients after discharge from the hospital, only 50% responded positively. However, when asked how long hospitalists were responsible for patients after discharge, 71% gave a response longer than hospital discharge, including 60% who believed this responsibility ended at 1 week or less following discharge. A minority (12%) felt it extended to 1 month following discharge (Table 3).

Responding hospitalists expressed confidence in a PDC to reduce postdischarge emergency department visits (74%). However, most felt they would require extra compensation to staff a PDC (77%). They were divided on whether they would discharge patients from the hospital earlier if they could see those patients in postdischarge follow‐up (51% would discharge patients earlier).

Compared to those who had not experienced a PDC, responding hospitalists who had provided care in a PDC trended toward responding more positively that hospitalists should provide postdischarge care (P = 0.054). Few responding hospitalists had such exposure (8.8%) at the time of the survey. Although 31% had considering starting a PDC, only 3% were starting in the next year. Of responding hospitalists with exposure to a PDC, 70% were satisfied with the experience and 85% felt their patients were satisfied. Responses did not vary by type of practice (academic vs nonacademic), group size, geographic location, or by exposure to a PDC except as above.

DISCUSSION

Responding hospitalists reported encountering significant difficulty arranging appropriate postdischarge appointments with primary care providers and feel this contributes to postdischarge complications. Nearly 75% of those surveyed felt a hospitalist‐run PDC would be effective in reducing postdischarge emergency department visits, presumably in part due to improved access to postdischarge care. However, 62% of responding hospitalists opposed providing this type of care, though those who had experienced a PDC were somewhat more likely to view providing care in a PDC favorably. Survey responses largely reflect attitudes rather than experience with PDCs, because very few respondents had ever worked in a PDC.

The juxtaposition of the confidence expressed in PDCs to reduce postdischarge emergency department visits with the less enthusiastic views of respondents about providing care in a PDC was surprising. Several explanations are possible. First, providing such care is outside the usual scope of practice of most hospitalists, and preliminary reports indicate hospitalists, as self‐selected inpatient providers, may not initially welcome this opportunity.[12] Second, responding hospitalists identified the need for extra compensation for providing this care, suggesting they would see staffing a PDC as a burden requiring extra payment. Third, only 12% of respondents felt their responsibility to their discharged patients extended to 1 month following discharge. Given this, hospitalists may not feel enough personal ownership over 30‐day readmission rates to justify the additional clinical demand of staffing a PDC.[13]

In fact, 29% of responding hospitalists felt their responsibility to the patient ended at the time of discharge. Respondents may have interpreted responsibility differently, and we cannot rule out response bias given our lower‐than‐expected response rate. However, we had anticipated many fewer hospitalists would respond this way given professional hospitalist societies have endorsed guidelines for improved transitions of care, which clearly delineate the key role hospitalists play in care transitions.[14]

Although fewer than 10% of respondents had worked in a PDC, nearly one‐third reported considering starting such a practice in the future, underscoring the importance of understanding hospitalist attitudes and experiences when creating a PDC and the significant barriers to arranging appropriate postdischarge care identified by survey respondents. The barriers to establishing a PDC may explain why few planned to start a PDC in the next year.

This study should be interpreted in the context of its design. Due to limitations in survey delivery, more rigorous sampling designs could not be used, and efforts were instead made to deliver the survey to a diverse group of hospitalists. The survey response rate was lower than anticipated and this increases the risk of response bias. Though this response rate is characteristic of other surveys of hospitalists, responses may have been from a selected population and therefore not representative of all hospitalists.[15] We sampled from a variety of practice venues, locations, academic and community practices, and practice group sizes to try to minimize this bias. Due to the low exposure rate to PDCs, hospitalist responses to experiences with PDCs should be considered exploratory.

We asked about similar content areas in the survey in multiple questions to maximize content validity; this resulted in variations in the degree of agreement or disagreement to similar prompts. For example, 62% of hospitalists opposed seeing patients in the clinic after discharge when directly asked, but nearly 50% said they would welcome the opportunity to work in a PDC if their employer required it. In another example, 50% of respondents said their responsibility for the patient ended at time of discharge, but when asked about duration of responsibility, 30% identified time of discharge as the limit. When reporting and interpreting results, we have tried to highlight responses to questions that ask most clearly and directly about the content of interest (rather than general themes), but this interpretation may also be subject to bias.

The time after hospital discharge is one of heightened risk for adverse events for the recently discharged patient. Hospitalist‐run postdischarge clinics may offer improved postdischarge care access and continuity; more research is needed on the effects of such clinics on patient outcomes, including postdischarge utilization. Until then, physicians and hospitals considering establishing PDCs should consider the barriers responding hospitalists identified to working in such a clinic, as well as the confidence they expressed in PDCs to reduce subsequent utilization.

Disclosures: Dr. Burke had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Veterans Affairs. Dr. Burke was supported by grant funding from the Colorado Research Enhancement Award Program to Improve Care Coordination for Veterans. Dr. Ryan has no conflicts of interest to disclose.

Transitions of care, which encompass the patient experience of hospital discharge to the community, are frequently associated with clinically and financially costly adverse events.[1, 2] One important element for reducing the risk of postdischarge adverse events is provision of timely follow‐up by a clinician familiar with the patient and hospital course.[3, 4]

However, achieving this ideal is becoming more difficult because of an increased demand for primary care services (due to expanding coverage of Medicare and Medicaid) and the decreased supply of primary care physicians.[5, 6] When a timely visit with a clinician is available postdischarge, the widening discontinuity between inpatient and outpatient care providers often means this clinician is lacking essential details of the hospitalization.[7, 8]

One increasingly common innovation to improve postdischarge care access and continuity is to extend the role of inpatient providers (usually hospitalists) to provide care after discharge in a postdischarge clinic (PDC).[9, 10, 11] These clinics require an expansion of a hospitalist's duties to the outpatient setting, a requirement that has met with hospitalist resistance in initial reports.[12] However, little is known about hospitalists' experience with PDCs or attitudes toward postdischarge care. We aimed to explore these attitudes and experiences surrounding postdischarge care and PDCs.

METHODS

We conducted a cross‐sectional 17‐question Web‐based survey of hospitalists at 20 academic and 17 VA medical centers across the United States. Hospital medicine faculty at each site were identified by their group leader; members of each group then received an email survey up to 3 times. To collect responses from nonacademic hospitalists, the survey was also distributed to a large national private hospitalist employer. Due to internal limitations at the employer site, sampling was not feasible, and thus a convenience sample was obtained. Hospitalists who were not clinically active or did not have computer access to complete the survey were excluded. Responses were initially gathered on a 4‐point Likert scale; for comparisons between groups the scale was collapsed to a binary comparison using Fisher exact or ² tests. We included questions answered in partially completed surveys in both the numerator and denominator; questions not answered were excluded from both numerator and denominator. The denominator of all responses was noted. All analyses were conducted using SAS 9.3 (SAS Institute, Inc., Cary, NC). The study was approved by the Colorado Multiple Institutional Review Board.

RESULTS

Of 814 hospitalists, 228 responded to the survey (28.3%). Table 1 illustrates characteristics of responding hospitalists, who were divided between university hospitals, community teaching hospitals, and community nonteaching hospitals in diverse practices in terms of location and group size.

Sixty‐one percent of responding hospitalists believed most patient problems after discharge were due to poor follow‐up with primary care providers, and 55% found it difficult to arrange timely primary care follow up (Table 2). Despite this, 87% thought patient problems after discharge should be cared for by primary care physicians, and 62% opposed the idea of hospitalists seeing patients in the clinic after discharge.

When asked if hospitalists were responsible for patients after discharge from the hospital, only 50% responded positively. However, when asked how long hospitalists were responsible for patients after discharge, 71% gave a response longer than hospital discharge, including 60% who believed this responsibility ended at 1 week or less following discharge. A minority (12%) felt it extended to 1 month following discharge (Table 3).

Responding hospitalists expressed confidence in a PDC to reduce postdischarge emergency department visits (74%). However, most felt they would require extra compensation to staff a PDC (77%). They were divided on whether they would discharge patients from the hospital earlier if they could see those patients in postdischarge follow‐up (51% would discharge patients earlier).

Compared to those who had not experienced a PDC, responding hospitalists who had provided care in a PDC trended toward responding more positively that hospitalists should provide postdischarge care (P = 0.054). Few responding hospitalists had such exposure (8.8%) at the time of the survey. Although 31% had considering starting a PDC, only 3% were starting in the next year. Of responding hospitalists with exposure to a PDC, 70% were satisfied with the experience and 85% felt their patients were satisfied. Responses did not vary by type of practice (academic vs nonacademic), group size, geographic location, or by exposure to a PDC except as above.

DISCUSSION

Responding hospitalists reported encountering significant difficulty arranging appropriate postdischarge appointments with primary care providers and feel this contributes to postdischarge complications. Nearly 75% of those surveyed felt a hospitalist‐run PDC would be effective in reducing postdischarge emergency department visits, presumably in part due to improved access to postdischarge care. However, 62% of responding hospitalists opposed providing this type of care, though those who had experienced a PDC were somewhat more likely to view providing care in a PDC favorably. Survey responses largely reflect attitudes rather than experience with PDCs, because very few respondents had ever worked in a PDC.

The juxtaposition of the confidence expressed in PDCs to reduce postdischarge emergency department visits with the less enthusiastic views of respondents about providing care in a PDC was surprising. Several explanations are possible. First, providing such care is outside the usual scope of practice of most hospitalists, and preliminary reports indicate hospitalists, as self‐selected inpatient providers, may not initially welcome this opportunity.[12] Second, responding hospitalists identified the need for extra compensation for providing this care, suggesting they would see staffing a PDC as a burden requiring extra payment. Third, only 12% of respondents felt their responsibility to their discharged patients extended to 1 month following discharge. Given this, hospitalists may not feel enough personal ownership over 30‐day readmission rates to justify the additional clinical demand of staffing a PDC.[13]

In fact, 29% of responding hospitalists felt their responsibility to the patient ended at the time of discharge. Respondents may have interpreted responsibility differently, and we cannot rule out response bias given our lower‐than‐expected response rate. However, we had anticipated many fewer hospitalists would respond this way given professional hospitalist societies have endorsed guidelines for improved transitions of care, which clearly delineate the key role hospitalists play in care transitions.[14]

Although fewer than 10% of respondents had worked in a PDC, nearly one‐third reported considering starting such a practice in the future, underscoring the importance of understanding hospitalist attitudes and experiences when creating a PDC and the significant barriers to arranging appropriate postdischarge care identified by survey respondents. The barriers to establishing a PDC may explain why few planned to start a PDC in the next year.

This study should be interpreted in the context of its design. Due to limitations in survey delivery, more rigorous sampling designs could not be used, and efforts were instead made to deliver the survey to a diverse group of hospitalists. The survey response rate was lower than anticipated and this increases the risk of response bias. Though this response rate is characteristic of other surveys of hospitalists, responses may have been from a selected population and therefore not representative of all hospitalists.[15] We sampled from a variety of practice venues, locations, academic and community practices, and practice group sizes to try to minimize this bias. Due to the low exposure rate to PDCs, hospitalist responses to experiences with PDCs should be considered exploratory.

We asked about similar content areas in the survey in multiple questions to maximize content validity; this resulted in variations in the degree of agreement or disagreement to similar prompts. For example, 62% of hospitalists opposed seeing patients in the clinic after discharge when directly asked, but nearly 50% said they would welcome the opportunity to work in a PDC if their employer required it. In another example, 50% of respondents said their responsibility for the patient ended at time of discharge, but when asked about duration of responsibility, 30% identified time of discharge as the limit. When reporting and interpreting results, we have tried to highlight responses to questions that ask most clearly and directly about the content of interest (rather than general themes), but this interpretation may also be subject to bias.

The time after hospital discharge is one of heightened risk for adverse events for the recently discharged patient. Hospitalist‐run postdischarge clinics may offer improved postdischarge care access and continuity; more research is needed on the effects of such clinics on patient outcomes, including postdischarge utilization. Until then, physicians and hospitals considering establishing PDCs should consider the barriers responding hospitalists identified to working in such a clinic, as well as the confidence they expressed in PDCs to reduce subsequent utilization.

Disclosures: Dr. Burke had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Veterans Affairs. Dr. Burke was supported by grant funding from the Colorado Research Enhancement Award Program to Improve Care Coordination for Veterans. Dr. Ryan has no conflicts of interest to disclose.

Transitions of care, which encompass the patient experience of hospital discharge to the community, are frequently associated with clinically and financially costly adverse events.[1, 2] One important element for reducing the risk of postdischarge adverse events is provision of timely follow‐up by a clinician familiar with the patient and hospital course.[3, 4]

However, achieving this ideal is becoming more difficult because of an increased demand for primary care services (due to expanding coverage of Medicare and Medicaid) and the decreased supply of primary care physicians.[5, 6] When a timely visit with a clinician is available postdischarge, the widening discontinuity between inpatient and outpatient care providers often means this clinician is lacking essential details of the hospitalization.[7, 8]

One increasingly common innovation to improve postdischarge care access and continuity is to extend the role of inpatient providers (usually hospitalists) to provide care after discharge in a postdischarge clinic (PDC).[9, 10, 11] These clinics require an expansion of a hospitalist's duties to the outpatient setting, a requirement that has met with hospitalist resistance in initial reports.[12] However, little is known about hospitalists' experience with PDCs or attitudes toward postdischarge care. We aimed to explore these attitudes and experiences surrounding postdischarge care and PDCs.

METHODS

We conducted a cross‐sectional 17‐question Web‐based survey of hospitalists at 20 academic and 17 VA medical centers across the United States. Hospital medicine faculty at each site were identified by their group leader; members of each group then received an email survey up to 3 times. To collect responses from nonacademic hospitalists, the survey was also distributed to a large national private hospitalist employer. Due to internal limitations at the employer site, sampling was not feasible, and thus a convenience sample was obtained. Hospitalists who were not clinically active or did not have computer access to complete the survey were excluded. Responses were initially gathered on a 4‐point Likert scale; for comparisons between groups the scale was collapsed to a binary comparison using Fisher exact or ² tests. We included questions answered in partially completed surveys in both the numerator and denominator; questions not answered were excluded from both numerator and denominator. The denominator of all responses was noted. All analyses were conducted using SAS 9.3 (SAS Institute, Inc., Cary, NC). The study was approved by the Colorado Multiple Institutional Review Board.

RESULTS

Of 814 hospitalists, 228 responded to the survey (28.3%). Table 1 illustrates characteristics of responding hospitalists, who were divided between university hospitals, community teaching hospitals, and community nonteaching hospitals in diverse practices in terms of location and group size.

Sixty‐one percent of responding hospitalists believed most patient problems after discharge were due to poor follow‐up with primary care providers, and 55% found it difficult to arrange timely primary care follow up (Table 2). Despite this, 87% thought patient problems after discharge should be cared for by primary care physicians, and 62% opposed the idea of hospitalists seeing patients in the clinic after discharge.

