Brief Reports

The Institute of Medicine (IOM) reported over a decade ago that between 44,000 and 98,000 deaths occurred every year due to preventable medical errors.[1] The report sparked an intense interest in identifying, measuring, and reporting hospital performance in patient safety.[2] The report also sparked the implementation of many initiatives aiming to improve patient safety.[3] Despite these efforts, there is still much room for improvement in the area of patient safety.[4] As the public has become more aware of patient safety issues, there has been an increased demand for information on hospital safety. The Leapfrog Group, a leading organization that examines and reports on hospital performance in patient safety, cites the IOM report as providing the focus that their newly formed organization required.[5]

Using 26 national measures of safety, The Leapfrog Group calculates a numeric Hospital Safety Score for over 2,600 acute care hospitals in the United States.[6] The primary data used to calculate this score are collected through the Leapfrog Hospital Survey, the Agency for Healthcare Research and Quality, the Centers for Disease Control and Prevention, and the Centers for Medicare and Medicaid Services (CMS). The American Hospital Association's (AHA) Annual Survey is used as a secondary data source as necessary. The Leapfrog Group conducts the survey annually, and substantial efforts are put forth to invite hospital administrators to participate in the survey. Participation in the Leapfrog survey is optional and free of charge.

Leapfrog recently moved a step further in their evaluation of hospital safety by releasing the Hidden Surcharge Calculator to enable employers to estimate the hidden surcharge they pay for their employees and dependents because of hospital errors.[7] The calculation depends largely on the letter grade (AF) that the hospital received from Leapfrog's Hospital Safety Score. For example, Leapfrog estimated a commercially insured patient admitted to a hospital with a grade of C or lower would incur $1845 additional cost per admission than if the same patient was admitted to a hospital with a grade of A.[7] The Leapfrog group encourages employers and payers to use this information to adjust benefits structures so that employees are discouraged from using hospitals that receive lower hospital safety scores. Leapfrog also encourages payers to negotiate lower reimbursement rates for hospitals with lower hospital safety scores.

The accuracy of Leapfrog's hospital safety grades warrants attention because of the methodology used to score hospitals that do not participate in the Leapfrog Survey. One common barrier that prevents hospitals from participating is the amount of effort required to complete the annual survey, including extensive inputs from hospital executives and staff. According to Leapfrog, 4 to 6 days are required for a hospital to compile the necessary survey data.[8] Leapfrog estimates a 90‐minute commitment for the hospital chief executive officer or designated administrator to enter the information into the online questionnaire. This is a significant commitment for many hospitals. As a result, among the approximately 2600 acute care hospitals covered by Leapfrog's 2012 to 2013 safety grading, only 1100 (or 42.3%) actually participated in the Leapfrog hospital survey. This limits Leapfrog's ability to provide accurate scores and assign fair safety grades to many hospitals.

METHODS

RESULTS

Out of these 35 top hospitals, those that participated in the Leapfrog Survey generally received higher scores than the nonparticipants (Table 2). The group of participating hospitals received an average grade of A (mean safety score, 3.165; standard error of the mean [SE], 0.081), whereas the nonparticipating hospitals received an average grade of B (mean safety score, 3.012; SE, 0.047). These grades were consistent whether mean or median scores were used.

To further examine the potential bias against nonparticipating hospitals, the safety scores for each of the 17 nonparticipating hospitals were estimated as if they had participated in the Leapfrog Survey. The letter grade of this group increased from an average of B (mean safety score, 3.012; SE, 0.047) to an average of A (mean safety score, 3.216; SE, 0.046). Among the 17 nonparticipating hospitals, 15 showed an increase in safety score, of which 8 hospitals rescored a change in score significant enough to receive 1 or 2 letter grades higher (Table 3). Only 2 hospitals had slight decreases in safety score, without any impact on letter grade.

We applied the same methods to test the top 17 Honor Roll Hospitals as designated by US News & World Report; among them, half are participating hospitals and another half nonparticipating hospitals. One hospital, Johns Hopkins Hospital was not scored by Leapfrog because no relevant Medicare data are available for Leapfrog to calculate its safety score. For this reason, Johns Hopkins was excluded from our comparison. The results persist even with this smaller sample of top hospitals. The group of 8 participating hospitals had an average grade of A (mean safety score, 3.145; SE, 0.146), whereas another 8 nonparticipating hospitals received an average grade of B (mean safety score, 3.011; SE, 0.075).

DISCUSSION

The Leapfrog Group's intent to provide patient safety information to patients, physicians, healthcare purchasers, and hospital executives should be commended. However, the current methodology may disadvantage nonparticipating hospitals. The combination of lower maximum scores and increased weight of the CPOE and IPS scores may result in a lower hospital safety score than is justified. Nonparticipating hospitals may also face more intensive pressure from employers and payors to lower their reimbursement rates due to the newly released Leapfrog Hidden Surcharge Calculator.

Leapfrog acknowledges that the more data points a hospital has to be scored on, the better its opportunity to achieve a higher score.[8] This justification may lead to bias against nonparticipating hospitals. On the other hand, it is possible that hospitals with good safety records are more likely to participate in the Leapfrog Survey than those with poorer safety records. Without detailed nonresponse analysis from Leapfrog, it is impossible to know if there is a selection bias. Regardless, the Leapfrog result can subsequently misguide the payment rate negotiation between insurers and hospitals.

With this consideration in mind, Leapfrog should explicitly acknowledge the limitations of its methodology and consider revising it in future studies. For example, Leapfrog could only report on those measures for which there are data available for both participating and nonparticipating hospitals. Pending this revision, every effort must be made to distinguish between participating and nonparticipating hospitals. The outcomes of Leapfrog's hospital safety grades are made available online to consumers without distinguishing between participating and nonparticipating hospitals. The only method to differentiate the categories is to examine the data sources in detail amid a large volume of data. It is unlikely that consumers comparing hospital safety grades will take note of this caveat. Thus, Leapfrog's grading system can drastically misrepresent many nonparticipating hospitals' patient safety performances.

This study of The Leapfrog Group's Hospital Safety Score is not without limitations. The small sample utilized in this study limited the power of statistical testing. The difference in mean scores between participating and nonparticipating hospitals is not statistically significant. However, The Leapfrog Group uses specific numerical cutoff points for each letter grade classification. In this classification system statistical significance is not considered when assigning hospitals with different letter grades. It was clear that nonparticipating hospitals were more likely to receive lower letter grades than participating hospitals.

The small sample also posed challenges when attempting to account for missing data when comparing participating hospitals versus nonparticipating hospitals. Although a multiple imputation approach may have been ideal to address this, the small sample size coupled with the large amount of missing data (58% of hospitals did not participate in the Leapfrog Survey) led us to question the accuracy of this approach in this situation.[12] Instead, a crude, mean imputation approach was utilized, relying on the assumption that nonresponding hospitals had the same mean performance as responding hospitals on those domains where data were missing. In this study, we purposely selected a sample of hospitals from U.S. News & World Report's top hospitals. We believe the mean imputation approach, although not perfect, is appropriate for this sample of hospitals. Future study, however, should examine if hospitals that anticipated lower performance scores would be less likely to participate in the Leapfrog Survey. This would help strengthen Leapfrog's methodology in dealing with nonresponsive hospitals.