When asked if hospitalists were responsible for patients after discharge from the hospital, only 50% responded positively. However, when asked how long hospitalists were responsible for patients after discharge, 71% gave a response longer than hospital discharge, including 60% who believed this responsibility ended at 1 week or less following discharge. A minority (12%) felt it extended to 1 month following discharge (Table 3).

Responding hospitalists expressed confidence in a PDC to reduce postdischarge emergency department visits (74%). However, most felt they would require extra compensation to staff a PDC (77%). They were divided on whether they would discharge patients from the hospital earlier if they could see those patients in postdischarge follow‐up (51% would discharge patients earlier).

Compared to those who had not experienced a PDC, responding hospitalists who had provided care in a PDC trended toward responding more positively that hospitalists should provide postdischarge care (P = 0.054). Few responding hospitalists had such exposure (8.8%) at the time of the survey. Although 31% had considering starting a PDC, only 3% were starting in the next year. Of responding hospitalists with exposure to a PDC, 70% were satisfied with the experience and 85% felt their patients were satisfied. Responses did not vary by type of practice (academic vs nonacademic), group size, geographic location, or by exposure to a PDC except as above.

DISCUSSION

Responding hospitalists reported encountering significant difficulty arranging appropriate postdischarge appointments with primary care providers and feel this contributes to postdischarge complications. Nearly 75% of those surveyed felt a hospitalist‐run PDC would be effective in reducing postdischarge emergency department visits, presumably in part due to improved access to postdischarge care. However, 62% of responding hospitalists opposed providing this type of care, though those who had experienced a PDC were somewhat more likely to view providing care in a PDC favorably. Survey responses largely reflect attitudes rather than experience with PDCs, because very few respondents had ever worked in a PDC.

The juxtaposition of the confidence expressed in PDCs to reduce postdischarge emergency department visits with the less enthusiastic views of respondents about providing care in a PDC was surprising. Several explanations are possible. First, providing such care is outside the usual scope of practice of most hospitalists, and preliminary reports indicate hospitalists, as self‐selected inpatient providers, may not initially welcome this opportunity.[12] Second, responding hospitalists identified the need for extra compensation for providing this care, suggesting they would see staffing a PDC as a burden requiring extra payment. Third, only 12% of respondents felt their responsibility to their discharged patients extended to 1 month following discharge. Given this, hospitalists may not feel enough personal ownership over 30‐day readmission rates to justify the additional clinical demand of staffing a PDC.[13]

In fact, 29% of responding hospitalists felt their responsibility to the patient ended at the time of discharge. Respondents may have interpreted responsibility differently, and we cannot rule out response bias given our lower‐than‐expected response rate. However, we had anticipated many fewer hospitalists would respond this way given professional hospitalist societies have endorsed guidelines for improved transitions of care, which clearly delineate the key role hospitalists play in care transitions.[14]

Although fewer than 10% of respondents had worked in a PDC, nearly one‐third reported considering starting such a practice in the future, underscoring the importance of understanding hospitalist attitudes and experiences when creating a PDC and the significant barriers to arranging appropriate postdischarge care identified by survey respondents. The barriers to establishing a PDC may explain why few planned to start a PDC in the next year.

This study should be interpreted in the context of its design. Due to limitations in survey delivery, more rigorous sampling designs could not be used, and efforts were instead made to deliver the survey to a diverse group of hospitalists. The survey response rate was lower than anticipated and this increases the risk of response bias. Though this response rate is characteristic of other surveys of hospitalists, responses may have been from a selected population and therefore not representative of all hospitalists.[15] We sampled from a variety of practice venues, locations, academic and community practices, and practice group sizes to try to minimize this bias. Due to the low exposure rate to PDCs, hospitalist responses to experiences with PDCs should be considered exploratory.

We asked about similar content areas in the survey in multiple questions to maximize content validity; this resulted in variations in the degree of agreement or disagreement to similar prompts. For example, 62% of hospitalists opposed seeing patients in the clinic after discharge when directly asked, but nearly 50% said they would welcome the opportunity to work in a PDC if their employer required it. In another example, 50% of respondents said their responsibility for the patient ended at time of discharge, but when asked about duration of responsibility, 30% identified time of discharge as the limit. When reporting and interpreting results, we have tried to highlight responses to questions that ask most clearly and directly about the content of interest (rather than general themes), but this interpretation may also be subject to bias.

The time after hospital discharge is one of heightened risk for adverse events for the recently discharged patient. Hospitalist‐run postdischarge clinics may offer improved postdischarge care access and continuity; more research is needed on the effects of such clinics on patient outcomes, including postdischarge utilization. Until then, physicians and hospitals considering establishing PDCs should consider the barriers responding hospitalists identified to working in such a clinic, as well as the confidence they expressed in PDCs to reduce subsequent utilization.

Disclosures: Dr. Burke had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Veterans Affairs. Dr. Burke was supported by grant funding from the Colorado Research Enhancement Award Program to Improve Care Coordination for Veterans. Dr. Ryan has no conflicts of interest to disclose.

Healthcare‐associated infections affect 5% to 10% of all hospitalized patients each year in the United States, account for nearly $45 billion in direct hospital costs, and cause nearly 100,000 deaths annually.[1, 2] Because catheter‐associated urinary tract infection (CAUTI) is one of the most common healthcare‐associated infections in the United States and is reasonably preventable, the Centers for Medicare and Medicaid Services stopped reimbursing hospitals in 2008 for the additional costs of caring for patients who develop CAUTI during hospitalization.[3] Still, strategies for reducing inappropriate urinary catheterization are infrequently implemented in practice; this is despite a consensus that such strategies are effective.[4]

To help motivate hospitals to reduce inappropriate urinary catheter use, we present a tool for estimating costs of CAUTI for individual hospitals. Although other tools for estimating the excess costs of healthcare‐associated infections are available (eg, the APIC Cost of Healthcare‐Associated Infections Model available at http://www.apic.org/Resources/Cost‐calculators), they do not provide estimates of potential cost savings. Our approach adds to the literature by providing estimates of a hospital's current costs based on a few simple inputs (eg, annual admissions and catheterization rate), and also yields projected costs after a hypothetical intervention to prevent infections. Results are derived by combining appropriate cost and risk estimates from the literature. Importantly, an online implementation of our approach is available that can be easily used by clinicians, hospital administrators, and national policymakers. Our implementation nicely complements efforts like the Society of Hospital Medicine's Choosing Wisely campaign, which highlights avoiding inappropriate urinary catheter use first on its list of Five Things Physicians and Patients Should Question, and aims to increase awareness about issues that could improve patient outcomes and reduce healthcare costs.[5] Although accounting for the full spectrum of institution‐specific costs (eg, actual intervention costs, opportunity costs) was beyond the scope of this work, the simple tool we present helps meet the primary goal of generating an awareness of the potential cost savings stemming from CAUTI prevention efforts.

METHODS

General Setup

We consider 4 possible events after urinary catheter placement: bacteriuria, symptomatic urinary tract infection (SUTI), bloodstream infection (BSI), and catheter removal. Conservatively, assuming that bacteriuria must precede SUTI and BSI, there are 5 possible trajectories for any hospitalized patient (Figure 1): (1) no infection, (2) only bacteriuria, (3) bacteriuria and SUTI, (4) bacteriuria and BSI, or (5) bacteriuria, SUTI, and BSI. The cost of CAUTI for a particular hospital is therefore the per‐patient cost of each trajectory multiplied by the number of patients experiencing each trajectory. Our approach for estimating hospital costs is based on factorizing the number of patients experiencing each trajectory into a product of terms for which estimates are available from the literature (see the Supporting Information, Appendix, in the online version of this article for all technical details).

Figure 1

RESULTS

Figure 2 shows the projected savings in hospital costs due to CAUTI across a range of interventions defined by percent decreases in placement and duration, for a hypothetical hospital with 3000 total patients, 15% with urinary catheters preintervention, and with all other default values listed in Table 1. The current costs for this hospital (ie, the costs when the percent reduction in placement and duration is zero) are estimated to be $37,868 (95% confidence interval [CI]: $9159‐$156,564). After an intervention resulting in 40% reductions in both urinary catheter placement and duration, this hospital would be expected to save $22,653 (95% CI: $5479‐$93,656). A less effective intervention yielding a 10% reduction in both urinary catheter placement and duration would result in more modest savings of $6376 (95% CI: $1542‐$26,360).

Figure 2

After an intervention resulting in 29% and 37% reductions in placement and duration, respectively, reflecting reductions seen in practice,[10, 11] our hypothetical hospital is estimated to save $19,126 (95% CI: $4626‐$79,074). This reflects an estimated savings of nearly 50%.

DISCUSSION

We have presented a tool for estimating customized hospital costs of CAUTI, both before and after a hypothetical intervention to reduce risk of infection. Our approach relies on mostly conservative assumptions, incorporates published risk estimates (properly accounting for their associated variability), and has easy‐to‐use online implementation. We believe this can play an important role in motivating hospitals to reduce inappropriate urinary catheter use.

The methodology employed here does have a few limitations. First and foremost, our results depend on the reliability of the input values, which are either provided by users or are based on estimates from the literature (see Table 1 for a complete list of suggested defaults). New information could potentially be incorporated if and when available. For example, substitution of more precise risk estimates could help reduce confidence interval length. Second, our approach essentially averages over hospital quality; we do not directly take into account quality of care or variation in underlying infection risk across hospitals in computing estimated costs. Finally, we only compute direct costs due to infection; other costs (eg, intervention costs) would typically also need to be considered for decision making.

Despite these limitations, we believe that our tool can help infection control professionals demonstrate the values of CAUTI prevention efforts to key administrators, particularly at a time where it has become increasingly necessary to develop a business case to initiate new interventions or justify the continued support for ongoing programs.[12] Additionally, we believe the proposed approach can be an important supplement to initiatives like the Society of Hospital Medicine's Choosing Wisely campaign, which aims to help reduce inappropriate urinary catheter use. Reducing catheter utilization has the potential to reduce costs associated with caring for CAUTI patients, but more importantly would help reduce CAUTI incidence as well as catheter‐related, noninfectious complications.[13, 14] We hope that our tool will greatly assist hospitals in promoting their CAUTI prevention efforts and improve the overall safety of hospitalized patients.

Healthcare‐associated infections affect 5% to 10% of all hospitalized patients each year in the United States, account for nearly $45 billion in direct hospital costs, and cause nearly 100,000 deaths annually.[1, 2] Because catheter‐associated urinary tract infection (CAUTI) is one of the most common healthcare‐associated infections in the United States and is reasonably preventable, the Centers for Medicare and Medicaid Services stopped reimbursing hospitals in 2008 for the additional costs of caring for patients who develop CAUTI during hospitalization.[3] Still, strategies for reducing inappropriate urinary catheterization are infrequently implemented in practice; this is despite a consensus that such strategies are effective.[4]

To help motivate hospitals to reduce inappropriate urinary catheter use, we present a tool for estimating costs of CAUTI for individual hospitals. Although other tools for estimating the excess costs of healthcare‐associated infections are available (eg, the APIC Cost of Healthcare‐Associated Infections Model available at http://www.apic.org/Resources/Cost‐calculators), they do not provide estimates of potential cost savings. Our approach adds to the literature by providing estimates of a hospital's current costs based on a few simple inputs (eg, annual admissions and catheterization rate), and also yields projected costs after a hypothetical intervention to prevent infections. Results are derived by combining appropriate cost and risk estimates from the literature. Importantly, an online implementation of our approach is available that can be easily used by clinicians, hospital administrators, and national policymakers. Our implementation nicely complements efforts like the Society of Hospital Medicine's Choosing Wisely campaign, which highlights avoiding inappropriate urinary catheter use first on its list of Five Things Physicians and Patients Should Question, and aims to increase awareness about issues that could improve patient outcomes and reduce healthcare costs.[5] Although accounting for the full spectrum of institution‐specific costs (eg, actual intervention costs, opportunity costs) was beyond the scope of this work, the simple tool we present helps meet the primary goal of generating an awareness of the potential cost savings stemming from CAUTI prevention efforts.

METHODS

General Setup

We consider 4 possible events after urinary catheter placement: bacteriuria, symptomatic urinary tract infection (SUTI), bloodstream infection (BSI), and catheter removal. Conservatively, assuming that bacteriuria must precede SUTI and BSI, there are 5 possible trajectories for any hospitalized patient (Figure 1): (1) no infection, (2) only bacteriuria, (3) bacteriuria and SUTI, (4) bacteriuria and BSI, or (5) bacteriuria, SUTI, and BSI. The cost of CAUTI for a particular hospital is therefore the per‐patient cost of each trajectory multiplied by the number of patients experiencing each trajectory. Our approach for estimating hospital costs is based on factorizing the number of patients experiencing each trajectory into a product of terms for which estimates are available from the literature (see the Supporting Information, Appendix, in the online version of this article for all technical details).

Figure 1

RESULTS

Figure 2 shows the projected savings in hospital costs due to CAUTI across a range of interventions defined by percent decreases in placement and duration, for a hypothetical hospital with 3000 total patients, 15% with urinary catheters preintervention, and with all other default values listed in Table 1. The current costs for this hospital (ie, the costs when the percent reduction in placement and duration is zero) are estimated to be $37,868 (95% confidence interval [CI]: $9159‐$156,564). After an intervention resulting in 40% reductions in both urinary catheter placement and duration, this hospital would be expected to save $22,653 (95% CI: $5479‐$93,656). A less effective intervention yielding a 10% reduction in both urinary catheter placement and duration would result in more modest savings of $6376 (95% CI: $1542‐$26,360).

Figure 2

After an intervention resulting in 29% and 37% reductions in placement and duration, respectively, reflecting reductions seen in practice,[10, 11] our hypothetical hospital is estimated to save $19,126 (95% CI: $4626‐$79,074). This reflects an estimated savings of nearly 50%.

DISCUSSION

We have presented a tool for estimating customized hospital costs of CAUTI, both before and after a hypothetical intervention to reduce risk of infection. Our approach relies on mostly conservative assumptions, incorporates published risk estimates (properly accounting for their associated variability), and has easy‐to‐use online implementation. We believe this can play an important role in motivating hospitals to reduce inappropriate urinary catheter use.

The methodology employed here does have a few limitations. First and foremost, our results depend on the reliability of the input values, which are either provided by users or are based on estimates from the literature (see Table 1 for a complete list of suggested defaults). New information could potentially be incorporated if and when available. For example, substitution of more precise risk estimates could help reduce confidence interval length. Second, our approach essentially averages over hospital quality; we do not directly take into account quality of care or variation in underlying infection risk across hospitals in computing estimated costs. Finally, we only compute direct costs due to infection; other costs (eg, intervention costs) would typically also need to be considered for decision making.

Despite these limitations, we believe that our tool can help infection control professionals demonstrate the values of CAUTI prevention efforts to key administrators, particularly at a time where it has become increasingly necessary to develop a business case to initiate new interventions or justify the continued support for ongoing programs.[12] Additionally, we believe the proposed approach can be an important supplement to initiatives like the Society of Hospital Medicine's Choosing Wisely campaign, which aims to help reduce inappropriate urinary catheter use. Reducing catheter utilization has the potential to reduce costs associated with caring for CAUTI patients, but more importantly would help reduce CAUTI incidence as well as catheter‐related, noninfectious complications.[13, 14] We hope that our tool will greatly assist hospitals in promoting their CAUTI prevention efforts and improve the overall safety of hospitalized patients.