The Institute of Medicine (IOM) reported over a decade ago that between 44,000 and 98,000 deaths occurred every year due to preventable medical errors.[1] The report sparked an intense interest in identifying, measuring, and reporting hospital performance in patient safety.[2] The report also sparked the implementation of many initiatives aiming to improve patient safety.[3] Despite these efforts, there is still much room for improvement in the area of patient safety.[4] As the public has become more aware of patient safety issues, there has been an increased demand for information on hospital safety. The Leapfrog Group, a leading organization that examines and reports on hospital performance in patient safety, cites the IOM report as providing the focus that their newly formed organization required.[5]

Using 26 national measures of safety, The Leapfrog Group calculates a numeric Hospital Safety Score for over 2,600 acute care hospitals in the United States.[6] The primary data used to calculate this score are collected through the Leapfrog Hospital Survey, the Agency for Healthcare Research and Quality, the Centers for Disease Control and Prevention, and the Centers for Medicare and Medicaid Services (CMS). The American Hospital Association's (AHA) Annual Survey is used as a secondary data source as necessary. The Leapfrog Group conducts the survey annually, and substantial efforts are put forth to invite hospital administrators to participate in the survey. Participation in the Leapfrog survey is optional and free of charge.

Leapfrog recently moved a step further in their evaluation of hospital safety by releasing the Hidden Surcharge Calculator to enable employers to estimate the hidden surcharge they pay for their employees and dependents because of hospital errors.[7] The calculation depends largely on the letter grade (AF) that the hospital received from Leapfrog's Hospital Safety Score. For example, Leapfrog estimated a commercially insured patient admitted to a hospital with a grade of C or lower would incur $1845 additional cost per admission than if the same patient was admitted to a hospital with a grade of A.[7] The Leapfrog group encourages employers and payers to use this information to adjust benefits structures so that employees are discouraged from using hospitals that receive lower hospital safety scores. Leapfrog also encourages payers to negotiate lower reimbursement rates for hospitals with lower hospital safety scores.

The accuracy of Leapfrog's hospital safety grades warrants attention because of the methodology used to score hospitals that do not participate in the Leapfrog Survey. One common barrier that prevents hospitals from participating is the amount of effort required to complete the annual survey, including extensive inputs from hospital executives and staff. According to Leapfrog, 4 to 6 days are required for a hospital to compile the necessary survey data.[8] Leapfrog estimates a 90‐minute commitment for the hospital chief executive officer or designated administrator to enter the information into the online questionnaire. This is a significant commitment for many hospitals. As a result, among the approximately 2600 acute care hospitals covered by Leapfrog's 2012 to 2013 safety grading, only 1100 (or 42.3%) actually participated in the Leapfrog hospital survey. This limits Leapfrog's ability to provide accurate scores and assign fair safety grades to many hospitals.

METHODS

RESULTS

Out of these 35 top hospitals, those that participated in the Leapfrog Survey generally received higher scores than the nonparticipants (Table 2). The group of participating hospitals received an average grade of A (mean safety score, 3.165; standard error of the mean [SE], 0.081), whereas the nonparticipating hospitals received an average grade of B (mean safety score, 3.012; SE, 0.047). These grades were consistent whether mean or median scores were used.

To further examine the potential bias against nonparticipating hospitals, the safety scores for each of the 17 nonparticipating hospitals were estimated as if they had participated in the Leapfrog Survey. The letter grade of this group increased from an average of B (mean safety score, 3.012; SE, 0.047) to an average of A (mean safety score, 3.216; SE, 0.046). Among the 17 nonparticipating hospitals, 15 showed an increase in safety score, of which 8 hospitals rescored a change in score significant enough to receive 1 or 2 letter grades higher (Table 3). Only 2 hospitals had slight decreases in safety score, without any impact on letter grade.

We applied the same methods to test the top 17 Honor Roll Hospitals as designated by US News & World Report; among them, half are participating hospitals and another half nonparticipating hospitals. One hospital, Johns Hopkins Hospital was not scored by Leapfrog because no relevant Medicare data are available for Leapfrog to calculate its safety score. For this reason, Johns Hopkins was excluded from our comparison. The results persist even with this smaller sample of top hospitals. The group of 8 participating hospitals had an average grade of A (mean safety score, 3.145; SE, 0.146), whereas another 8 nonparticipating hospitals received an average grade of B (mean safety score, 3.011; SE, 0.075).

DISCUSSION

The Leapfrog Group's intent to provide patient safety information to patients, physicians, healthcare purchasers, and hospital executives should be commended. However, the current methodology may disadvantage nonparticipating hospitals. The combination of lower maximum scores and increased weight of the CPOE and IPS scores may result in a lower hospital safety score than is justified. Nonparticipating hospitals may also face more intensive pressure from employers and payors to lower their reimbursement rates due to the newly released Leapfrog Hidden Surcharge Calculator.

Leapfrog acknowledges that the more data points a hospital has to be scored on, the better its opportunity to achieve a higher score.[8] This justification may lead to bias against nonparticipating hospitals. On the other hand, it is possible that hospitals with good safety records are more likely to participate in the Leapfrog Survey than those with poorer safety records. Without detailed nonresponse analysis from Leapfrog, it is impossible to know if there is a selection bias. Regardless, the Leapfrog result can subsequently misguide the payment rate negotiation between insurers and hospitals.

With this consideration in mind, Leapfrog should explicitly acknowledge the limitations of its methodology and consider revising it in future studies. For example, Leapfrog could only report on those measures for which there are data available for both participating and nonparticipating hospitals. Pending this revision, every effort must be made to distinguish between participating and nonparticipating hospitals. The outcomes of Leapfrog's hospital safety grades are made available online to consumers without distinguishing between participating and nonparticipating hospitals. The only method to differentiate the categories is to examine the data sources in detail amid a large volume of data. It is unlikely that consumers comparing hospital safety grades will take note of this caveat. Thus, Leapfrog's grading system can drastically misrepresent many nonparticipating hospitals' patient safety performances.

This study of The Leapfrog Group's Hospital Safety Score is not without limitations. The small sample utilized in this study limited the power of statistical testing. The difference in mean scores between participating and nonparticipating hospitals is not statistically significant. However, The Leapfrog Group uses specific numerical cutoff points for each letter grade classification. In this classification system statistical significance is not considered when assigning hospitals with different letter grades. It was clear that nonparticipating hospitals were more likely to receive lower letter grades than participating hospitals.

The small sample also posed challenges when attempting to account for missing data when comparing participating hospitals versus nonparticipating hospitals. Although a multiple imputation approach may have been ideal to address this, the small sample size coupled with the large amount of missing data (58% of hospitals did not participate in the Leapfrog Survey) led us to question the accuracy of this approach in this situation.[12] Instead, a crude, mean imputation approach was utilized, relying on the assumption that nonresponding hospitals had the same mean performance as responding hospitals on those domains where data were missing. In this study, we purposely selected a sample of hospitals from U.S. News & World Report's top hospitals. We believe the mean imputation approach, although not perfect, is appropriate for this sample of hospitals. Future study, however, should examine if hospitals that anticipated lower performance scores would be less likely to participate in the Leapfrog Survey. This would help strengthen Leapfrog's methodology in dealing with nonresponsive hospitals.

The Institute of Medicine (IOM) reported over a decade ago that between 44,000 and 98,000 deaths occurred every year due to preventable medical errors.[1] The report sparked an intense interest in identifying, measuring, and reporting hospital performance in patient safety.[2] The report also sparked the implementation of many initiatives aiming to improve patient safety.[3] Despite these efforts, there is still much room for improvement in the area of patient safety.[4] As the public has become more aware of patient safety issues, there has been an increased demand for information on hospital safety. The Leapfrog Group, a leading organization that examines and reports on hospital performance in patient safety, cites the IOM report as providing the focus that their newly formed organization required.[5]

Using 26 national measures of safety, The Leapfrog Group calculates a numeric Hospital Safety Score for over 2,600 acute care hospitals in the United States.[6] The primary data used to calculate this score are collected through the Leapfrog Hospital Survey, the Agency for Healthcare Research and Quality, the Centers for Disease Control and Prevention, and the Centers for Medicare and Medicaid Services (CMS). The American Hospital Association's (AHA) Annual Survey is used as a secondary data source as necessary. The Leapfrog Group conducts the survey annually, and substantial efforts are put forth to invite hospital administrators to participate in the survey. Participation in the Leapfrog survey is optional and free of charge.

Leapfrog recently moved a step further in their evaluation of hospital safety by releasing the Hidden Surcharge Calculator to enable employers to estimate the hidden surcharge they pay for their employees and dependents because of hospital errors.[7] The calculation depends largely on the letter grade (AF) that the hospital received from Leapfrog's Hospital Safety Score. For example, Leapfrog estimated a commercially insured patient admitted to a hospital with a grade of C or lower would incur $1845 additional cost per admission than if the same patient was admitted to a hospital with a grade of A.[7] The Leapfrog group encourages employers and payers to use this information to adjust benefits structures so that employees are discouraged from using hospitals that receive lower hospital safety scores. Leapfrog also encourages payers to negotiate lower reimbursement rates for hospitals with lower hospital safety scores.

The accuracy of Leapfrog's hospital safety grades warrants attention because of the methodology used to score hospitals that do not participate in the Leapfrog Survey. One common barrier that prevents hospitals from participating is the amount of effort required to complete the annual survey, including extensive inputs from hospital executives and staff. According to Leapfrog, 4 to 6 days are required for a hospital to compile the necessary survey data.[8] Leapfrog estimates a 90‐minute commitment for the hospital chief executive officer or designated administrator to enter the information into the online questionnaire. This is a significant commitment for many hospitals. As a result, among the approximately 2600 acute care hospitals covered by Leapfrog's 2012 to 2013 safety grading, only 1100 (or 42.3%) actually participated in the Leapfrog hospital survey. This limits Leapfrog's ability to provide accurate scores and assign fair safety grades to many hospitals.

METHODS

RESULTS

Out of these 35 top hospitals, those that participated in the Leapfrog Survey generally received higher scores than the nonparticipants (Table 2). The group of participating hospitals received an average grade of A (mean safety score, 3.165; standard error of the mean [SE], 0.081), whereas the nonparticipating hospitals received an average grade of B (mean safety score, 3.012; SE, 0.047). These grades were consistent whether mean or median scores were used.

To further examine the potential bias against nonparticipating hospitals, the safety scores for each of the 17 nonparticipating hospitals were estimated as if they had participated in the Leapfrog Survey. The letter grade of this group increased from an average of B (mean safety score, 3.012; SE, 0.047) to an average of A (mean safety score, 3.216; SE, 0.046). Among the 17 nonparticipating hospitals, 15 showed an increase in safety score, of which 8 hospitals rescored a change in score significant enough to receive 1 or 2 letter grades higher (Table 3). Only 2 hospitals had slight decreases in safety score, without any impact on letter grade.

We applied the same methods to test the top 17 Honor Roll Hospitals as designated by US News & World Report; among them, half are participating hospitals and another half nonparticipating hospitals. One hospital, Johns Hopkins Hospital was not scored by Leapfrog because no relevant Medicare data are available for Leapfrog to calculate its safety score. For this reason, Johns Hopkins was excluded from our comparison. The results persist even with this smaller sample of top hospitals. The group of 8 participating hospitals had an average grade of A (mean safety score, 3.145; SE, 0.146), whereas another 8 nonparticipating hospitals received an average grade of B (mean safety score, 3.011; SE, 0.075).

DISCUSSION

The Leapfrog Group's intent to provide patient safety information to patients, physicians, healthcare purchasers, and hospital executives should be commended. However, the current methodology may disadvantage nonparticipating hospitals. The combination of lower maximum scores and increased weight of the CPOE and IPS scores may result in a lower hospital safety score than is justified. Nonparticipating hospitals may also face more intensive pressure from employers and payors to lower their reimbursement rates due to the newly released Leapfrog Hidden Surcharge Calculator.

Leapfrog acknowledges that the more data points a hospital has to be scored on, the better its opportunity to achieve a higher score.[8] This justification may lead to bias against nonparticipating hospitals. On the other hand, it is possible that hospitals with good safety records are more likely to participate in the Leapfrog Survey than those with poorer safety records. Without detailed nonresponse analysis from Leapfrog, it is impossible to know if there is a selection bias. Regardless, the Leapfrog result can subsequently misguide the payment rate negotiation between insurers and hospitals.

With this consideration in mind, Leapfrog should explicitly acknowledge the limitations of its methodology and consider revising it in future studies. For example, Leapfrog could only report on those measures for which there are data available for both participating and nonparticipating hospitals. Pending this revision, every effort must be made to distinguish between participating and nonparticipating hospitals. The outcomes of Leapfrog's hospital safety grades are made available online to consumers without distinguishing between participating and nonparticipating hospitals. The only method to differentiate the categories is to examine the data sources in detail amid a large volume of data. It is unlikely that consumers comparing hospital safety grades will take note of this caveat. Thus, Leapfrog's grading system can drastically misrepresent many nonparticipating hospitals' patient safety performances.

This study of The Leapfrog Group's Hospital Safety Score is not without limitations. The small sample utilized in this study limited the power of statistical testing. The difference in mean scores between participating and nonparticipating hospitals is not statistically significant. However, The Leapfrog Group uses specific numerical cutoff points for each letter grade classification. In this classification system statistical significance is not considered when assigning hospitals with different letter grades. It was clear that nonparticipating hospitals were more likely to receive lower letter grades than participating hospitals.

The small sample also posed challenges when attempting to account for missing data when comparing participating hospitals versus nonparticipating hospitals. Although a multiple imputation approach may have been ideal to address this, the small sample size coupled with the large amount of missing data (58% of hospitals did not participate in the Leapfrog Survey) led us to question the accuracy of this approach in this situation.[12] Instead, a crude, mean imputation approach was utilized, relying on the assumption that nonresponding hospitals had the same mean performance as responding hospitals on those domains where data were missing. In this study, we purposely selected a sample of hospitals from U.S. News & World Report's top hospitals. We believe the mean imputation approach, although not perfect, is appropriate for this sample of hospitals. Future study, however, should examine if hospitals that anticipated lower performance scores would be less likely to participate in the Leapfrog Survey. This would help strengthen Leapfrog's methodology in dealing with nonresponsive hospitals.

Suicide is the tenth leading cause of death in the United States,[1] resulting in the deaths of over 34,000 people each year.[2] In 2007, 165,997 individuals were hospitalized for self‐inflicted injuries, and 395,320 people were treated for self‐harm in emergency departments.[2] In 2003, the American Psychiatric Association reported that approximately 1500 suicides take place within hospital facilities in the United States each year.[3]

Although a number of studies have examined inpatient suicides that occurred on psychiatric units,[4, 5, 6, 7, 8] fewer have focused on suicides occurring on medical units. A Joint Commission review of inpatient suicide on medical/surgical units[9] found that 14.25% of all inpatient suicides occurred while the patient was on a medical unit, and now recommends that all hospitals identify individuals at risk for suicide, develop interventions for suicidal patients, and educate staff about the risk factors of suicide. Bostwick and Rackley[10] reviewed studies of suicide on medical/surgical units and found that few of the patients had histories of mental illness or suicidal ideation and recommend close attention to agitated patients, aggressively treating depression and pain, modifying the environment where possible, and observation of patients thought to be at risk. Wint and Alil[5] also report a high level of depression in patients who commit suicide in general hospitals and suggest that improved recognition of depression in general hospital patients will reduce suicide.