Healthcare‐associated infections affect 5% to 10% of all hospitalized patients each year in the United States, account for nearly $45 billion in direct hospital costs, and cause nearly 100,000 deaths annually.[1, 2] Because catheter‐associated urinary tract infection (CAUTI) is one of the most common healthcare‐associated infections in the United States and is reasonably preventable, the Centers for Medicare and Medicaid Services stopped reimbursing hospitals in 2008 for the additional costs of caring for patients who develop CAUTI during hospitalization.[3] Still, strategies for reducing inappropriate urinary catheterization are infrequently implemented in practice; this is despite a consensus that such strategies are effective.[4]

To help motivate hospitals to reduce inappropriate urinary catheter use, we present a tool for estimating costs of CAUTI for individual hospitals. Although other tools for estimating the excess costs of healthcare‐associated infections are available (eg, the APIC Cost of Healthcare‐Associated Infections Model available at http://www.apic.org/Resources/Cost‐calculators), they do not provide estimates of potential cost savings. Our approach adds to the literature by providing estimates of a hospital's current costs based on a few simple inputs (eg, annual admissions and catheterization rate), and also yields projected costs after a hypothetical intervention to prevent infections. Results are derived by combining appropriate cost and risk estimates from the literature. Importantly, an online implementation of our approach is available that can be easily used by clinicians, hospital administrators, and national policymakers. Our implementation nicely complements efforts like the Society of Hospital Medicine's Choosing Wisely campaign, which highlights avoiding inappropriate urinary catheter use first on its list of Five Things Physicians and Patients Should Question, and aims to increase awareness about issues that could improve patient outcomes and reduce healthcare costs.[5] Although accounting for the full spectrum of institution‐specific costs (eg, actual intervention costs, opportunity costs) was beyond the scope of this work, the simple tool we present helps meet the primary goal of generating an awareness of the potential cost savings stemming from CAUTI prevention efforts.

METHODS

General Setup

We consider 4 possible events after urinary catheter placement: bacteriuria, symptomatic urinary tract infection (SUTI), bloodstream infection (BSI), and catheter removal. Conservatively, assuming that bacteriuria must precede SUTI and BSI, there are 5 possible trajectories for any hospitalized patient (Figure 1): (1) no infection, (2) only bacteriuria, (3) bacteriuria and SUTI, (4) bacteriuria and BSI, or (5) bacteriuria, SUTI, and BSI. The cost of CAUTI for a particular hospital is therefore the per‐patient cost of each trajectory multiplied by the number of patients experiencing each trajectory. Our approach for estimating hospital costs is based on factorizing the number of patients experiencing each trajectory into a product of terms for which estimates are available from the literature (see the Supporting Information, Appendix, in the online version of this article for all technical details).

Figure 1

RESULTS

Figure 2 shows the projected savings in hospital costs due to CAUTI across a range of interventions defined by percent decreases in placement and duration, for a hypothetical hospital with 3000 total patients, 15% with urinary catheters preintervention, and with all other default values listed in Table 1. The current costs for this hospital (ie, the costs when the percent reduction in placement and duration is zero) are estimated to be $37,868 (95% confidence interval [CI]: $9159‐$156,564). After an intervention resulting in 40% reductions in both urinary catheter placement and duration, this hospital would be expected to save $22,653 (95% CI: $5479‐$93,656). A less effective intervention yielding a 10% reduction in both urinary catheter placement and duration would result in more modest savings of $6376 (95% CI: $1542‐$26,360).

Figure 2

After an intervention resulting in 29% and 37% reductions in placement and duration, respectively, reflecting reductions seen in practice,[10, 11] our hypothetical hospital is estimated to save $19,126 (95% CI: $4626‐$79,074). This reflects an estimated savings of nearly 50%.

DISCUSSION

We have presented a tool for estimating customized hospital costs of CAUTI, both before and after a hypothetical intervention to reduce risk of infection. Our approach relies on mostly conservative assumptions, incorporates published risk estimates (properly accounting for their associated variability), and has easy‐to‐use online implementation. We believe this can play an important role in motivating hospitals to reduce inappropriate urinary catheter use.

The methodology employed here does have a few limitations. First and foremost, our results depend on the reliability of the input values, which are either provided by users or are based on estimates from the literature (see Table 1 for a complete list of suggested defaults). New information could potentially be incorporated if and when available. For example, substitution of more precise risk estimates could help reduce confidence interval length. Second, our approach essentially averages over hospital quality; we do not directly take into account quality of care or variation in underlying infection risk across hospitals in computing estimated costs. Finally, we only compute direct costs due to infection; other costs (eg, intervention costs) would typically also need to be considered for decision making.

Despite these limitations, we believe that our tool can help infection control professionals demonstrate the values of CAUTI prevention efforts to key administrators, particularly at a time where it has become increasingly necessary to develop a business case to initiate new interventions or justify the continued support for ongoing programs.[12] Additionally, we believe the proposed approach can be an important supplement to initiatives like the Society of Hospital Medicine's Choosing Wisely campaign, which aims to help reduce inappropriate urinary catheter use. Reducing catheter utilization has the potential to reduce costs associated with caring for CAUTI patients, but more importantly would help reduce CAUTI incidence as well as catheter‐related, noninfectious complications.[13, 14] We hope that our tool will greatly assist hospitals in promoting their CAUTI prevention efforts and improve the overall safety of hospitalized patients.

For at least 25 years, approximately 20% of Medicare fee‐for‐service discharges have been followed by a hospital readmission within 30 days.[1, 2] Section 3025 of the Patient Protection and Affordable Care Act (ACA)[3] created escalating penalties for hospitals with higher than expected 30‐day readmission rates, and the Congressional Budget Office estimated this will reduce Medicare spending by over $7 billion between 2010 and 2019.[4]

Hospitals and physicians have begun developing strategies to identify which Medicare beneficiaries are most likely to be readmitted and use this information to design programs to reduce their readmission rate. Initially, penalties will be based on readmission rates after an index discharge with heart failure, myocardial infarction, and pneumonia.[5] Recently, the Centers for Medicare and Medicaid Services (CMS) released the Inpatient Prospective Payment System FY2014 proposed rule, which proposes to add 2 new readmission penalties beginning in FY2015: readmissions for hip/knee arthroplasty and chronic obstructive pulmonary disease.[6] Other countries are already penalizing hospitals with high readmission rates; for example, Germany is penalizing all readmissions that occur within a 30‐day period following admission.[7] In this brief report, we examine the characteristics of Medicare beneficiaries most likely to be readmitted within 30 days. We focus on readmission rates for all discharge conditions and all patient readmission rates, because we believe the language in the ACA ultimately points to an all‐inclusive approach.

METHODS

We used a nationally random 5% sample of all Medicare beneficiaries for the period between January 1, 2008 and September 30, 2008. To be included, beneficiaries must have both Part A and B coverage and live within the United States. Medicare Advantage patients were excluded because Medicare Advantage plans do not report the data in the same way as fee for service. We calculated the readmission rate as the number of admissions that were preceded by an at‐risk discharge within 30 days divided by the total number of at‐risk discharges. This definition included admissions to and discharges from sole community providers, Medicare‐dependent small rural hospitals, and critical access hospitals. We counted as at risk all live discharges from short‐term acute care hospitals that were not discharged against medical advice, discharged to a rehabilitation unit within an acute care hospital, or readmitted on day 0 (due to inconsistency with use of transfer coding). We only included discharges and readmissions to acute care hospitals and excluded hospitalizations in long‐term care facilities, rehabilitation facilities, skilled nursing homes, and other non‐acute care hospital facilities from being an index hospitalization. However, if the beneficiary was discharged to 1 of these facilities and then readmitted to an acute care hospital, the readmission was counted.

Each discharge was recorded as an independent event and we reset the readmission clock for a fresh 30‐day count each time the beneficiary was discharged. We examined the admission and readmission rate to determine if the rates varied by age, gender, reason for entitlement, racial characteristics, region of the country, number of chronic conditions, and whether the beneficiary is also enrolled in Medicaid (dual eligibles). We calculated the mean readmission rate for each diagnosis‐related group (DRG) and then used the probability of having a readmission for each DRG to calculate a case mix adjustment for each hospital. To calculate the chronic illness burden, we used a previously developed methodology for counting the number of chronic disease categories reported for the patient in the preceding year (2007).[8, 9] The classification system is maintained by the Agency for Health Care Research and Quality. We then used logistic regression to calculate the odds ratio of a discharge being readmitted based on these factors. We preformed statistical analysis using SAS version 9.1.3 (SAS Institute Inc., Cary, NC).

RESULTS

There were 434,999 hospital discharges that occurred in the first 9 months of 2008 in the 5% sample. There were 20.6% of Medicare beneficiaries hospitalized, and the overall readmission rate was 19.5%. Table 1 shows the odds ratios and 95% confidence intervals for the probability that a Medicare beneficiary will be readmitted within 30 days for variables including: age, sex, race, dual‐eligibility status, number of comorbid conditions, geographic region, and reason for entitlement. Of note, beneficiaries with 10 or more chronic conditions were more than 6 times more likely, and beneficiaries with 5 to 9 chronic conditions were more than 2.5 times more likely, to be readmitted than beneficiaries with 1 to 4 chronic conditions.

DISCUSSION

The most interesting finding is that beneficiaries with 10 or more chronic conditions were more than 6 times more likely to be readmitted than beneficiaries with 1 to 4 chronic conditions. Beneficiaries with 10 or more chronic conditions represent only 8.9% of all Medicare beneficiaries (31.0% of all hospitalizations), but they were responsible for 50.2% of all readmissions. The 31.8% of beneficiaries with 5 to 9 chronic conditions (55.5% of all hospitalizations) had the second highest odds ratio (2.5) and were responsible for 45% of all readmissions. The 59.3% of beneficiaries with 5 comorbidities (13.6% of all hospitalizations) were associated with only 4.7% of all readmissions. This strongly suggests that hospitals focus their attention on beneficiaries with 10 or more comorbidities. These results were despite correction for DRG diagnosis in the model.

We recognize that the number of chronic conditions is a crude measure of health status because it weighs hundreds of different clinical conditions equally; however, it seems a good proxy for 3 closely allied concepts: (1) the overall burden of chronic illness carried by the patient, (2) the patient's level of engagement with the healthcare system (including number of unique providers), and (3) the number of conditions being treated. By providing a 1‐year window of a patient's health status, it is a more complete picture than any single hospital claim submission or indices based solely on hospital discharge data.

The other variables are less predictive of 30‐day readmissions. Beneficiaries over 85 years old are only 14% more likely, whereas disabled Medicare beneficiaries 44 years old are 63% more likely to be readmitted than beneficiaries between 65 and 74 years old. Men are 20% more likely to be readmitted than women. Black race and dual‐eligibility slightly increase rates of readmission. Beneficiaries located in the West have the lowest readmission rates. In comparison to those who are aged, those with end‐stage renal disease (ESRD) have a higher rate of readmission, and those with a disability have a lower rate of readmission. In considering the age and reason for entitlement findings, one would assume that ESRD was the driver of higher readmission rates in the younger Medicare population.

CMS will need to analyze which hospitals have higher than expected readmission rates, and this will require risk adjustment at each hospital. In addition to the number of chronic conditions and other variables shown in Table 1, other factors CMS might want to include when it starts doing readmissions for all discharges is the discharge diagnosis (because our results suggest there are significant differences in the probability of a readmission across DRGs). In addition, CMS will need to consider how to capture additional data not currently in the claims data, such as social factors like homelessness.

We recognize significant limitations to these findings. First, this analysis uses only information that is available from Medicare claims and administrative data. Claims give almost no information on how well the hospital planned the discharge, instructed the patient and family, or engaged follow‐up providers. Also, claims data tell us virtually nothing about a patient's health literacy or social situation. Second, the analysis relies on claims data, but this has little clinical detail. Third, these data are limited to persons enrolled in fee‐for‐service Medicare. Fourth, we included all readmissions, including some readmissions (such as chemotherapy and staged percutaneous coronary interventions) that were part of a planned treatment protocol.[10] Fifth, we were unable to distinguish same‐day readmissions versus transfers, and therefore excluded all same‐day readmissions from measurement.

As hospitals and physicians begin to plan for the regulations that will penalize hospitals with high readmission rates, they will need to strongly consider targeting beneficiaries with more than 10 chronic conditions.

For at least 25 years, approximately 20% of Medicare fee‐for‐service discharges have been followed by a hospital readmission within 30 days.[1, 2] Section 3025 of the Patient Protection and Affordable Care Act (ACA)[3] created escalating penalties for hospitals with higher than expected 30‐day readmission rates, and the Congressional Budget Office estimated this will reduce Medicare spending by over $7 billion between 2010 and 2019.[4]

Hospitals and physicians have begun developing strategies to identify which Medicare beneficiaries are most likely to be readmitted and use this information to design programs to reduce their readmission rate. Initially, penalties will be based on readmission rates after an index discharge with heart failure, myocardial infarction, and pneumonia.[5] Recently, the Centers for Medicare and Medicaid Services (CMS) released the Inpatient Prospective Payment System FY2014 proposed rule, which proposes to add 2 new readmission penalties beginning in FY2015: readmissions for hip/knee arthroplasty and chronic obstructive pulmonary disease.[6] Other countries are already penalizing hospitals with high readmission rates; for example, Germany is penalizing all readmissions that occur within a 30‐day period following admission.[7] In this brief report, we examine the characteristics of Medicare beneficiaries most likely to be readmitted within 30 days. We focus on readmission rates for all discharge conditions and all patient readmission rates, because we believe the language in the ACA ultimately points to an all‐inclusive approach.

METHODS

We used a nationally random 5% sample of all Medicare beneficiaries for the period between January 1, 2008 and September 30, 2008. To be included, beneficiaries must have both Part A and B coverage and live within the United States. Medicare Advantage patients were excluded because Medicare Advantage plans do not report the data in the same way as fee for service. We calculated the readmission rate as the number of admissions that were preceded by an at‐risk discharge within 30 days divided by the total number of at‐risk discharges. This definition included admissions to and discharges from sole community providers, Medicare‐dependent small rural hospitals, and critical access hospitals. We counted as at risk all live discharges from short‐term acute care hospitals that were not discharged against medical advice, discharged to a rehabilitation unit within an acute care hospital, or readmitted on day 0 (due to inconsistency with use of transfer coding). We only included discharges and readmissions to acute care hospitals and excluded hospitalizations in long‐term care facilities, rehabilitation facilities, skilled nursing homes, and other non‐acute care hospital facilities from being an index hospitalization. However, if the beneficiary was discharged to 1 of these facilities and then readmitted to an acute care hospital, the readmission was counted.