GOALS FOR THIS STUDY

Few studies have examined suicide on acute medical and surgical and intensive care units (ICUs), and there are no large studies conducted in the United States. The goal of this study was to describe suicide attempts and completions in the medical setting using Root Cause Analysis (RCA) reports of these events in the Veterans Health Administration (VHA).

METHODS

RESULTS

Our search resulted in 525 RCA reports of inpatient suicide attempts and completions. These were obtained from the 14,851 total RCA reports in the RCA dataset. Of the 525, we identified 50 cases that occurred while the patient was on the acute medical‐surgical unit (43 cases) or ICU (7 cases). Other cases occurred on mental health units, emergency department, or other areas of the hospital. Five cases were completed suicides, and 45 were suicide attempts. Based on the number of admissions per year reported above, the approximate rate of completed inpatient suicides on medical‐surgical and ICUs is 0.6 per million admissions. (For comparison, the rate of completed suicide on psychiatric units in the VA has been estimated to be 8.7 per million admissions.[14] Table 1 displays the admitting diagnosis and demographic data for those RCA reports that contained this information. The most common admitting diagnoses were alcohol detoxification and chest pain or rule out myocardial infarction (MI); note that 12 reports did not contain an admitting diagnosis. Table 2 displays the methods and root causes for the 50 cases; there were 118 root causes generated. The most common methods were cutting, overdose, and hanging; and the most common root causes were poor communication, need for staff training in suicide assessment, and need to improve suicide risk assessment.

DISCUSSION

This study examined the specific systemic factors involved in suicide attempts and completions in medical‐surgical and intensive care units in a large, national hospital sample. Overall, the number of completed suicides over the 13‐year period was small (5 in total). The most common reason for admission was alcohol detoxification. Many patients going through alcohol detoxification experience agitation, which is a risk factor for suicide among medical patients.[10] This hypothesis is further supported by the fact that 2 of the 5 completed suicides were admitted for alcohol detoxification. Interestingly, only 2 of the patients who attempted suicide in the hospital were also admitted for medical conditions related to a prior suicide attempt. It is likely the case that patients admitted for a suicide attempt are closely watched throughout the admission and so may have fewer opportunities to repeat the suicide attempt.

The most common method of suicide attempts was cutting with a sharp object. However, cutting did not result in death, whereas overdose, hanging, and gunshot did. As a precaution, especially with patients with a known history of suicidal ideation, removing sharp objects such as razor blades and knives as well as extra medications is a reasonable first step. It may also be possible to create safer bathroom environments, at least in some medical rooms for potentially suicidal patients, which have break‐away shower curtains, sealed grab‐bars, and a general reduction of anchor points for hanging (see https://www.naphs.org/resources/home.aspx?product‐tab=1). Another intervention that should be considered is systematic contraband searches, especially for sharp objects and firearms that could easily be used for self‐harm. Hospitals should develop clear plans for what steps should be taken if patients are identified to be a risk for suicide, and medical staff should be trained in the initiation and execution of this suicide prevention process.

As with other studies of RCAs,[15, 16] we found that problems communicating risk was the most common identified root cause for suicide attempts and completions. Problems communicating risk most often involved knowledge of suicide risk or specific suicide mitigation plans that were not shared by the treatment team or communicated during handoffs. Most frequently, this communication problem involved team members assessing a patient to be at high risk for suicide, but that information was not provided to other care team members. This root cause also included situations in which the treatment plan for suicide prevention was inadequately disseminated to the entire treatment team. It is critical that good systems are in place so that staff members have the time to communicate critical information about patients. In addition, the system should be standardized so that the same information is communicated each time there is a handoff. The lack of clear steps to mitigate suicide risk when a patient was identified at high risk was also a commonly cited root cause. The most extreme examples involved completed suicides occurring with a patient receiving 1‐on‐1 staffing. This 1‐on‐1 staffing did not include specific guidance for the sitters such as the need to remove personal items that could be used for self‐harm. We also saw that staff on medical units needed to learn more about risk factors for suicide and how to conduct a suicide assessment with their patients. Another root cause was the stress caused by the medical and psychiatric conditions of the patients. It is notable that no completed suicides occurred in ICUs, suggesting that closer observation and/or a higher level of medical incapacitation can reduce the risk of completed suicides.

To address these root causes, staff should be educated about risk factors for suicide, and standardized high‐risk for suicide order sets and checklists should be used to ensure staff execute the desired care processes and communicate them to all staff. In addition, specific training in suicide prevention should be provided to staff involved in 1‐on‐1 observation for high‐risk patients. Again, this may be aided by a checklist to help staff remember the protocol for what may be a low‐frequency event. A high risk suicide care process may include:

1. Conducting contraband searches for items that could be used for self‐harm, modifying the environment of a small percentage of toilet rooms on medical floors to reduce anchor points for hanging. A high risk patient could then be moved to these rooms.
2. Regular psychiatric input into the treatment plan.
3. Discharge planning that includes attention to the potential for depression and suicidal ideation upon discharge.

Suicide is the tenth leading cause of death in the United States,[1] resulting in the deaths of over 34,000 people each year.[2] In 2007, 165,997 individuals were hospitalized for self‐inflicted injuries, and 395,320 people were treated for self‐harm in emergency departments.[2] In 2003, the American Psychiatric Association reported that approximately 1500 suicides take place within hospital facilities in the United States each year.[3]

Although a number of studies have examined inpatient suicides that occurred on psychiatric units,[4, 5, 6, 7, 8] fewer have focused on suicides occurring on medical units. A Joint Commission review of inpatient suicide on medical/surgical units[9] found that 14.25% of all inpatient suicides occurred while the patient was on a medical unit, and now recommends that all hospitals identify individuals at risk for suicide, develop interventions for suicidal patients, and educate staff about the risk factors of suicide. Bostwick and Rackley[10] reviewed studies of suicide on medical/surgical units and found that few of the patients had histories of mental illness or suicidal ideation and recommend close attention to agitated patients, aggressively treating depression and pain, modifying the environment where possible, and observation of patients thought to be at risk. Wint and Alil[5] also report a high level of depression in patients who commit suicide in general hospitals and suggest that improved recognition of depression in general hospital patients will reduce suicide.

GOALS FOR THIS STUDY

Few studies have examined suicide on acute medical and surgical and intensive care units (ICUs), and there are no large studies conducted in the United States. The goal of this study was to describe suicide attempts and completions in the medical setting using Root Cause Analysis (RCA) reports of these events in the Veterans Health Administration (VHA).

METHODS

RESULTS

Our search resulted in 525 RCA reports of inpatient suicide attempts and completions. These were obtained from the 14,851 total RCA reports in the RCA dataset. Of the 525, we identified 50 cases that occurred while the patient was on the acute medical‐surgical unit (43 cases) or ICU (7 cases). Other cases occurred on mental health units, emergency department, or other areas of the hospital. Five cases were completed suicides, and 45 were suicide attempts. Based on the number of admissions per year reported above, the approximate rate of completed inpatient suicides on medical‐surgical and ICUs is 0.6 per million admissions. (For comparison, the rate of completed suicide on psychiatric units in the VA has been estimated to be 8.7 per million admissions.[14] Table 1 displays the admitting diagnosis and demographic data for those RCA reports that contained this information. The most common admitting diagnoses were alcohol detoxification and chest pain or rule out myocardial infarction (MI); note that 12 reports did not contain an admitting diagnosis. Table 2 displays the methods and root causes for the 50 cases; there were 118 root causes generated. The most common methods were cutting, overdose, and hanging; and the most common root causes were poor communication, need for staff training in suicide assessment, and need to improve suicide risk assessment.