Each discharge was recorded as an independent event and we reset the readmission clock for a fresh 30‐day count each time the beneficiary was discharged. We examined the admission and readmission rate to determine if the rates varied by age, gender, reason for entitlement, racial characteristics, region of the country, number of chronic conditions, and whether the beneficiary is also enrolled in Medicaid (dual eligibles). We calculated the mean readmission rate for each diagnosis‐related group (DRG) and then used the probability of having a readmission for each DRG to calculate a case mix adjustment for each hospital. To calculate the chronic illness burden, we used a previously developed methodology for counting the number of chronic disease categories reported for the patient in the preceding year (2007).[8, 9] The classification system is maintained by the Agency for Health Care Research and Quality. We then used logistic regression to calculate the odds ratio of a discharge being readmitted based on these factors. We preformed statistical analysis using SAS version 9.1.3 (SAS Institute Inc., Cary, NC).

RESULTS

There were 434,999 hospital discharges that occurred in the first 9 months of 2008 in the 5% sample. There were 20.6% of Medicare beneficiaries hospitalized, and the overall readmission rate was 19.5%. Table 1 shows the odds ratios and 95% confidence intervals for the probability that a Medicare beneficiary will be readmitted within 30 days for variables including: age, sex, race, dual‐eligibility status, number of comorbid conditions, geographic region, and reason for entitlement. Of note, beneficiaries with 10 or more chronic conditions were more than 6 times more likely, and beneficiaries with 5 to 9 chronic conditions were more than 2.5 times more likely, to be readmitted than beneficiaries with 1 to 4 chronic conditions.

DISCUSSION

The most interesting finding is that beneficiaries with 10 or more chronic conditions were more than 6 times more likely to be readmitted than beneficiaries with 1 to 4 chronic conditions. Beneficiaries with 10 or more chronic conditions represent only 8.9% of all Medicare beneficiaries (31.0% of all hospitalizations), but they were responsible for 50.2% of all readmissions. The 31.8% of beneficiaries with 5 to 9 chronic conditions (55.5% of all hospitalizations) had the second highest odds ratio (2.5) and were responsible for 45% of all readmissions. The 59.3% of beneficiaries with 5 comorbidities (13.6% of all hospitalizations) were associated with only 4.7% of all readmissions. This strongly suggests that hospitals focus their attention on beneficiaries with 10 or more comorbidities. These results were despite correction for DRG diagnosis in the model.

We recognize that the number of chronic conditions is a crude measure of health status because it weighs hundreds of different clinical conditions equally; however, it seems a good proxy for 3 closely allied concepts: (1) the overall burden of chronic illness carried by the patient, (2) the patient's level of engagement with the healthcare system (including number of unique providers), and (3) the number of conditions being treated. By providing a 1‐year window of a patient's health status, it is a more complete picture than any single hospital claim submission or indices based solely on hospital discharge data.

The other variables are less predictive of 30‐day readmissions. Beneficiaries over 85 years old are only 14% more likely, whereas disabled Medicare beneficiaries 44 years old are 63% more likely to be readmitted than beneficiaries between 65 and 74 years old. Men are 20% more likely to be readmitted than women. Black race and dual‐eligibility slightly increase rates of readmission. Beneficiaries located in the West have the lowest readmission rates. In comparison to those who are aged, those with end‐stage renal disease (ESRD) have a higher rate of readmission, and those with a disability have a lower rate of readmission. In considering the age and reason for entitlement findings, one would assume that ESRD was the driver of higher readmission rates in the younger Medicare population.

CMS will need to analyze which hospitals have higher than expected readmission rates, and this will require risk adjustment at each hospital. In addition to the number of chronic conditions and other variables shown in Table 1, other factors CMS might want to include when it starts doing readmissions for all discharges is the discharge diagnosis (because our results suggest there are significant differences in the probability of a readmission across DRGs). In addition, CMS will need to consider how to capture additional data not currently in the claims data, such as social factors like homelessness.

We recognize significant limitations to these findings. First, this analysis uses only information that is available from Medicare claims and administrative data. Claims give almost no information on how well the hospital planned the discharge, instructed the patient and family, or engaged follow‐up providers. Also, claims data tell us virtually nothing about a patient's health literacy or social situation. Second, the analysis relies on claims data, but this has little clinical detail. Third, these data are limited to persons enrolled in fee‐for‐service Medicare. Fourth, we included all readmissions, including some readmissions (such as chemotherapy and staged percutaneous coronary interventions) that were part of a planned treatment protocol.[10] Fifth, we were unable to distinguish same‐day readmissions versus transfers, and therefore excluded all same‐day readmissions from measurement.

As hospitals and physicians begin to plan for the regulations that will penalize hospitals with high readmission rates, they will need to strongly consider targeting beneficiaries with more than 10 chronic conditions.

For at least 25 years, approximately 20% of Medicare fee‐for‐service discharges have been followed by a hospital readmission within 30 days.[1, 2] Section 3025 of the Patient Protection and Affordable Care Act (ACA)[3] created escalating penalties for hospitals with higher than expected 30‐day readmission rates, and the Congressional Budget Office estimated this will reduce Medicare spending by over $7 billion between 2010 and 2019.[4]

Hospitals and physicians have begun developing strategies to identify which Medicare beneficiaries are most likely to be readmitted and use this information to design programs to reduce their readmission rate. Initially, penalties will be based on readmission rates after an index discharge with heart failure, myocardial infarction, and pneumonia.[5] Recently, the Centers for Medicare and Medicaid Services (CMS) released the Inpatient Prospective Payment System FY2014 proposed rule, which proposes to add 2 new readmission penalties beginning in FY2015: readmissions for hip/knee arthroplasty and chronic obstructive pulmonary disease.[6] Other countries are already penalizing hospitals with high readmission rates; for example, Germany is penalizing all readmissions that occur within a 30‐day period following admission.[7] In this brief report, we examine the characteristics of Medicare beneficiaries most likely to be readmitted within 30 days. We focus on readmission rates for all discharge conditions and all patient readmission rates, because we believe the language in the ACA ultimately points to an all‐inclusive approach.

METHODS

We used a nationally random 5% sample of all Medicare beneficiaries for the period between January 1, 2008 and September 30, 2008. To be included, beneficiaries must have both Part A and B coverage and live within the United States. Medicare Advantage patients were excluded because Medicare Advantage plans do not report the data in the same way as fee for service. We calculated the readmission rate as the number of admissions that were preceded by an at‐risk discharge within 30 days divided by the total number of at‐risk discharges. This definition included admissions to and discharges from sole community providers, Medicare‐dependent small rural hospitals, and critical access hospitals. We counted as at risk all live discharges from short‐term acute care hospitals that were not discharged against medical advice, discharged to a rehabilitation unit within an acute care hospital, or readmitted on day 0 (due to inconsistency with use of transfer coding). We only included discharges and readmissions to acute care hospitals and excluded hospitalizations in long‐term care facilities, rehabilitation facilities, skilled nursing homes, and other non‐acute care hospital facilities from being an index hospitalization. However, if the beneficiary was discharged to 1 of these facilities and then readmitted to an acute care hospital, the readmission was counted.

Each discharge was recorded as an independent event and we reset the readmission clock for a fresh 30‐day count each time the beneficiary was discharged. We examined the admission and readmission rate to determine if the rates varied by age, gender, reason for entitlement, racial characteristics, region of the country, number of chronic conditions, and whether the beneficiary is also enrolled in Medicaid (dual eligibles). We calculated the mean readmission rate for each diagnosis‐related group (DRG) and then used the probability of having a readmission for each DRG to calculate a case mix adjustment for each hospital. To calculate the chronic illness burden, we used a previously developed methodology for counting the number of chronic disease categories reported for the patient in the preceding year (2007).[8, 9] The classification system is maintained by the Agency for Health Care Research and Quality. We then used logistic regression to calculate the odds ratio of a discharge being readmitted based on these factors. We preformed statistical analysis using SAS version 9.1.3 (SAS Institute Inc., Cary, NC).

RESULTS

There were 434,999 hospital discharges that occurred in the first 9 months of 2008 in the 5% sample. There were 20.6% of Medicare beneficiaries hospitalized, and the overall readmission rate was 19.5%. Table 1 shows the odds ratios and 95% confidence intervals for the probability that a Medicare beneficiary will be readmitted within 30 days for variables including: age, sex, race, dual‐eligibility status, number of comorbid conditions, geographic region, and reason for entitlement. Of note, beneficiaries with 10 or more chronic conditions were more than 6 times more likely, and beneficiaries with 5 to 9 chronic conditions were more than 2.5 times more likely, to be readmitted than beneficiaries with 1 to 4 chronic conditions.

DISCUSSION

The most interesting finding is that beneficiaries with 10 or more chronic conditions were more than 6 times more likely to be readmitted than beneficiaries with 1 to 4 chronic conditions. Beneficiaries with 10 or more chronic conditions represent only 8.9% of all Medicare beneficiaries (31.0% of all hospitalizations), but they were responsible for 50.2% of all readmissions. The 31.8% of beneficiaries with 5 to 9 chronic conditions (55.5% of all hospitalizations) had the second highest odds ratio (2.5) and were responsible for 45% of all readmissions. The 59.3% of beneficiaries with 5 comorbidities (13.6% of all hospitalizations) were associated with only 4.7% of all readmissions. This strongly suggests that hospitals focus their attention on beneficiaries with 10 or more comorbidities. These results were despite correction for DRG diagnosis in the model.

We recognize that the number of chronic conditions is a crude measure of health status because it weighs hundreds of different clinical conditions equally; however, it seems a good proxy for 3 closely allied concepts: (1) the overall burden of chronic illness carried by the patient, (2) the patient's level of engagement with the healthcare system (including number of unique providers), and (3) the number of conditions being treated. By providing a 1‐year window of a patient's health status, it is a more complete picture than any single hospital claim submission or indices based solely on hospital discharge data.

The other variables are less predictive of 30‐day readmissions. Beneficiaries over 85 years old are only 14% more likely, whereas disabled Medicare beneficiaries 44 years old are 63% more likely to be readmitted than beneficiaries between 65 and 74 years old. Men are 20% more likely to be readmitted than women. Black race and dual‐eligibility slightly increase rates of readmission. Beneficiaries located in the West have the lowest readmission rates. In comparison to those who are aged, those with end‐stage renal disease (ESRD) have a higher rate of readmission, and those with a disability have a lower rate of readmission. In considering the age and reason for entitlement findings, one would assume that ESRD was the driver of higher readmission rates in the younger Medicare population.

CMS will need to analyze which hospitals have higher than expected readmission rates, and this will require risk adjustment at each hospital. In addition to the number of chronic conditions and other variables shown in Table 1, other factors CMS might want to include when it starts doing readmissions for all discharges is the discharge diagnosis (because our results suggest there are significant differences in the probability of a readmission across DRGs). In addition, CMS will need to consider how to capture additional data not currently in the claims data, such as social factors like homelessness.

We recognize significant limitations to these findings. First, this analysis uses only information that is available from Medicare claims and administrative data. Claims give almost no information on how well the hospital planned the discharge, instructed the patient and family, or engaged follow‐up providers. Also, claims data tell us virtually nothing about a patient's health literacy or social situation. Second, the analysis relies on claims data, but this has little clinical detail. Third, these data are limited to persons enrolled in fee‐for‐service Medicare. Fourth, we included all readmissions, including some readmissions (such as chemotherapy and staged percutaneous coronary interventions) that were part of a planned treatment protocol.[10] Fifth, we were unable to distinguish same‐day readmissions versus transfers, and therefore excluded all same‐day readmissions from measurement.

As hospitals and physicians begin to plan for the regulations that will penalize hospitals with high readmission rates, they will need to strongly consider targeting beneficiaries with more than 10 chronic conditions.

Unprofessional behavior in the inpatient setting has the potential to impact care delivery and the quality of trainee's educational experience. These behaviors, from disparaging colleagues to blocking admissions, can negatively impact the learning environment. The learning environment or conditions created by the patient care team's actions play a critical role in the development of trainees.[1, 2] The rising presence of hospitalists in the inpatient setting raises the question of how their actions impact the learning environment. Professional behavior has been defined as a core competency for hospitalists by the Society of Hospital Medicine.[3] Professional behavior of all team members, from faculty to trainee, can impact the learning environment and patient safety.[4, 5] However, few educational materials exist to train faculty and housestaff on recognizing and ameliorating unprofessional behaviors.

A prior assessment regarding hospitalists' lapses in professionalism identified scenarios that demonstrated increased participation by hospitalists at 3 institutions.[6] Participants reported observation or participation in specific unprofessional behaviors and rated their perception of these behaviors. Additional work within those residency environments demonstrated that residents' perceptions of and participation in these behaviors increased throughout training, with environmental characteristics, specifically faculty behavior, influencing trainee professional development and acclimation of these behaviors.[7, 8]

Although overall participation in egregious behavior was low, resident participation in 3 categories of unprofessional behavior increased during internship. Those scenarios included disparaging the emergency room or primary care physician for missed findings or management decisions, blocking or not taking admissions appropriate for the service in question, and misrepresenting a test as urgent to expedite obtaining the test. We developed our intervention focused on these areas to address professionalism lapses that occur during internship. Our earlier work showed faculty role models influenced trainee behavior. For this reason, we provided education to both residents and hospitalists to maximize the impact of the intervention.

We present here a novel, interactive, video‐based workshop curriculum for faculty and trainees that aims to illustrate unprofessional behaviors and outlines the role faculty may play in promoting such behaviors. In addition, we review the result of postworkshop evaluation on intent to change behavior and satisfaction.

METHODS

A grant from the American Board of Internal Medicine Foundation supported this project. The working group that resulted, the Chicago Professional Practice Project and Outcomes, included faculty representation from 3 Chicago‐area hospitals: the University of Chicago, Northwestern University, and NorthShore University HealthSystem. Academic hospitalists at these sites were invited to participate. Each site also has an internal medicine residency program in which hospitalists were expected to attend the teaching service. Given this, resident trainees at all participating sites, and 1 community teaching affiliate program (Mercy Hospital and Medical Center) where academic hospitalists at the University of Chicago rotate, were recruited for participation. Faculty champions were identified for each site, and 1 internal and external faculty representative from the working group served to debrief and facilitate. Trainee workshops were administered by 1 internal and external collaborator, and for the community site, 2 external faculty members. Workshops were held during established educational conference times, and lunch was provided.

Scripts highlighting each of the behaviors identified in the prior survey were developed and peer reviewed for clarity and face validity across the 3 sites. Medical student and resident actors were trained utilizing the finalized scripts, and a performance artist affiliated with the Screen Actors Guild assisted in their preparation for filming. All videos were filmed at the University of Chicago Pritzker School of Medicine Clinical Performance Center. The final videos ranged in length from 4 to 7 minutes and included title, cast, and funding source. As an example, 1 video highlighted the unprofessional behavior of misrepresenting a test as urgent to prioritize one's patient in the queue. This video included a resident, intern, and attending on inpatient rounds during which the resident encouraged the intern to misrepresent the patient's status to expedite obtaining the study and facilitate the patient's discharge. The resident stressed that he would be in the clinic and had many patients to see, highlighting the impact of workload on unprofessional behavior, and aggressively persuaded the intern to sell her test to have it performed the same day. When this occurred, the attending applauded the intern for her strong work.

A moderator guide and debriefing tools were developed to facilitate discussion. The duration of each of the workshops was approximately 60 minutes. After welcoming remarks, participants were provided tools to utilize during the viewing of each video. These checklists noted the roles of those depicted in the video, asked to identify positive or negative behaviors displayed, and included questions regarding how behaviors could be detrimental and how the situation could have been prevented. After viewing the videos, participants divided into small groups to discuss the individual exhibiting the unprofessional behavior, their perceived motivation for said behavior, and its impact on the team culture and patient care. Following a small‐group discussion, large‐group debriefing was performed, addressing the barriers and facilitators to professional behavior. Two videos were shown at each workshop, and participants completed a postworkshop evaluation. Videos chosen for viewing were based upon preworkshop survey results that highlighted areas of concern at that specific site.

Postworkshop paper‐based evaluations assessed participants' perception of displayed behaviors on a Likert‐type scale (1=unprofessional to 5=professional) utilizing items validated in prior work,[6, 7, 8] their level of agreement regarding the impact of video‐based exercises, and intent to change behavior using a Likert‐type scale (1=strongly disagree to 5=strongly agree). A constructed‐response section for comments regarding their experience was included. Descriptive statistics and Wilcoxon rank sum analyses were performed.

RESULTS

Forty‐four academic hospitalist faculty members (44/83; 53%) and 244 resident trainees (244/356; 68%) participated. When queried regarding their perception of the displayed behaviors in the videos, nearly 100% of faculty and trainees felt disparaging the emergency department or primary care physician for missed findings or clinical decisions was somewhat unprofessional or unprofessional. Ninety percent of hospitalists and 93% of trainees rated celebrating a blocked admission as somewhat unprofessional or unprofessional (Table 1).

The scenarios portrayed were well received, with more than 85% of faculty and trainees agreeing that the behaviors displayed were realistic. Those who perceived videos as very realistic were more likely to report intent to change behavior (93% vs 53%, P=0.01). Nearly two‐thirds of faculty and 67% of housestaff expressed agreement that they intended to change behavior based upon the experience (Table 2).

Qualitative comments in the constructed‐response portion of the evaluation noted the effectiveness of the interactive materials. In addition, the need for focused faculty development was identified by 1 respondent who stated: If unprofessional behavior is the unwritten curriculum, there needs to be an explicit, written curriculum to address it. Finally, the aim of facilitating self‐reflection is echoed in this faculty respondent's comment: Always good to be reminded of our behaviors and the influence they have on others and from this resident physician It helps to re‐evaluate how you talk to people.

CONCLUSIONS

Faculty can be a large determinant of the learning environment and impact trainees' professional development.[9] Hospitalists should be encouraged to embrace faculty role‐modeling of effective professional behaviors, especially given their increased presence in the inpatient learning environment. In addition, resident trainees and their behaviors contribute to the learning environment and influence the further professional development of more junior trainees.[10] Targeting professionalism education toward previously identified and prevalent unprofessional behaviors in the inpatient care of patients may serve to affect the most change among providers who practice in this setting. Individualized assessment of the learning environment may aid in identifying common scenarios that may plague a specific learning culture, allowing for relevant and targeted discussion of factors that promote and perpetuate such behaviors.[11]

Interactive, video‐based modules provided an effective way to promote interactive reflection and robust discussion. This model of experiential learning is an effective form of professional development as it engages the learner and stimulates ongoing incorporation of the topics addressed.[12, 13] Creating a shared concrete experience among targeted learners, using the video‐based scenarios, stimulates reflective observation, and ultimately experimentation, or incorporation into practice.[14]

There are several limitations to our evaluation including that we focused solely on academic hospitalist programs, and our sample size for faculty and residents was small. Also, we only addressed a small, though representative, sample of unprofessional behaviors and have not yet linked intervention to actual behavior change. Finally, the script scenarios that we used in this study were not previously published as they were created specifically for this intervention. Validity evidence for these scenarios include that they were based upon the results of earlier work from our institutions and underwent thorough peer review for content and clarity. Further studies will be required to do this. However, we do believe that these are positive findings for utilizing this type of interactive curriculum for professionalism education to promote self‐reflection and behavior change.

Video‐based professionalism education is a feasible, interactive mechanism to encourage self‐reflection and intent to change behavior among faculty and resident physicians. Future study is underway to conduct longitudinal assessments of the learning environments at the participating institutions to assess culture change, perceptions of behaviors, and sustainability of this type of intervention.

Disclosures: The authors acknowledge funding from the American Board of Internal Medicine. The funders had no role in the design of the study; the collection, analysis, and interpretation of the data; or the decision to approve publication of the finished manuscript. Results from this work have been presented at the Midwest Society of General Internal Medicine Regional Meeting, Chicago, Illinois, September 2011; Midwest Society of Hospital Medicine Regional Meeting, Chicago, Illinois, October 2011, and Society of Hospital Medicine Annual Meeting, San Diego, California, April 2012. The authors declare that they do not have any conflicts of interest to disclose.

Unprofessional behavior in the inpatient setting has the potential to impact care delivery and the quality of trainee's educational experience. These behaviors, from disparaging colleagues to blocking admissions, can negatively impact the learning environment. The learning environment or conditions created by the patient care team's actions play a critical role in the development of trainees.[1, 2] The rising presence of hospitalists in the inpatient setting raises the question of how their actions impact the learning environment. Professional behavior has been defined as a core competency for hospitalists by the Society of Hospital Medicine.[3] Professional behavior of all team members, from faculty to trainee, can impact the learning environment and patient safety.[4, 5] However, few educational materials exist to train faculty and housestaff on recognizing and ameliorating unprofessional behaviors.

A prior assessment regarding hospitalists' lapses in professionalism identified scenarios that demonstrated increased participation by hospitalists at 3 institutions.[6] Participants reported observation or participation in specific unprofessional behaviors and rated their perception of these behaviors. Additional work within those residency environments demonstrated that residents' perceptions of and participation in these behaviors increased throughout training, with environmental characteristics, specifically faculty behavior, influencing trainee professional development and acclimation of these behaviors.[7, 8]

Although overall participation in egregious behavior was low, resident participation in 3 categories of unprofessional behavior increased during internship. Those scenarios included disparaging the emergency room or primary care physician for missed findings or management decisions, blocking or not taking admissions appropriate for the service in question, and misrepresenting a test as urgent to expedite obtaining the test. We developed our intervention focused on these areas to address professionalism lapses that occur during internship. Our earlier work showed faculty role models influenced trainee behavior. For this reason, we provided education to both residents and hospitalists to maximize the impact of the intervention.

We present here a novel, interactive, video‐based workshop curriculum for faculty and trainees that aims to illustrate unprofessional behaviors and outlines the role faculty may play in promoting such behaviors. In addition, we review the result of postworkshop evaluation on intent to change behavior and satisfaction.

METHODS

A grant from the American Board of Internal Medicine Foundation supported this project. The working group that resulted, the Chicago Professional Practice Project and Outcomes, included faculty representation from 3 Chicago‐area hospitals: the University of Chicago, Northwestern University, and NorthShore University HealthSystem. Academic hospitalists at these sites were invited to participate. Each site also has an internal medicine residency program in which hospitalists were expected to attend the teaching service. Given this, resident trainees at all participating sites, and 1 community teaching affiliate program (Mercy Hospital and Medical Center) where academic hospitalists at the University of Chicago rotate, were recruited for participation. Faculty champions were identified for each site, and 1 internal and external faculty representative from the working group served to debrief and facilitate. Trainee workshops were administered by 1 internal and external collaborator, and for the community site, 2 external faculty members. Workshops were held during established educational conference times, and lunch was provided.

Scripts highlighting each of the behaviors identified in the prior survey were developed and peer reviewed for clarity and face validity across the 3 sites. Medical student and resident actors were trained utilizing the finalized scripts, and a performance artist affiliated with the Screen Actors Guild assisted in their preparation for filming. All videos were filmed at the University of Chicago Pritzker School of Medicine Clinical Performance Center. The final videos ranged in length from 4 to 7 minutes and included title, cast, and funding source. As an example, 1 video highlighted the unprofessional behavior of misrepresenting a test as urgent to prioritize one's patient in the queue. This video included a resident, intern, and attending on inpatient rounds during which the resident encouraged the intern to misrepresent the patient's status to expedite obtaining the study and facilitate the patient's discharge. The resident stressed that he would be in the clinic and had many patients to see, highlighting the impact of workload on unprofessional behavior, and aggressively persuaded the intern to sell her test to have it performed the same day. When this occurred, the attending applauded the intern for her strong work.

A moderator guide and debriefing tools were developed to facilitate discussion. The duration of each of the workshops was approximately 60 minutes. After welcoming remarks, participants were provided tools to utilize during the viewing of each video. These checklists noted the roles of those depicted in the video, asked to identify positive or negative behaviors displayed, and included questions regarding how behaviors could be detrimental and how the situation could have been prevented. After viewing the videos, participants divided into small groups to discuss the individual exhibiting the unprofessional behavior, their perceived motivation for said behavior, and its impact on the team culture and patient care. Following a small‐group discussion, large‐group debriefing was performed, addressing the barriers and facilitators to professional behavior. Two videos were shown at each workshop, and participants completed a postworkshop evaluation. Videos chosen for viewing were based upon preworkshop survey results that highlighted areas of concern at that specific site.

Postworkshop paper‐based evaluations assessed participants' perception of displayed behaviors on a Likert‐type scale (1=unprofessional to 5=professional) utilizing items validated in prior work,[6, 7, 8] their level of agreement regarding the impact of video‐based exercises, and intent to change behavior using a Likert‐type scale (1=strongly disagree to 5=strongly agree). A constructed‐response section for comments regarding their experience was included. Descriptive statistics and Wilcoxon rank sum analyses were performed.

RESULTS

Forty‐four academic hospitalist faculty members (44/83; 53%) and 244 resident trainees (244/356; 68%) participated. When queried regarding their perception of the displayed behaviors in the videos, nearly 100% of faculty and trainees felt disparaging the emergency department or primary care physician for missed findings or clinical decisions was somewhat unprofessional or unprofessional. Ninety percent of hospitalists and 93% of trainees rated celebrating a blocked admission as somewhat unprofessional or unprofessional (Table 1).

The scenarios portrayed were well received, with more than 85% of faculty and trainees agreeing that the behaviors displayed were realistic. Those who perceived videos as very realistic were more likely to report intent to change behavior (93% vs 53%, P=0.01). Nearly two‐thirds of faculty and 67% of housestaff expressed agreement that they intended to change behavior based upon the experience (Table 2).

Qualitative comments in the constructed‐response portion of the evaluation noted the effectiveness of the interactive materials. In addition, the need for focused faculty development was identified by 1 respondent who stated: If unprofessional behavior is the unwritten curriculum, there needs to be an explicit, written curriculum to address it. Finally, the aim of facilitating self‐reflection is echoed in this faculty respondent's comment: Always good to be reminded of our behaviors and the influence they have on others and from this resident physician It helps to re‐evaluate how you talk to people.

CONCLUSIONS

Faculty can be a large determinant of the learning environment and impact trainees' professional development.[9] Hospitalists should be encouraged to embrace faculty role‐modeling of effective professional behaviors, especially given their increased presence in the inpatient learning environment. In addition, resident trainees and their behaviors contribute to the learning environment and influence the further professional development of more junior trainees.[10] Targeting professionalism education toward previously identified and prevalent unprofessional behaviors in the inpatient care of patients may serve to affect the most change among providers who practice in this setting. Individualized assessment of the learning environment may aid in identifying common scenarios that may plague a specific learning culture, allowing for relevant and targeted discussion of factors that promote and perpetuate such behaviors.[11]

Interactive, video‐based modules provided an effective way to promote interactive reflection and robust discussion. This model of experiential learning is an effective form of professional development as it engages the learner and stimulates ongoing incorporation of the topics addressed.[12, 13] Creating a shared concrete experience among targeted learners, using the video‐based scenarios, stimulates reflective observation, and ultimately experimentation, or incorporation into practice.[14]

There are several limitations to our evaluation including that we focused solely on academic hospitalist programs, and our sample size for faculty and residents was small. Also, we only addressed a small, though representative, sample of unprofessional behaviors and have not yet linked intervention to actual behavior change. Finally, the script scenarios that we used in this study were not previously published as they were created specifically for this intervention. Validity evidence for these scenarios include that they were based upon the results of earlier work from our institutions and underwent thorough peer review for content and clarity. Further studies will be required to do this. However, we do believe that these are positive findings for utilizing this type of interactive curriculum for professionalism education to promote self‐reflection and behavior change.

Video‐based professionalism education is a feasible, interactive mechanism to encourage self‐reflection and intent to change behavior among faculty and resident physicians. Future study is underway to conduct longitudinal assessments of the learning environments at the participating institutions to assess culture change, perceptions of behaviors, and sustainability of this type of intervention.

Disclosures: The authors acknowledge funding from the American Board of Internal Medicine. The funders had no role in the design of the study; the collection, analysis, and interpretation of the data; or the decision to approve publication of the finished manuscript. Results from this work have been presented at the Midwest Society of General Internal Medicine Regional Meeting, Chicago, Illinois, September 2011; Midwest Society of Hospital Medicine Regional Meeting, Chicago, Illinois, October 2011, and Society of Hospital Medicine Annual Meeting, San Diego, California, April 2012. The authors declare that they do not have any conflicts of interest to disclose.

Unprofessional behavior in the inpatient setting has the potential to impact care delivery and the quality of trainee's educational experience. These behaviors, from disparaging colleagues to blocking admissions, can negatively impact the learning environment. The learning environment or conditions created by the patient care team's actions play a critical role in the development of trainees.[1, 2] The rising presence of hospitalists in the inpatient setting raises the question of how their actions impact the learning environment. Professional behavior has been defined as a core competency for hospitalists by the Society of Hospital Medicine.[3] Professional behavior of all team members, from faculty to trainee, can impact the learning environment and patient safety.[4, 5] However, few educational materials exist to train faculty and housestaff on recognizing and ameliorating unprofessional behaviors.

A prior assessment regarding hospitalists' lapses in professionalism identified scenarios that demonstrated increased participation by hospitalists at 3 institutions.[6] Participants reported observation or participation in specific unprofessional behaviors and rated their perception of these behaviors. Additional work within those residency environments demonstrated that residents' perceptions of and participation in these behaviors increased throughout training, with environmental characteristics, specifically faculty behavior, influencing trainee professional development and acclimation of these behaviors.[7, 8]

Although overall participation in egregious behavior was low, resident participation in 3 categories of unprofessional behavior increased during internship. Those scenarios included disparaging the emergency room or primary care physician for missed findings or management decisions, blocking or not taking admissions appropriate for the service in question, and misrepresenting a test as urgent to expedite obtaining the test. We developed our intervention focused on these areas to address professionalism lapses that occur during internship. Our earlier work showed faculty role models influenced trainee behavior. For this reason, we provided education to both residents and hospitalists to maximize the impact of the intervention.

We present here a novel, interactive, video‐based workshop curriculum for faculty and trainees that aims to illustrate unprofessional behaviors and outlines the role faculty may play in promoting such behaviors. In addition, we review the result of postworkshop evaluation on intent to change behavior and satisfaction.