DISCUSSION

This study examined the specific systemic factors involved in suicide attempts and completions in medical‐surgical and intensive care units in a large, national hospital sample. Overall, the number of completed suicides over the 13‐year period was small (5 in total). The most common reason for admission was alcohol detoxification. Many patients going through alcohol detoxification experience agitation, which is a risk factor for suicide among medical patients.[10] This hypothesis is further supported by the fact that 2 of the 5 completed suicides were admitted for alcohol detoxification. Interestingly, only 2 of the patients who attempted suicide in the hospital were also admitted for medical conditions related to a prior suicide attempt. It is likely the case that patients admitted for a suicide attempt are closely watched throughout the admission and so may have fewer opportunities to repeat the suicide attempt.

The most common method of suicide attempts was cutting with a sharp object. However, cutting did not result in death, whereas overdose, hanging, and gunshot did. As a precaution, especially with patients with a known history of suicidal ideation, removing sharp objects such as razor blades and knives as well as extra medications is a reasonable first step. It may also be possible to create safer bathroom environments, at least in some medical rooms for potentially suicidal patients, which have break‐away shower curtains, sealed grab‐bars, and a general reduction of anchor points for hanging (see https://www.naphs.org/resources/home.aspx?product‐tab=1). Another intervention that should be considered is systematic contraband searches, especially for sharp objects and firearms that could easily be used for self‐harm. Hospitals should develop clear plans for what steps should be taken if patients are identified to be a risk for suicide, and medical staff should be trained in the initiation and execution of this suicide prevention process.

As with other studies of RCAs,[15, 16] we found that problems communicating risk was the most common identified root cause for suicide attempts and completions. Problems communicating risk most often involved knowledge of suicide risk or specific suicide mitigation plans that were not shared by the treatment team or communicated during handoffs. Most frequently, this communication problem involved team members assessing a patient to be at high risk for suicide, but that information was not provided to other care team members. This root cause also included situations in which the treatment plan for suicide prevention was inadequately disseminated to the entire treatment team. It is critical that good systems are in place so that staff members have the time to communicate critical information about patients. In addition, the system should be standardized so that the same information is communicated each time there is a handoff. The lack of clear steps to mitigate suicide risk when a patient was identified at high risk was also a commonly cited root cause. The most extreme examples involved completed suicides occurring with a patient receiving 1‐on‐1 staffing. This 1‐on‐1 staffing did not include specific guidance for the sitters such as the need to remove personal items that could be used for self‐harm. We also saw that staff on medical units needed to learn more about risk factors for suicide and how to conduct a suicide assessment with their patients. Another root cause was the stress caused by the medical and psychiatric conditions of the patients. It is notable that no completed suicides occurred in ICUs, suggesting that closer observation and/or a higher level of medical incapacitation can reduce the risk of completed suicides.

To address these root causes, staff should be educated about risk factors for suicide, and standardized high‐risk for suicide order sets and checklists should be used to ensure staff execute the desired care processes and communicate them to all staff. In addition, specific training in suicide prevention should be provided to staff involved in 1‐on‐1 observation for high‐risk patients. Again, this may be aided by a checklist to help staff remember the protocol for what may be a low‐frequency event. A high risk suicide care process may include:

1. Conducting contraband searches for items that could be used for self‐harm, modifying the environment of a small percentage of toilet rooms on medical floors to reduce anchor points for hanging. A high risk patient could then be moved to these rooms.
2. Regular psychiatric input into the treatment plan.
3. Discharge planning that includes attention to the potential for depression and suicidal ideation upon discharge.

Suicide is the tenth leading cause of death in the United States,[1] resulting in the deaths of over 34,000 people each year.[2] In 2007, 165,997 individuals were hospitalized for self‐inflicted injuries, and 395,320 people were treated for self‐harm in emergency departments.[2] In 2003, the American Psychiatric Association reported that approximately 1500 suicides take place within hospital facilities in the United States each year.[3]

Although a number of studies have examined inpatient suicides that occurred on psychiatric units,[4, 5, 6, 7, 8] fewer have focused on suicides occurring on medical units. A Joint Commission review of inpatient suicide on medical/surgical units[9] found that 14.25% of all inpatient suicides occurred while the patient was on a medical unit, and now recommends that all hospitals identify individuals at risk for suicide, develop interventions for suicidal patients, and educate staff about the risk factors of suicide. Bostwick and Rackley[10] reviewed studies of suicide on medical/surgical units and found that few of the patients had histories of mental illness or suicidal ideation and recommend close attention to agitated patients, aggressively treating depression and pain, modifying the environment where possible, and observation of patients thought to be at risk. Wint and Alil[5] also report a high level of depression in patients who commit suicide in general hospitals and suggest that improved recognition of depression in general hospital patients will reduce suicide.

GOALS FOR THIS STUDY

Few studies have examined suicide on acute medical and surgical and intensive care units (ICUs), and there are no large studies conducted in the United States. The goal of this study was to describe suicide attempts and completions in the medical setting using Root Cause Analysis (RCA) reports of these events in the Veterans Health Administration (VHA).

METHODS

RESULTS

Our search resulted in 525 RCA reports of inpatient suicide attempts and completions. These were obtained from the 14,851 total RCA reports in the RCA dataset. Of the 525, we identified 50 cases that occurred while the patient was on the acute medical‐surgical unit (43 cases) or ICU (7 cases). Other cases occurred on mental health units, emergency department, or other areas of the hospital. Five cases were completed suicides, and 45 were suicide attempts. Based on the number of admissions per year reported above, the approximate rate of completed inpatient suicides on medical‐surgical and ICUs is 0.6 per million admissions. (For comparison, the rate of completed suicide on psychiatric units in the VA has been estimated to be 8.7 per million admissions.[14] Table 1 displays the admitting diagnosis and demographic data for those RCA reports that contained this information. The most common admitting diagnoses were alcohol detoxification and chest pain or rule out myocardial infarction (MI); note that 12 reports did not contain an admitting diagnosis. Table 2 displays the methods and root causes for the 50 cases; there were 118 root causes generated. The most common methods were cutting, overdose, and hanging; and the most common root causes were poor communication, need for staff training in suicide assessment, and need to improve suicide risk assessment.

DISCUSSION

This study examined the specific systemic factors involved in suicide attempts and completions in medical‐surgical and intensive care units in a large, national hospital sample. Overall, the number of completed suicides over the 13‐year period was small (5 in total). The most common reason for admission was alcohol detoxification. Many patients going through alcohol detoxification experience agitation, which is a risk factor for suicide among medical patients.[10] This hypothesis is further supported by the fact that 2 of the 5 completed suicides were admitted for alcohol detoxification. Interestingly, only 2 of the patients who attempted suicide in the hospital were also admitted for medical conditions related to a prior suicide attempt. It is likely the case that patients admitted for a suicide attempt are closely watched throughout the admission and so may have fewer opportunities to repeat the suicide attempt.

The most common method of suicide attempts was cutting with a sharp object. However, cutting did not result in death, whereas overdose, hanging, and gunshot did. As a precaution, especially with patients with a known history of suicidal ideation, removing sharp objects such as razor blades and knives as well as extra medications is a reasonable first step. It may also be possible to create safer bathroom environments, at least in some medical rooms for potentially suicidal patients, which have break‐away shower curtains, sealed grab‐bars, and a general reduction of anchor points for hanging (see https://www.naphs.org/resources/home.aspx?product‐tab=1). Another intervention that should be considered is systematic contraband searches, especially for sharp objects and firearms that could easily be used for self‐harm. Hospitals should develop clear plans for what steps should be taken if patients are identified to be a risk for suicide, and medical staff should be trained in the initiation and execution of this suicide prevention process.

As with other studies of RCAs,[15, 16] we found that problems communicating risk was the most common identified root cause for suicide attempts and completions. Problems communicating risk most often involved knowledge of suicide risk or specific suicide mitigation plans that were not shared by the treatment team or communicated during handoffs. Most frequently, this communication problem involved team members assessing a patient to be at high risk for suicide, but that information was not provided to other care team members. This root cause also included situations in which the treatment plan for suicide prevention was inadequately disseminated to the entire treatment team. It is critical that good systems are in place so that staff members have the time to communicate critical information about patients. In addition, the system should be standardized so that the same information is communicated each time there is a handoff. The lack of clear steps to mitigate suicide risk when a patient was identified at high risk was also a commonly cited root cause. The most extreme examples involved completed suicides occurring with a patient receiving 1‐on‐1 staffing. This 1‐on‐1 staffing did not include specific guidance for the sitters such as the need to remove personal items that could be used for self‐harm. We also saw that staff on medical units needed to learn more about risk factors for suicide and how to conduct a suicide assessment with their patients. Another root cause was the stress caused by the medical and psychiatric conditions of the patients. It is notable that no completed suicides occurred in ICUs, suggesting that closer observation and/or a higher level of medical incapacitation can reduce the risk of completed suicides.

To address these root causes, staff should be educated about risk factors for suicide, and standardized high‐risk for suicide order sets and checklists should be used to ensure staff execute the desired care processes and communicate them to all staff. In addition, specific training in suicide prevention should be provided to staff involved in 1‐on‐1 observation for high‐risk patients. Again, this may be aided by a checklist to help staff remember the protocol for what may be a low‐frequency event. A high risk suicide care process may include:

1. Conducting contraband searches for items that could be used for self‐harm, modifying the environment of a small percentage of toilet rooms on medical floors to reduce anchor points for hanging. A high risk patient could then be moved to these rooms.
2. Regular psychiatric input into the treatment plan.
3. Discharge planning that includes attention to the potential for depression and suicidal ideation upon discharge.

It's hard to imagine a busy urban hospital without its chorus of beepers.[1] This statement, the first sentence of an article published in 1988, rings (or beeps or buzzes) true to any resident physician today. At that time, pagers had replaced overhead paging, and provided a rapid method to contact physicians who were often scattered throughout the hospital. Still, it was an imperfect solution as the ubiquitous pager constantly interrupted patient care and other tasks, failed to prioritize information, and added to an already stressful working environment. Notably, interns were paged on average once per hour, and occasionally 5 or more times per hour, a frequency that was felt to be detrimental to patient care and to the working environment of resident physicians.[1]

Little has changed. Despite the instant, multidirectional communication platforms available today, alphanumeric paging remains a mainstay of communication between physicians and other members of the care team. Importantly, paging contributes to communication errors (eg, by failing to convey urgency, having incomplete information, or being missed entirely by coverage gaps),[2, 3] and interrupts resident workflow, thereby negatively affecting work efficiency and educational activities, and adding to perceived workload.[4, 5]

In this era of duty hour restrictions, there has been concern that residents experience increased workload due to having fewer hours to do the same amount of work.[6, 7] As such, the Accreditation Council of Graduate Medical Education emphasizes the quality of those hours, with a focus on several aspects of the resident working environment as key to improved educational and patient safety outcomes.[8, 9, 10]

Geographic localization of physicians to patient care units has been proposed as a means to improve communication and agreement on plans of care,[11, 12] and also to reduce resident workload by decreasing inefficiencies attributable to traveling throughout the hospital.[13] O'Leary, et al. (2009) found that when physicians were localized to 1 hospital unit, there was greater agreement between physicians and nurses on various aspects of care, such as planned tests and anticipated length of stay. In addition, members of the patient care team were better able to identify one another, and there was a perceived increase in face‐to‐face communication, and a perceived decrease in text paging.[11]

In consideration of these factors, in July 2011, at New YorkPresbyterian Hospital/Weill Cornell (NYPH/WC), an 800‐bed tertiary care teaching hospital in New York, New York, we geographically localized 2 internal medicine resident teams, and partially localized 2 additional teams. We investigated whether interns on teams that were geographically localized received fewer pages than interns on teams that were not localized. This study was reviewed by the institutional review board of Weill Cornell Medical College and met the requirements for exemption.

METHODS

We conducted a retrospective analysis of the number of pages received by interns during the day (7:00 _am to 7:00 _pm) on 5 general internal medicine teams during a 1‐month ward rotation between October 17, 2011 and November 13, 2011 at NYPH/WC. The general medicine teams were composed of 1 attending, 1 resident, and 2 interns each. Two teams were geographically localized to a 32‐bed unit (geographic localization model [GLM]). Two teams were partially localized to a 26‐bed unit, which included a respiratory care step‐down unit (partial localization model [PLM]). A fifth and final team admitted patients irrespective of their assigned bed location (standard model [SM]). Both the GLM and the PLM occasionally carried patients on other units to allow for overall census management and patient throughput. The total number of pages received by each intern over the study period was collected by retrospective analysis of electronic paging logs. Night pages (7 _pm7 _am) were excluded because of night float coverage. Weekend pages were excluded because data were inaccurate due to coverage for days off.

The daily number of admissions and daily census per team were recorded by physician assistants, who also assigned new patients to appropriate teams according to an admissions algorithm (see Supporting Figure 1 in the online version of this article). The percent of geographically localized patients on each team was estimated from the percentage of localized patients on the day of discharge averaged over the study period. For the SM team, percent localization was defined as the number of patients on the patient care unit that contained the team's work area.

Figure 1

Standard multivariate linear regression techniques were used to analyze the relationship between the number of pages received per intern and the type of team, controlling for the potential effect of total census and number of admissions. The regression model was used to determine adjusted marginal point estimates and 95% confidence intervals (CIs) for the average number of pages per intern per hour for each type of team. All statistical analyses were conducted using Stata version 12 (StataCorp, College Station, TX).

RESULTS

Over the 28‐day study period, a total of 6652 pages were received by 10 interns on 5 general internal medicine teams from 7 _am to 7 _pm Monday through Friday. The average daily census, average daily admissions, and percent of patients localized to patient care units for the individual teams are shown in Table 1. In univariate analysis, the mean daily pages per intern were not significantly different between the 2 teams within the GLM, nor between the 2 teams in the PLM, allowing them to be combined in multivariate analysis (data not shown). The number of pages received per intern per hour, adjusted for team census and number of admissions, was 2.2 (95% CI: 2.02.4) in the GLM, 2.8 (95% CI: 2.6‐3.0) in the PLM, and 3.9 (95% CI: 3.6‐4.2) in the SM (Table 1). All of these differences were statistically significant (P0.001).

Figure 1 shows the pattern of daytime paging for each model. The GLM and PLM had a similar pattern, with an initial ramp up in the first 2 hours of the day, holding steady until approximately 4 _pm, and then decrease until 7 _pm. The SM had a steeper initial rise, and then continued to increase slowly until a peak at 4 _pm.

DISCUSSION

This study corroborates that of Singh et al. (2012), who found that geographic localization led to significantly fewer pages.[14] Our results strengthen the evidence by demonstrating that even modest differences between the percent of patients localized to a care unit led to a significant decrease in the number of pages, indicating a dose‐response effect. The paging frequency we measured is higher than described in Singh et al. (1.4 pages per hour for localized teams), yet our average census appears to be 4 patients higher, which may account for some of that difference. We also show that interns on teams whose patients are more widely scattered throughout the hospital may experience upward of 5 pages per hour, an interruption by pager every 12 minutes, all day long.