METHODS

A grant from the American Board of Internal Medicine Foundation supported this project. The working group that resulted, the Chicago Professional Practice Project and Outcomes, included faculty representation from 3 Chicago‐area hospitals: the University of Chicago, Northwestern University, and NorthShore University HealthSystem. Academic hospitalists at these sites were invited to participate. Each site also has an internal medicine residency program in which hospitalists were expected to attend the teaching service. Given this, resident trainees at all participating sites, and 1 community teaching affiliate program (Mercy Hospital and Medical Center) where academic hospitalists at the University of Chicago rotate, were recruited for participation. Faculty champions were identified for each site, and 1 internal and external faculty representative from the working group served to debrief and facilitate. Trainee workshops were administered by 1 internal and external collaborator, and for the community site, 2 external faculty members. Workshops were held during established educational conference times, and lunch was provided.

Scripts highlighting each of the behaviors identified in the prior survey were developed and peer reviewed for clarity and face validity across the 3 sites. Medical student and resident actors were trained utilizing the finalized scripts, and a performance artist affiliated with the Screen Actors Guild assisted in their preparation for filming. All videos were filmed at the University of Chicago Pritzker School of Medicine Clinical Performance Center. The final videos ranged in length from 4 to 7 minutes and included title, cast, and funding source. As an example, 1 video highlighted the unprofessional behavior of misrepresenting a test as urgent to prioritize one's patient in the queue. This video included a resident, intern, and attending on inpatient rounds during which the resident encouraged the intern to misrepresent the patient's status to expedite obtaining the study and facilitate the patient's discharge. The resident stressed that he would be in the clinic and had many patients to see, highlighting the impact of workload on unprofessional behavior, and aggressively persuaded the intern to sell her test to have it performed the same day. When this occurred, the attending applauded the intern for her strong work.

A moderator guide and debriefing tools were developed to facilitate discussion. The duration of each of the workshops was approximately 60 minutes. After welcoming remarks, participants were provided tools to utilize during the viewing of each video. These checklists noted the roles of those depicted in the video, asked to identify positive or negative behaviors displayed, and included questions regarding how behaviors could be detrimental and how the situation could have been prevented. After viewing the videos, participants divided into small groups to discuss the individual exhibiting the unprofessional behavior, their perceived motivation for said behavior, and its impact on the team culture and patient care. Following a small‐group discussion, large‐group debriefing was performed, addressing the barriers and facilitators to professional behavior. Two videos were shown at each workshop, and participants completed a postworkshop evaluation. Videos chosen for viewing were based upon preworkshop survey results that highlighted areas of concern at that specific site.

Postworkshop paper‐based evaluations assessed participants' perception of displayed behaviors on a Likert‐type scale (1=unprofessional to 5=professional) utilizing items validated in prior work,[6, 7, 8] their level of agreement regarding the impact of video‐based exercises, and intent to change behavior using a Likert‐type scale (1=strongly disagree to 5=strongly agree). A constructed‐response section for comments regarding their experience was included. Descriptive statistics and Wilcoxon rank sum analyses were performed.

RESULTS

Forty‐four academic hospitalist faculty members (44/83; 53%) and 244 resident trainees (244/356; 68%) participated. When queried regarding their perception of the displayed behaviors in the videos, nearly 100% of faculty and trainees felt disparaging the emergency department or primary care physician for missed findings or clinical decisions was somewhat unprofessional or unprofessional. Ninety percent of hospitalists and 93% of trainees rated celebrating a blocked admission as somewhat unprofessional or unprofessional (Table 1).

The scenarios portrayed were well received, with more than 85% of faculty and trainees agreeing that the behaviors displayed were realistic. Those who perceived videos as very realistic were more likely to report intent to change behavior (93% vs 53%, P=0.01). Nearly two‐thirds of faculty and 67% of housestaff expressed agreement that they intended to change behavior based upon the experience (Table 2).

Qualitative comments in the constructed‐response portion of the evaluation noted the effectiveness of the interactive materials. In addition, the need for focused faculty development was identified by 1 respondent who stated: If unprofessional behavior is the unwritten curriculum, there needs to be an explicit, written curriculum to address it. Finally, the aim of facilitating self‐reflection is echoed in this faculty respondent's comment: Always good to be reminded of our behaviors and the influence they have on others and from this resident physician It helps to re‐evaluate how you talk to people.

CONCLUSIONS

Faculty can be a large determinant of the learning environment and impact trainees' professional development.[9] Hospitalists should be encouraged to embrace faculty role‐modeling of effective professional behaviors, especially given their increased presence in the inpatient learning environment. In addition, resident trainees and their behaviors contribute to the learning environment and influence the further professional development of more junior trainees.[10] Targeting professionalism education toward previously identified and prevalent unprofessional behaviors in the inpatient care of patients may serve to affect the most change among providers who practice in this setting. Individualized assessment of the learning environment may aid in identifying common scenarios that may plague a specific learning culture, allowing for relevant and targeted discussion of factors that promote and perpetuate such behaviors.[11]

Interactive, video‐based modules provided an effective way to promote interactive reflection and robust discussion. This model of experiential learning is an effective form of professional development as it engages the learner and stimulates ongoing incorporation of the topics addressed.[12, 13] Creating a shared concrete experience among targeted learners, using the video‐based scenarios, stimulates reflective observation, and ultimately experimentation, or incorporation into practice.[14]

There are several limitations to our evaluation including that we focused solely on academic hospitalist programs, and our sample size for faculty and residents was small. Also, we only addressed a small, though representative, sample of unprofessional behaviors and have not yet linked intervention to actual behavior change. Finally, the script scenarios that we used in this study were not previously published as they were created specifically for this intervention. Validity evidence for these scenarios include that they were based upon the results of earlier work from our institutions and underwent thorough peer review for content and clarity. Further studies will be required to do this. However, we do believe that these are positive findings for utilizing this type of interactive curriculum for professionalism education to promote self‐reflection and behavior change.

Video‐based professionalism education is a feasible, interactive mechanism to encourage self‐reflection and intent to change behavior among faculty and resident physicians. Future study is underway to conduct longitudinal assessments of the learning environments at the participating institutions to assess culture change, perceptions of behaviors, and sustainability of this type of intervention.

Disclosures: The authors acknowledge funding from the American Board of Internal Medicine. The funders had no role in the design of the study; the collection, analysis, and interpretation of the data; or the decision to approve publication of the finished manuscript. Results from this work have been presented at the Midwest Society of General Internal Medicine Regional Meeting, Chicago, Illinois, September 2011; Midwest Society of Hospital Medicine Regional Meeting, Chicago, Illinois, October 2011, and Society of Hospital Medicine Annual Meeting, San Diego, California, April 2012. The authors declare that they do not have any conflicts of interest to disclose.

Children with complex chronic conditions comprise an increasing proportion of hospital admissions, readmissions, and resource use.[1, 2, 3] Dependence on technology or medical devices is a frequent characteristic of children in this group.[4] Adverse medical device events (AMDEs) are estimated to occur in as many as 8% of all adult admissions, depending on the methods used to identify them.[5] These events may result in hospitalizations or complicate hospital stays. To date, however, the burden of AMDEs among hospitalized children is little described, even though children may be at increased risk for device events as compared to adults.[6] Although some medical devices are intended solely or primarily for use with children, most devices used with children have been initially developed for, tested with, and most frequently employed to treat adults.[6] Assessing the continued safety and effectiveness of medical devices marketed in the Unites States is the responsibility of the Center for Devices and Radiologic Health of the US Food and Drug Administration (FDA). Its existing mechanisms for postmarket device surveillance rely primarily on passive reporting systems and specific observational studies.[7]

The objective of this study was to utilize administrative data from children's hospitals to explore the prevalence and nature of AMDEs in tertiary care children's hospitals that treat significant numbers of children with complex needs requiring medical devices.

METHODS

Data were obtained from the Pediatric Health Information System (PHIS), an administrative database containing inpatient data from 44 not‐for‐profit, tertiary care, pediatric hospitals affiliated with the Children's Hospital Association. Data are deidentified at the time of submission, and are subjected to a number of reliability and validity checks.[8] Individual admission records have both a deidentified visit identification (ID) and patient ID, allowing for linkage of multiple admissions by the same patient.

AMDEs were defined by International Classification of Diseases, Ninth Revision (ICD‐9) codes, using a methodology developed by Samore et al., who identified a set of such codes that specified devices in their definitions and therefore were considered to have a high likelihood of indicating a device problem (see Supporting Information, Table S1, in the online version of this article).[5] The diagnosis codes were grouped into device categories (eg, nervous system, orthopedic, cardiac).

From the 44 hospitals, the primary study cohort consisted of any patient with an admission between January 1, 2004 and December 31, 2011 with 1 AMDE ICD‐9 code as a primary or secondary diagnosis.

Descriptive statistics for patient demographics and visit characteristics of AMDE admissions were generated and stratified by device category. We reported these as counts and percentages for categorical variables and as median and interquartile range for length of stay. We also reported on how frequently patients with AMDEs have a top 10 most common diagnosis and top 10 most common procedure during the AMDE admission. We also reported the presence or absence of a complex chronic condition.[9] We generated the list of most common principal diagnoses and procedures by a separate query of PHIS from 2004 to 2009. Our top 10 most common diagnoses included ICD‐9 codes 486 (pneumonia), 466.11 (acute bronchiolitis due to respiratory syncytial virus), V58.11 (chemotherapy encounter), 493.92 (asthma exacerbation), 493.91 (asthma with status asthmaticus), 466.19 (acute bronchiolitis due to other organism), 780.39 (other convulsions), 540.9 (acute appendicitis), 282.62 (sickle cell disease with crisis), and 276.51 (dehydration). Our top 10 most common procedures included ICD‐9 codes 38.93 (venous catheterization), 03.31 (spinal tap), 99.04 (packed blood cell transfusion), 99.15 (parenteral nutrition), 99.25 (cancer chemotherapy), 96.71 (invasive mechanical ventilation, 96 hours), 96.04 (endotracheal intubation), 96.72 (invasive mechanical ventilation,95 hours), 96.6 (enteral nutrition), and 99.05 (platelet transfusion). Analyses were performed using SAS Enterprise Guide version 4.2 for Windows (SAS Institute, Cary, NC).

This study was approved by Cincinnati Children's Hospital Medical Center Institutional Review Board.

RESULTS

Of the 4,115,755 admissions during the study period, 136,465 (3.3%) had at least 1 AMDE. Over our study period, AMDEs were associated with a mean 17,058 inpatient stays annually. The number of AMDE‐related admissions decreased the last 4 years of our study period despite generally increasing admissions at PHIS hospitals (Figure 1). For 55% of the admissions (75,206/136,465), this AMDE code represented the primary diagnosis. Of these visits with a primary AMDE diagnosis, 39,874 (53%) were related to nervous system devices. The visits associated with AMDEs were comprised of 88,908 unique patients, 55% of whom were male (Table 1). The median age on admission was 6 years, and the interquartile range was 1 to 14 years of age.

Figure 1

Among admissions with AMDEs, 2.9% ended in death. The mortality was 0.5% when an AMDE was the primary diagnosis and 5.7% when the AMDE was a secondary diagnosis. The median length of inpatient stays was 6 days, with an interquartile range of 2 to 17 days.

Vascular access AMDEs were the most common event associated with admissions (26.6%), followed by nervous system devices (17.8%) (Table 2). The majority (75.5%) of patients admitted with AMDEs had a complex chronic condition. Less than half (46.8%) of AMDE admissions had an associated code for 1 of the 10 most common principal procedures. A minority (14.3%) of admissions had an associated ICD‐9 code for 1 of the top 10 most common principal diagnoses.

DISCUSSION

To our knowledge, our study is the first to report the burden of AMDEs among children requiring hospitalization. AMDEs are common in this population of children cared for at tertiary care children's hospitals, accounting for or complicating 3.3% of inpatient stays in these 44 hospitals. AMDEs were associated with a mean of >17,000 total visits per year. Vascular access devices and nervous system devices were the most common device categories linked to AMDEs. Similar to published literature, we found that the youngest children accounted for the highest proportion of AMDEs.[10, 11]

The majority (>75%) of children with an AMDE admission had diagnoses indicating complex chronic conditions during the admission. Over a partially overlapping study period, Feudtner and colleagues found 25.2% of patients admitted to PHIS hospitals had complex chronic conditions.[12] This finding, combined with the uncommon association of the most prevalent diagnoses and procedures, suggests that the burden of AMDEs falls disproportionately on this population of children. Death occurred considerably less commonly when AMDE diagnosis was the primary versus a secondary diagnosis (0.5% vs 5.7%). This finding likely illustrates 2 distinct populations: children with an AMDE that causes admission who have a relatively low risk of mortality and a second group who have AMDE‐complicated hospitalizations that may have an already high risk of mortality.

Our findings complement those of Wang and colleagues who employed the National Electronic Injury Surveillance System All Injury Program database to provide national estimates of medical device‐associated adverse events.[11] Importantly, this group used a different population (patients presenting to the emergency department) and a different methodology. These authors reported on device‐associated events, as they did not collect information to discriminate the device's role in the event. A walker that malfunctioned leading to patient injury would be a device‐related event; however, a patient who has a walker suffering a fall would be device‐associated, even if the walker's role in the injury was uncertain. We believe our methodology, established by Samore et al., more accurately identifies device‐related events.[5] Wang et al. found that 6.3% of pediatric patients who presented to emergency departments with medical device‐associated events were admitted to the hospital.[11] This resulted in national estimates of 9,082 events with 95% confidence intervals of 2,990 to 25,373 hospitalizations. Our findings of >17,000 AMDE‐related inpatient stays per annum included not only AMDEs leading to admissions but also those that were complications during stays.

Our study has several limitations, most related to the possibility of misclassification present in administrative data. Our approach only captured events that led to or complicated admissions. We suspect that ICD‐9 codes likely missed some AMDEs and that our estimates may therefore under‐represent this problem in our population. Future studies should compare our methodology, which has produced the first across‐center estimates of AMDE admissions, to alternative event capture techniques. We were unable to determine which events were present on admission and which complicated hospital stays, and it is likely that differing interventions would be required to reduce these 2 types of AMDEs. Another important limitation is that the PHIS database, comprised of data on children receiving care at tertiary academic medical centers with large numbers of pediatric subspecialists, is not representative of the population of children overall. The individual ICD‐9 codes for AMDEs are sufficiently nonspecific to limit the ability to characterize device events from administrative data alone. The high prevalence of unspecified device‐related admissions is an additional limitation. Although the estimates of these types of AMDEs are important in describing the frequency of these events, the unspecified category limits the ability to fully stratify based on device type and then implement monitoring strategies and interventions based on each.