A pager interruption is not solely limited to a disruption by noxious sound or vibration. The page recipient must then read the page and respond accordingly, which may involve a phone call, placing an order, walking to another location, or other work tasks. Although some of these interruptions must be handled immediately, such as a clinically deteriorating patient, many are not urgent, and could wait until the physician's current task or thought process is complete. There is also the potentially risky assumption on the part of the sender that the message has been received and will be acted upon. Furthermore, frequent paging is a common interruption to physician workflow; interruptions contribute to increased perceived physician workload[4, 5] and are likely detrimental to patient safety.[15, 16]

The most common metrics used to measure resident workload are patient census and number of admissions,[13] but these metrics have provided a mixed and likely incomplete picture. Recent research suggests that other factors, such as work efficiency (including interruptions, time spent obtaining test results, and time in transit) and work intensity (such as the acuity and complexity of patients), contribute significantly to actual and perceived resident workload.[13]

Our analysis was a single‐site, retrospective study, which occurred over 1 month and was limited to internal medicine teams. Additionally, geographic localization logically should lead to increased face‐to‐face interruptions, which we were unable to measure with this project, but direct communication is more efficient and less prone to error, which would likely lead to fewer overall interruptions. Although we anticipate that our findings are applicable to geographically localized patient care units in other hospitals, further investigation is warranted.

The paging chorus has only grown louder over the last 25 years, with likely downstream effects on patient safety and resident education. To mitigate these effects, it is incumbent upon us to approach our training and patient care environments with a critical and creative lens, and to explore opportunities to decrease interruptions and streamline our communication systems.

It's hard to imagine a busy urban hospital without its chorus of beepers.[1] This statement, the first sentence of an article published in 1988, rings (or beeps or buzzes) true to any resident physician today. At that time, pagers had replaced overhead paging, and provided a rapid method to contact physicians who were often scattered throughout the hospital. Still, it was an imperfect solution as the ubiquitous pager constantly interrupted patient care and other tasks, failed to prioritize information, and added to an already stressful working environment. Notably, interns were paged on average once per hour, and occasionally 5 or more times per hour, a frequency that was felt to be detrimental to patient care and to the working environment of resident physicians.[1]

Little has changed. Despite the instant, multidirectional communication platforms available today, alphanumeric paging remains a mainstay of communication between physicians and other members of the care team. Importantly, paging contributes to communication errors (eg, by failing to convey urgency, having incomplete information, or being missed entirely by coverage gaps),[2, 3] and interrupts resident workflow, thereby negatively affecting work efficiency and educational activities, and adding to perceived workload.[4, 5]

In this era of duty hour restrictions, there has been concern that residents experience increased workload due to having fewer hours to do the same amount of work.[6, 7] As such, the Accreditation Council of Graduate Medical Education emphasizes the quality of those hours, with a focus on several aspects of the resident working environment as key to improved educational and patient safety outcomes.[8, 9, 10]

Geographic localization of physicians to patient care units has been proposed as a means to improve communication and agreement on plans of care,[11, 12] and also to reduce resident workload by decreasing inefficiencies attributable to traveling throughout the hospital.[13] O'Leary, et al. (2009) found that when physicians were localized to 1 hospital unit, there was greater agreement between physicians and nurses on various aspects of care, such as planned tests and anticipated length of stay. In addition, members of the patient care team were better able to identify one another, and there was a perceived increase in face‐to‐face communication, and a perceived decrease in text paging.[11]

In consideration of these factors, in July 2011, at New YorkPresbyterian Hospital/Weill Cornell (NYPH/WC), an 800‐bed tertiary care teaching hospital in New York, New York, we geographically localized 2 internal medicine resident teams, and partially localized 2 additional teams. We investigated whether interns on teams that were geographically localized received fewer pages than interns on teams that were not localized. This study was reviewed by the institutional review board of Weill Cornell Medical College and met the requirements for exemption.

METHODS

We conducted a retrospective analysis of the number of pages received by interns during the day (7:00 _am to 7:00 _pm) on 5 general internal medicine teams during a 1‐month ward rotation between October 17, 2011 and November 13, 2011 at NYPH/WC. The general medicine teams were composed of 1 attending, 1 resident, and 2 interns each. Two teams were geographically localized to a 32‐bed unit (geographic localization model [GLM]). Two teams were partially localized to a 26‐bed unit, which included a respiratory care step‐down unit (partial localization model [PLM]). A fifth and final team admitted patients irrespective of their assigned bed location (standard model [SM]). Both the GLM and the PLM occasionally carried patients on other units to allow for overall census management and patient throughput. The total number of pages received by each intern over the study period was collected by retrospective analysis of electronic paging logs. Night pages (7 _pm7 _am) were excluded because of night float coverage. Weekend pages were excluded because data were inaccurate due to coverage for days off.

The daily number of admissions and daily census per team were recorded by physician assistants, who also assigned new patients to appropriate teams according to an admissions algorithm (see Supporting Figure 1 in the online version of this article). The percent of geographically localized patients on each team was estimated from the percentage of localized patients on the day of discharge averaged over the study period. For the SM team, percent localization was defined as the number of patients on the patient care unit that contained the team's work area.

Figure 1

Standard multivariate linear regression techniques were used to analyze the relationship between the number of pages received per intern and the type of team, controlling for the potential effect of total census and number of admissions. The regression model was used to determine adjusted marginal point estimates and 95% confidence intervals (CIs) for the average number of pages per intern per hour for each type of team. All statistical analyses were conducted using Stata version 12 (StataCorp, College Station, TX).

RESULTS

Over the 28‐day study period, a total of 6652 pages were received by 10 interns on 5 general internal medicine teams from 7 _am to 7 _pm Monday through Friday. The average daily census, average daily admissions, and percent of patients localized to patient care units for the individual teams are shown in Table 1. In univariate analysis, the mean daily pages per intern were not significantly different between the 2 teams within the GLM, nor between the 2 teams in the PLM, allowing them to be combined in multivariate analysis (data not shown). The number of pages received per intern per hour, adjusted for team census and number of admissions, was 2.2 (95% CI: 2.02.4) in the GLM, 2.8 (95% CI: 2.6‐3.0) in the PLM, and 3.9 (95% CI: 3.6‐4.2) in the SM (Table 1). All of these differences were statistically significant (P0.001).

Figure 1 shows the pattern of daytime paging for each model. The GLM and PLM had a similar pattern, with an initial ramp up in the first 2 hours of the day, holding steady until approximately 4 _pm, and then decrease until 7 _pm. The SM had a steeper initial rise, and then continued to increase slowly until a peak at 4 _pm.

DISCUSSION

This study corroborates that of Singh et al. (2012), who found that geographic localization led to significantly fewer pages.[14] Our results strengthen the evidence by demonstrating that even modest differences between the percent of patients localized to a care unit led to a significant decrease in the number of pages, indicating a dose‐response effect. The paging frequency we measured is higher than described in Singh et al. (1.4 pages per hour for localized teams), yet our average census appears to be 4 patients higher, which may account for some of that difference. We also show that interns on teams whose patients are more widely scattered throughout the hospital may experience upward of 5 pages per hour, an interruption by pager every 12 minutes, all day long.

A pager interruption is not solely limited to a disruption by noxious sound or vibration. The page recipient must then read the page and respond accordingly, which may involve a phone call, placing an order, walking to another location, or other work tasks. Although some of these interruptions must be handled immediately, such as a clinically deteriorating patient, many are not urgent, and could wait until the physician's current task or thought process is complete. There is also the potentially risky assumption on the part of the sender that the message has been received and will be acted upon. Furthermore, frequent paging is a common interruption to physician workflow; interruptions contribute to increased perceived physician workload[4, 5] and are likely detrimental to patient safety.[15, 16]

The most common metrics used to measure resident workload are patient census and number of admissions,[13] but these metrics have provided a mixed and likely incomplete picture. Recent research suggests that other factors, such as work efficiency (including interruptions, time spent obtaining test results, and time in transit) and work intensity (such as the acuity and complexity of patients), contribute significantly to actual and perceived resident workload.[13]

Our analysis was a single‐site, retrospective study, which occurred over 1 month and was limited to internal medicine teams. Additionally, geographic localization logically should lead to increased face‐to‐face interruptions, which we were unable to measure with this project, but direct communication is more efficient and less prone to error, which would likely lead to fewer overall interruptions. Although we anticipate that our findings are applicable to geographically localized patient care units in other hospitals, further investigation is warranted.