To our knowledge, this study is the first multicenter analysis of the spectrum of pediatric AMDEs in hospitalized children. The AMDE prevalence is substantial, and the burden of these events largely falls on children with complex chronic conditions. Despite its limitations, this study complements recent efforts to enhance postmarket surveillance of pediatric devices including that of the FDA's Office of Pediatric Therapeutics, the recent FDA report Strengthening Our National System for Medical Device Postmarket Surveillance (http://www.fda.gov/MedicalDevices/default.htm), and the proposed rule for a unique device identification (UDI) system.[13] Establishment of UDI systems and their eventual incorporation into electronic health‐related databases will greatly expand postmarket surveillance capabilities.[13]

Our description of AMDEs by device category and patient characteristics is a first and necessary step to understanding the public health burden associated with device use in the pediatric population. Further developments in refined coding and device designation (eg, UDI systems) are needed to refine these estimates.

Children with complex chronic conditions comprise an increasing proportion of hospital admissions, readmissions, and resource use.[1, 2, 3] Dependence on technology or medical devices is a frequent characteristic of children in this group.[4] Adverse medical device events (AMDEs) are estimated to occur in as many as 8% of all adult admissions, depending on the methods used to identify them.[5] These events may result in hospitalizations or complicate hospital stays. To date, however, the burden of AMDEs among hospitalized children is little described, even though children may be at increased risk for device events as compared to adults.[6] Although some medical devices are intended solely or primarily for use with children, most devices used with children have been initially developed for, tested with, and most frequently employed to treat adults.[6] Assessing the continued safety and effectiveness of medical devices marketed in the Unites States is the responsibility of the Center for Devices and Radiologic Health of the US Food and Drug Administration (FDA). Its existing mechanisms for postmarket device surveillance rely primarily on passive reporting systems and specific observational studies.[7]

The objective of this study was to utilize administrative data from children's hospitals to explore the prevalence and nature of AMDEs in tertiary care children's hospitals that treat significant numbers of children with complex needs requiring medical devices.

METHODS

Data were obtained from the Pediatric Health Information System (PHIS), an administrative database containing inpatient data from 44 not‐for‐profit, tertiary care, pediatric hospitals affiliated with the Children's Hospital Association. Data are deidentified at the time of submission, and are subjected to a number of reliability and validity checks.[8] Individual admission records have both a deidentified visit identification (ID) and patient ID, allowing for linkage of multiple admissions by the same patient.

AMDEs were defined by International Classification of Diseases, Ninth Revision (ICD‐9) codes, using a methodology developed by Samore et al., who identified a set of such codes that specified devices in their definitions and therefore were considered to have a high likelihood of indicating a device problem (see Supporting Information, Table S1, in the online version of this article).[5] The diagnosis codes were grouped into device categories (eg, nervous system, orthopedic, cardiac).

From the 44 hospitals, the primary study cohort consisted of any patient with an admission between January 1, 2004 and December 31, 2011 with 1 AMDE ICD‐9 code as a primary or secondary diagnosis.

Descriptive statistics for patient demographics and visit characteristics of AMDE admissions were generated and stratified by device category. We reported these as counts and percentages for categorical variables and as median and interquartile range for length of stay. We also reported on how frequently patients with AMDEs have a top 10 most common diagnosis and top 10 most common procedure during the AMDE admission. We also reported the presence or absence of a complex chronic condition.[9] We generated the list of most common principal diagnoses and procedures by a separate query of PHIS from 2004 to 2009. Our top 10 most common diagnoses included ICD‐9 codes 486 (pneumonia), 466.11 (acute bronchiolitis due to respiratory syncytial virus), V58.11 (chemotherapy encounter), 493.92 (asthma exacerbation), 493.91 (asthma with status asthmaticus), 466.19 (acute bronchiolitis due to other organism), 780.39 (other convulsions), 540.9 (acute appendicitis), 282.62 (sickle cell disease with crisis), and 276.51 (dehydration). Our top 10 most common procedures included ICD‐9 codes 38.93 (venous catheterization), 03.31 (spinal tap), 99.04 (packed blood cell transfusion), 99.15 (parenteral nutrition), 99.25 (cancer chemotherapy), 96.71 (invasive mechanical ventilation, 96 hours), 96.04 (endotracheal intubation), 96.72 (invasive mechanical ventilation,95 hours), 96.6 (enteral nutrition), and 99.05 (platelet transfusion). Analyses were performed using SAS Enterprise Guide version 4.2 for Windows (SAS Institute, Cary, NC).

This study was approved by Cincinnati Children's Hospital Medical Center Institutional Review Board.

RESULTS

Of the 4,115,755 admissions during the study period, 136,465 (3.3%) had at least 1 AMDE. Over our study period, AMDEs were associated with a mean 17,058 inpatient stays annually. The number of AMDE‐related admissions decreased the last 4 years of our study period despite generally increasing admissions at PHIS hospitals (Figure 1). For 55% of the admissions (75,206/136,465), this AMDE code represented the primary diagnosis. Of these visits with a primary AMDE diagnosis, 39,874 (53%) were related to nervous system devices. The visits associated with AMDEs were comprised of 88,908 unique patients, 55% of whom were male (Table 1). The median age on admission was 6 years, and the interquartile range was 1 to 14 years of age.

Figure 1

Among admissions with AMDEs, 2.9% ended in death. The mortality was 0.5% when an AMDE was the primary diagnosis and 5.7% when the AMDE was a secondary diagnosis. The median length of inpatient stays was 6 days, with an interquartile range of 2 to 17 days.

Vascular access AMDEs were the most common event associated with admissions (26.6%), followed by nervous system devices (17.8%) (Table 2). The majority (75.5%) of patients admitted with AMDEs had a complex chronic condition. Less than half (46.8%) of AMDE admissions had an associated code for 1 of the 10 most common principal procedures. A minority (14.3%) of admissions had an associated ICD‐9 code for 1 of the top 10 most common principal diagnoses.

DISCUSSION

To our knowledge, our study is the first to report the burden of AMDEs among children requiring hospitalization. AMDEs are common in this population of children cared for at tertiary care children's hospitals, accounting for or complicating 3.3% of inpatient stays in these 44 hospitals. AMDEs were associated with a mean of >17,000 total visits per year. Vascular access devices and nervous system devices were the most common device categories linked to AMDEs. Similar to published literature, we found that the youngest children accounted for the highest proportion of AMDEs.[10, 11]

The majority (>75%) of children with an AMDE admission had diagnoses indicating complex chronic conditions during the admission. Over a partially overlapping study period, Feudtner and colleagues found 25.2% of patients admitted to PHIS hospitals had complex chronic conditions.[12] This finding, combined with the uncommon association of the most prevalent diagnoses and procedures, suggests that the burden of AMDEs falls disproportionately on this population of children. Death occurred considerably less commonly when AMDE diagnosis was the primary versus a secondary diagnosis (0.5% vs 5.7%). This finding likely illustrates 2 distinct populations: children with an AMDE that causes admission who have a relatively low risk of mortality and a second group who have AMDE‐complicated hospitalizations that may have an already high risk of mortality.

Our findings complement those of Wang and colleagues who employed the National Electronic Injury Surveillance System All Injury Program database to provide national estimates of medical device‐associated adverse events.[11] Importantly, this group used a different population (patients presenting to the emergency department) and a different methodology. These authors reported on device‐associated events, as they did not collect information to discriminate the device's role in the event. A walker that malfunctioned leading to patient injury would be a device‐related event; however, a patient who has a walker suffering a fall would be device‐associated, even if the walker's role in the injury was uncertain. We believe our methodology, established by Samore et al., more accurately identifies device‐related events.[5] Wang et al. found that 6.3% of pediatric patients who presented to emergency departments with medical device‐associated events were admitted to the hospital.[11] This resulted in national estimates of 9,082 events with 95% confidence intervals of 2,990 to 25,373 hospitalizations. Our findings of >17,000 AMDE‐related inpatient stays per annum included not only AMDEs leading to admissions but also those that were complications during stays.

Our study has several limitations, most related to the possibility of misclassification present in administrative data. Our approach only captured events that led to or complicated admissions. We suspect that ICD‐9 codes likely missed some AMDEs and that our estimates may therefore under‐represent this problem in our population. Future studies should compare our methodology, which has produced the first across‐center estimates of AMDE admissions, to alternative event capture techniques. We were unable to determine which events were present on admission and which complicated hospital stays, and it is likely that differing interventions would be required to reduce these 2 types of AMDEs. Another important limitation is that the PHIS database, comprised of data on children receiving care at tertiary academic medical centers with large numbers of pediatric subspecialists, is not representative of the population of children overall. The individual ICD‐9 codes for AMDEs are sufficiently nonspecific to limit the ability to characterize device events from administrative data alone. The high prevalence of unspecified device‐related admissions is an additional limitation. Although the estimates of these types of AMDEs are important in describing the frequency of these events, the unspecified category limits the ability to fully stratify based on device type and then implement monitoring strategies and interventions based on each.

To our knowledge, this study is the first multicenter analysis of the spectrum of pediatric AMDEs in hospitalized children. The AMDE prevalence is substantial, and the burden of these events largely falls on children with complex chronic conditions. Despite its limitations, this study complements recent efforts to enhance postmarket surveillance of pediatric devices including that of the FDA's Office of Pediatric Therapeutics, the recent FDA report Strengthening Our National System for Medical Device Postmarket Surveillance (http://www.fda.gov/MedicalDevices/default.htm), and the proposed rule for a unique device identification (UDI) system.[13] Establishment of UDI systems and their eventual incorporation into electronic health‐related databases will greatly expand postmarket surveillance capabilities.[13]

Our description of AMDEs by device category and patient characteristics is a first and necessary step to understanding the public health burden associated with device use in the pediatric population. Further developments in refined coding and device designation (eg, UDI systems) are needed to refine these estimates.

Children with complex chronic conditions comprise an increasing proportion of hospital admissions, readmissions, and resource use.[1, 2, 3] Dependence on technology or medical devices is a frequent characteristic of children in this group.[4] Adverse medical device events (AMDEs) are estimated to occur in as many as 8% of all adult admissions, depending on the methods used to identify them.[5] These events may result in hospitalizations or complicate hospital stays. To date, however, the burden of AMDEs among hospitalized children is little described, even though children may be at increased risk for device events as compared to adults.[6] Although some medical devices are intended solely or primarily for use with children, most devices used with children have been initially developed for, tested with, and most frequently employed to treat adults.[6] Assessing the continued safety and effectiveness of medical devices marketed in the Unites States is the responsibility of the Center for Devices and Radiologic Health of the US Food and Drug Administration (FDA). Its existing mechanisms for postmarket device surveillance rely primarily on passive reporting systems and specific observational studies.[7]

The objective of this study was to utilize administrative data from children's hospitals to explore the prevalence and nature of AMDEs in tertiary care children's hospitals that treat significant numbers of children with complex needs requiring medical devices.

METHODS

Data were obtained from the Pediatric Health Information System (PHIS), an administrative database containing inpatient data from 44 not‐for‐profit, tertiary care, pediatric hospitals affiliated with the Children's Hospital Association. Data are deidentified at the time of submission, and are subjected to a number of reliability and validity checks.[8] Individual admission records have both a deidentified visit identification (ID) and patient ID, allowing for linkage of multiple admissions by the same patient.

AMDEs were defined by International Classification of Diseases, Ninth Revision (ICD‐9) codes, using a methodology developed by Samore et al., who identified a set of such codes that specified devices in their definitions and therefore were considered to have a high likelihood of indicating a device problem (see Supporting Information, Table S1, in the online version of this article).[5] The diagnosis codes were grouped into device categories (eg, nervous system, orthopedic, cardiac).

From the 44 hospitals, the primary study cohort consisted of any patient with an admission between January 1, 2004 and December 31, 2011 with 1 AMDE ICD‐9 code as a primary or secondary diagnosis.

Descriptive statistics for patient demographics and visit characteristics of AMDE admissions were generated and stratified by device category. We reported these as counts and percentages for categorical variables and as median and interquartile range for length of stay. We also reported on how frequently patients with AMDEs have a top 10 most common diagnosis and top 10 most common procedure during the AMDE admission. We also reported the presence or absence of a complex chronic condition.[9] We generated the list of most common principal diagnoses and procedures by a separate query of PHIS from 2004 to 2009. Our top 10 most common diagnoses included ICD‐9 codes 486 (pneumonia), 466.11 (acute bronchiolitis due to respiratory syncytial virus), V58.11 (chemotherapy encounter), 493.92 (asthma exacerbation), 493.91 (asthma with status asthmaticus), 466.19 (acute bronchiolitis due to other organism), 780.39 (other convulsions), 540.9 (acute appendicitis), 282.62 (sickle cell disease with crisis), and 276.51 (dehydration). Our top 10 most common procedures included ICD‐9 codes 38.93 (venous catheterization), 03.31 (spinal tap), 99.04 (packed blood cell transfusion), 99.15 (parenteral nutrition), 99.25 (cancer chemotherapy), 96.71 (invasive mechanical ventilation, 96 hours), 96.04 (endotracheal intubation), 96.72 (invasive mechanical ventilation,95 hours), 96.6 (enteral nutrition), and 99.05 (platelet transfusion). Analyses were performed using SAS Enterprise Guide version 4.2 for Windows (SAS Institute, Cary, NC).

This study was approved by Cincinnati Children's Hospital Medical Center Institutional Review Board.

RESULTS

Of the 4,115,755 admissions during the study period, 136,465 (3.3%) had at least 1 AMDE. Over our study period, AMDEs were associated with a mean 17,058 inpatient stays annually. The number of AMDE‐related admissions decreased the last 4 years of our study period despite generally increasing admissions at PHIS hospitals (Figure 1). For 55% of the admissions (75,206/136,465), this AMDE code represented the primary diagnosis. Of these visits with a primary AMDE diagnosis, 39,874 (53%) were related to nervous system devices. The visits associated with AMDEs were comprised of 88,908 unique patients, 55% of whom were male (Table 1). The median age on admission was 6 years, and the interquartile range was 1 to 14 years of age.

Figure 1

Among admissions with AMDEs, 2.9% ended in death. The mortality was 0.5% when an AMDE was the primary diagnosis and 5.7% when the AMDE was a secondary diagnosis. The median length of inpatient stays was 6 days, with an interquartile range of 2 to 17 days.

Vascular access AMDEs were the most common event associated with admissions (26.6%), followed by nervous system devices (17.8%) (Table 2). The majority (75.5%) of patients admitted with AMDEs had a complex chronic condition. Less than half (46.8%) of AMDE admissions had an associated code for 1 of the 10 most common principal procedures. A minority (14.3%) of admissions had an associated ICD‐9 code for 1 of the top 10 most common principal diagnoses.

DISCUSSION

To our knowledge, our study is the first to report the burden of AMDEs among children requiring hospitalization. AMDEs are common in this population of children cared for at tertiary care children's hospitals, accounting for or complicating 3.3% of inpatient stays in these 44 hospitals. AMDEs were associated with a mean of >17,000 total visits per year. Vascular access devices and nervous system devices were the most common device categories linked to AMDEs. Similar to published literature, we found that the youngest children accounted for the highest proportion of AMDEs.[10, 11]

The majority (>75%) of children with an AMDE admission had diagnoses indicating complex chronic conditions during the admission. Over a partially overlapping study period, Feudtner and colleagues found 25.2% of patients admitted to PHIS hospitals had complex chronic conditions.[12] This finding, combined with the uncommon association of the most prevalent diagnoses and procedures, suggests that the burden of AMDEs falls disproportionately on this population of children. Death occurred considerably less commonly when AMDE diagnosis was the primary versus a secondary diagnosis (0.5% vs 5.7%). This finding likely illustrates 2 distinct populations: children with an AMDE that causes admission who have a relatively low risk of mortality and a second group who have AMDE‐complicated hospitalizations that may have an already high risk of mortality.

Our findings complement those of Wang and colleagues who employed the National Electronic Injury Surveillance System All Injury Program database to provide national estimates of medical device‐associated adverse events.[11] Importantly, this group used a different population (patients presenting to the emergency department) and a different methodology. These authors reported on device‐associated events, as they did not collect information to discriminate the device's role in the event. A walker that malfunctioned leading to patient injury would be a device‐related event; however, a patient who has a walker suffering a fall would be device‐associated, even if the walker's role in the injury was uncertain. We believe our methodology, established by Samore et al., more accurately identifies device‐related events.[5] Wang et al. found that 6.3% of pediatric patients who presented to emergency departments with medical device‐associated events were admitted to the hospital.[11] This resulted in national estimates of 9,082 events with 95% confidence intervals of 2,990 to 25,373 hospitalizations. Our findings of >17,000 AMDE‐related inpatient stays per annum included not only AMDEs leading to admissions but also those that were complications during stays.

Our study has several limitations, most related to the possibility of misclassification present in administrative data. Our approach only captured events that led to or complicated admissions. We suspect that ICD‐9 codes likely missed some AMDEs and that our estimates may therefore under‐represent this problem in our population. Future studies should compare our methodology, which has produced the first across‐center estimates of AMDE admissions, to alternative event capture techniques. We were unable to determine which events were present on admission and which complicated hospital stays, and it is likely that differing interventions would be required to reduce these 2 types of AMDEs. Another important limitation is that the PHIS database, comprised of data on children receiving care at tertiary academic medical centers with large numbers of pediatric subspecialists, is not representative of the population of children overall. The individual ICD‐9 codes for AMDEs are sufficiently nonspecific to limit the ability to characterize device events from administrative data alone. The high prevalence of unspecified device‐related admissions is an additional limitation. Although the estimates of these types of AMDEs are important in describing the frequency of these events, the unspecified category limits the ability to fully stratify based on device type and then implement monitoring strategies and interventions based on each.

To our knowledge, this study is the first multicenter analysis of the spectrum of pediatric AMDEs in hospitalized children. The AMDE prevalence is substantial, and the burden of these events largely falls on children with complex chronic conditions. Despite its limitations, this study complements recent efforts to enhance postmarket surveillance of pediatric devices including that of the FDA's Office of Pediatric Therapeutics, the recent FDA report Strengthening Our National System for Medical Device Postmarket Surveillance (http://www.fda.gov/MedicalDevices/default.htm), and the proposed rule for a unique device identification (UDI) system.[13] Establishment of UDI systems and their eventual incorporation into electronic health‐related databases will greatly expand postmarket surveillance capabilities.[13]

Our description of AMDEs by device category and patient characteristics is a first and necessary step to understanding the public health burden associated with device use in the pediatric population. Further developments in refined coding and device designation (eg, UDI systems) are needed to refine these estimates.

Self‐reported penicillin allergy is common and frequently limits the available antimicrobial agents to choose from. This often results in the use of more expensive, potentially more toxic, and possibly less efficacious agents.[1, 2]

For over 30 years, penicilloyl‐polylysine (PPL) penicillin skin testing (PST) was widely used to diagnose penicillin allergy with a negative predictive value (NPV) of about 97% to 99%.[3] After being off the market for 5 years, PPL PST was reapproved in 2009 as PRE‐PEN.[4] However, many clinicians still fail to utilize PST despite its simplicity and substantial clinical impact. The main purpose of this study was to describe the predictive value of PST and impact on antibiotic selection in a sample of hospitalized patients with a reported history of penicillin allergy.

METHODS

In 2010, PST was introduced as a quality‐improvement measure after approval and support from the chief of professional services and the medical staff executive committee at Vidant Medical Center, an 861‐bed tertiary care and teaching hospital. Our antimicrobial stewardship program is regularly contacted for approval of alternative therapies in penicillin allergic patients. The PST quality‐improvement intervention was implemented to avoid resorting to less appropriate therapies in these situations. Following approval by the University and Medical Center Institutional Review Board, we designed a 4‐month study to assess the impact of this ongoing quality improvement measure from March 2012 to July 2012.

Hospitalized patients of all ages with reported penicillin allergies were obtained from our antimicrobial stewardship database. Their charts were reviewed for demographics, antibiotic use, clinical infection, and allergic description. Deciding whether to alter antibiotic therapy to a ‐lactam regimen was based on microbiologic results, laboratory values, clinical infection, and history of immunoglobulin E (IgE)‐mediated reactions, as defined by the updated drug allergy practice parameters.[5] IgE‐mediated reactions included: (1) immediate urticaria, laryngeal edema, or hypotension; (2) anemia; and (3) fever, arthralgias, lymphadenopathy, and an urticarial rash after 7 to 21 days.[5, 6, 7] We defined anaphylaxis as the development of angioedema or hemodynamic instability within 1 hour of penicillin administration. A true negative reaction was a lack of an IgE‐mediated reaction to all the drug challenges.

Patients in the medical, surgical, labor, and delivery wards; intensive care units; and emergency department underwent testing. The ‐lactam agent used after a negative PST was recorded, and the patients were followed for 24 hours after transitioning their therapy to a ‐lactam regimen. Excluded subjects included those with (1) nonIgE‐mediated reactions, (2) skin conditions that can give false positive results, (3) medications that may interfere with anaphylactic therapy, (4) history of severe exfoliative reactions to ‐lactams, (5) anaphylaxis less than 4 weeks prior, (6) allergies to antibiotics other than penicillin, and (7) uncertain allergy history.

Skin Testing Procedure

An infectious diseases fellow (R.H.R. or B.K.) was supervised in preparing for potential anaphylaxis, applying the reagents and interpreting the results based on drug allergy practice parameters.[5] The preliminary epicutaneous prick/puncture test was performed with a lancet in subjects without prior anaphylaxis using full‐strength PPL and penicillin G potassium reagents. If there was no response within 15 minutes, which we defined as a lack of wheal formation 3 mm or greater than that of the negative control, 0.02 to 0.03 mL of each reagent was injected intradermally using a tuberculin syringe and examined for 15 minutes.[5] If there was no response, patients were then challenged with either a single oral dose of penicillin V potassium 250 mg or whichever oral penicillin agent they previously reported an allergy to. If no reaction was appreciated within 2 hours, their therapy was changed to a ‐lactam agent including penicillins, cephalosporins, and carbapenems for the remaining duration of therapy (Figure 1) An estimate of NPV was obtained after 24 hours follow‐up.

Figure 1

RESULTS

A total of 4031 allergy histories were reviewed during the 5‐month study period to achieve the sample size of 146 patients (Table 1). Of those, 3885 were excluded (Figure 2). Common infections included pneumonias (26%) and urinary tract infections (20%) (Table 2) Only 1 subject had a positive reaction with hives, edema, and itching approximately 6 minutes after the agents were injected intradermally. The remaining 145 (99%) had negative reactions to the PST and oral challenge and were then successfully transitioned to a ‐lactam agent without any reaction at 24 hours, giving an NPV of 100%. Ten subjects were switched from intravenous to oral ‐lactam agents (Figure 1). Avoidance of PICC placement ($1,200) and removal ($65), dressing changes, weekly drug‐level testing, laboratory technician, and pharmaceutical drug calibration costs allowed for a healthcare reduction of $5,233 ($520/patient) based on the 146 patients studied. The total cost of therapy would have been $113,991 if the PST had not been performed. However, the cost of altered therapy following a negative PST was $81,180, a difference of $32,811 ($225/per patient) in a 5‐month period. The total estimated annual difference, including antibiotic alteration and associated drug‐costs, would be $82,000.

Figure 2

DISCUSSION

PST is the most rapid, sensitive, and cost‐effective modality for evaluating patients with immediate allergic reactions to penicillin. Over 90% of individuals with a true history of penicillin allergy have confirmed sensitivity with a PST, implying most patients who are skin tested negative are truly not allergic.[7, 9, 10, 11, 12] Our study shows that the reapproved PST with the PPL and penicillin G determinants continues to have a high NPV. A patient with a negative PST result is generally at a low risk of developing an immediate‐type hypersensitivity reaction to penicillin.[2, 11] PST frequently allowed for less expensive agents that would have been avoided due to a reported allergy. The estimated annual savings of $82,000 dollars from antibiotic alteration with successful transition to a ‐lactam agent after a negative PST illustrates its value, supports its validity, and makes this study novel.

Many ‐lactamase inhibitors (ie, piperacillin‐tazobactam), fourth generation cephalosporins (ie, cefepime), and carbapenems still remain costly. Despite this, we were still able to achieve a significant reduction in overall cost. In addition to financial benefits, PST allowed for the use of more appropriate agents with less potential adverse effects. Narrow‐spectrum, non‐lactam agents were sometimes altered to a broader‐spectrum ‐lactam agent. We also frequently tailored 2 agents to just 1 broad‐spectrum ‐lactam. This led to more patients being given broad‐spectrum agents after the PST (72 vs 89 patients). However, we were able to avoid using second‐line agents, such as aztreonam, vancomycin, linezolid, daptomycin, and tobramycin, in many patients with infections that are often best treated with penicillin‐based antibiotics (ie, syphilis, group B Streptococcus infections). With increasing incidence and recovery of multidrug‐resistant bacteria, PST may also allow use of potentially more effective antimicrobial agents.

A possible limitation is that our prevalence of a true penicillin allergy was 1%, whereas Bousquet et al. illustrate a higher prevalence of about 20%.[7] Although our prevalence may not be generalizable, Bousquet's study only assessed patients with allergies 5 years prior.

The introduction of PST into clinical practice will allow trained healthcare providers to prescribe cheaper, more appropriate, less toxic antimicrobial agents. The overall benefit of reintroducing penicillin agents when needed in the future is far more cost‐effective than what is described here. PST should become a standard of care when prescribing antibiotics to patients with a history of penicillin allergy. Medical providers should be aware of its utility, acquire training, and incorporate it into their practice.

Self‐reported penicillin allergy is common and frequently limits the available antimicrobial agents to choose from. This often results in the use of more expensive, potentially more toxic, and possibly less efficacious agents.[1, 2]

For over 30 years, penicilloyl‐polylysine (PPL) penicillin skin testing (PST) was widely used to diagnose penicillin allergy with a negative predictive value (NPV) of about 97% to 99%.[3] After being off the market for 5 years, PPL PST was reapproved in 2009 as PRE‐PEN.[4] However, many clinicians still fail to utilize PST despite its simplicity and substantial clinical impact. The main purpose of this study was to describe the predictive value of PST and impact on antibiotic selection in a sample of hospitalized patients with a reported history of penicillin allergy.

METHODS

In 2010, PST was introduced as a quality‐improvement measure after approval and support from the chief of professional services and the medical staff executive committee at Vidant Medical Center, an 861‐bed tertiary care and teaching hospital. Our antimicrobial stewardship program is regularly contacted for approval of alternative therapies in penicillin allergic patients. The PST quality‐improvement intervention was implemented to avoid resorting to less appropriate therapies in these situations. Following approval by the University and Medical Center Institutional Review Board, we designed a 4‐month study to assess the impact of this ongoing quality improvement measure from March 2012 to July 2012.

Hospitalized patients of all ages with reported penicillin allergies were obtained from our antimicrobial stewardship database. Their charts were reviewed for demographics, antibiotic use, clinical infection, and allergic description. Deciding whether to alter antibiotic therapy to a ‐lactam regimen was based on microbiologic results, laboratory values, clinical infection, and history of immunoglobulin E (IgE)‐mediated reactions, as defined by the updated drug allergy practice parameters.[5] IgE‐mediated reactions included: (1) immediate urticaria, laryngeal edema, or hypotension; (2) anemia; and (3) fever, arthralgias, lymphadenopathy, and an urticarial rash after 7 to 21 days.[5, 6, 7] We defined anaphylaxis as the development of angioedema or hemodynamic instability within 1 hour of penicillin administration. A true negative reaction was a lack of an IgE‐mediated reaction to all the drug challenges.

Patients in the medical, surgical, labor, and delivery wards; intensive care units; and emergency department underwent testing. The ‐lactam agent used after a negative PST was recorded, and the patients were followed for 24 hours after transitioning their therapy to a ‐lactam regimen. Excluded subjects included those with (1) nonIgE‐mediated reactions, (2) skin conditions that can give false positive results, (3) medications that may interfere with anaphylactic therapy, (4) history of severe exfoliative reactions to ‐lactams, (5) anaphylaxis less than 4 weeks prior, (6) allergies to antibiotics other than penicillin, and (7) uncertain allergy history.

Skin Testing Procedure

An infectious diseases fellow (R.H.R. or B.K.) was supervised in preparing for potential anaphylaxis, applying the reagents and interpreting the results based on drug allergy practice parameters.[5] The preliminary epicutaneous prick/puncture test was performed with a lancet in subjects without prior anaphylaxis using full‐strength PPL and penicillin G potassium reagents. If there was no response within 15 minutes, which we defined as a lack of wheal formation 3 mm or greater than that of the negative control, 0.02 to 0.03 mL of each reagent was injected intradermally using a tuberculin syringe and examined for 15 minutes.[5] If there was no response, patients were then challenged with either a single oral dose of penicillin V potassium 250 mg or whichever oral penicillin agent they previously reported an allergy to. If no reaction was appreciated within 2 hours, their therapy was changed to a ‐lactam agent including penicillins, cephalosporins, and carbapenems for the remaining duration of therapy (Figure 1) An estimate of NPV was obtained after 24 hours follow‐up.

Figure 1

RESULTS

A total of 4031 allergy histories were reviewed during the 5‐month study period to achieve the sample size of 146 patients (Table 1). Of those, 3885 were excluded (Figure 2). Common infections included pneumonias (26%) and urinary tract infections (20%) (Table 2) Only 1 subject had a positive reaction with hives, edema, and itching approximately 6 minutes after the agents were injected intradermally. The remaining 145 (99%) had negative reactions to the PST and oral challenge and were then successfully transitioned to a ‐lactam agent without any reaction at 24 hours, giving an NPV of 100%. Ten subjects were switched from intravenous to oral ‐lactam agents (Figure 1). Avoidance of PICC placement ($1,200) and removal ($65), dressing changes, weekly drug‐level testing, laboratory technician, and pharmaceutical drug calibration costs allowed for a healthcare reduction of $5,233 ($520/patient) based on the 146 patients studied. The total cost of therapy would have been $113,991 if the PST had not been performed. However, the cost of altered therapy following a negative PST was $81,180, a difference of $32,811 ($225/per patient) in a 5‐month period. The total estimated annual difference, including antibiotic alteration and associated drug‐costs, would be $82,000.

Figure 2