The paging chorus has only grown louder over the last 25 years, with likely downstream effects on patient safety and resident education. To mitigate these effects, it is incumbent upon us to approach our training and patient care environments with a critical and creative lens, and to explore opportunities to decrease interruptions and streamline our communication systems.

It's hard to imagine a busy urban hospital without its chorus of beepers.[1] This statement, the first sentence of an article published in 1988, rings (or beeps or buzzes) true to any resident physician today. At that time, pagers had replaced overhead paging, and provided a rapid method to contact physicians who were often scattered throughout the hospital. Still, it was an imperfect solution as the ubiquitous pager constantly interrupted patient care and other tasks, failed to prioritize information, and added to an already stressful working environment. Notably, interns were paged on average once per hour, and occasionally 5 or more times per hour, a frequency that was felt to be detrimental to patient care and to the working environment of resident physicians.[1]

Little has changed. Despite the instant, multidirectional communication platforms available today, alphanumeric paging remains a mainstay of communication between physicians and other members of the care team. Importantly, paging contributes to communication errors (eg, by failing to convey urgency, having incomplete information, or being missed entirely by coverage gaps),[2, 3] and interrupts resident workflow, thereby negatively affecting work efficiency and educational activities, and adding to perceived workload.[4, 5]

In this era of duty hour restrictions, there has been concern that residents experience increased workload due to having fewer hours to do the same amount of work.[6, 7] As such, the Accreditation Council of Graduate Medical Education emphasizes the quality of those hours, with a focus on several aspects of the resident working environment as key to improved educational and patient safety outcomes.[8, 9, 10]

Geographic localization of physicians to patient care units has been proposed as a means to improve communication and agreement on plans of care,[11, 12] and also to reduce resident workload by decreasing inefficiencies attributable to traveling throughout the hospital.[13] O'Leary, et al. (2009) found that when physicians were localized to 1 hospital unit, there was greater agreement between physicians and nurses on various aspects of care, such as planned tests and anticipated length of stay. In addition, members of the patient care team were better able to identify one another, and there was a perceived increase in face‐to‐face communication, and a perceived decrease in text paging.[11]

In consideration of these factors, in July 2011, at New YorkPresbyterian Hospital/Weill Cornell (NYPH/WC), an 800‐bed tertiary care teaching hospital in New York, New York, we geographically localized 2 internal medicine resident teams, and partially localized 2 additional teams. We investigated whether interns on teams that were geographically localized received fewer pages than interns on teams that were not localized. This study was reviewed by the institutional review board of Weill Cornell Medical College and met the requirements for exemption.

METHODS

We conducted a retrospective analysis of the number of pages received by interns during the day (7:00 _am to 7:00 _pm) on 5 general internal medicine teams during a 1‐month ward rotation between October 17, 2011 and November 13, 2011 at NYPH/WC. The general medicine teams were composed of 1 attending, 1 resident, and 2 interns each. Two teams were geographically localized to a 32‐bed unit (geographic localization model [GLM]). Two teams were partially localized to a 26‐bed unit, which included a respiratory care step‐down unit (partial localization model [PLM]). A fifth and final team admitted patients irrespective of their assigned bed location (standard model [SM]). Both the GLM and the PLM occasionally carried patients on other units to allow for overall census management and patient throughput. The total number of pages received by each intern over the study period was collected by retrospective analysis of electronic paging logs. Night pages (7 _pm7 _am) were excluded because of night float coverage. Weekend pages were excluded because data were inaccurate due to coverage for days off.

The daily number of admissions and daily census per team were recorded by physician assistants, who also assigned new patients to appropriate teams according to an admissions algorithm (see Supporting Figure 1 in the online version of this article). The percent of geographically localized patients on each team was estimated from the percentage of localized patients on the day of discharge averaged over the study period. For the SM team, percent localization was defined as the number of patients on the patient care unit that contained the team's work area.

Figure 1

Standard multivariate linear regression techniques were used to analyze the relationship between the number of pages received per intern and the type of team, controlling for the potential effect of total census and number of admissions. The regression model was used to determine adjusted marginal point estimates and 95% confidence intervals (CIs) for the average number of pages per intern per hour for each type of team. All statistical analyses were conducted using Stata version 12 (StataCorp, College Station, TX).

RESULTS

Over the 28‐day study period, a total of 6652 pages were received by 10 interns on 5 general internal medicine teams from 7 _am to 7 _pm Monday through Friday. The average daily census, average daily admissions, and percent of patients localized to patient care units for the individual teams are shown in Table 1. In univariate analysis, the mean daily pages per intern were not significantly different between the 2 teams within the GLM, nor between the 2 teams in the PLM, allowing them to be combined in multivariate analysis (data not shown). The number of pages received per intern per hour, adjusted for team census and number of admissions, was 2.2 (95% CI: 2.02.4) in the GLM, 2.8 (95% CI: 2.6‐3.0) in the PLM, and 3.9 (95% CI: 3.6‐4.2) in the SM (Table 1). All of these differences were statistically significant (P0.001).

Figure 1 shows the pattern of daytime paging for each model. The GLM and PLM had a similar pattern, with an initial ramp up in the first 2 hours of the day, holding steady until approximately 4 _pm, and then decrease until 7 _pm. The SM had a steeper initial rise, and then continued to increase slowly until a peak at 4 _pm.

DISCUSSION

This study corroborates that of Singh et al. (2012), who found that geographic localization led to significantly fewer pages.[14] Our results strengthen the evidence by demonstrating that even modest differences between the percent of patients localized to a care unit led to a significant decrease in the number of pages, indicating a dose‐response effect. The paging frequency we measured is higher than described in Singh et al. (1.4 pages per hour for localized teams), yet our average census appears to be 4 patients higher, which may account for some of that difference. We also show that interns on teams whose patients are more widely scattered throughout the hospital may experience upward of 5 pages per hour, an interruption by pager every 12 minutes, all day long.

A pager interruption is not solely limited to a disruption by noxious sound or vibration. The page recipient must then read the page and respond accordingly, which may involve a phone call, placing an order, walking to another location, or other work tasks. Although some of these interruptions must be handled immediately, such as a clinically deteriorating patient, many are not urgent, and could wait until the physician's current task or thought process is complete. There is also the potentially risky assumption on the part of the sender that the message has been received and will be acted upon. Furthermore, frequent paging is a common interruption to physician workflow; interruptions contribute to increased perceived physician workload[4, 5] and are likely detrimental to patient safety.[15, 16]

The most common metrics used to measure resident workload are patient census and number of admissions,[13] but these metrics have provided a mixed and likely incomplete picture. Recent research suggests that other factors, such as work efficiency (including interruptions, time spent obtaining test results, and time in transit) and work intensity (such as the acuity and complexity of patients), contribute significantly to actual and perceived resident workload.[13]

Our analysis was a single‐site, retrospective study, which occurred over 1 month and was limited to internal medicine teams. Additionally, geographic localization logically should lead to increased face‐to‐face interruptions, which we were unable to measure with this project, but direct communication is more efficient and less prone to error, which would likely lead to fewer overall interruptions. Although we anticipate that our findings are applicable to geographically localized patient care units in other hospitals, further investigation is warranted.

The paging chorus has only grown louder over the last 25 years, with likely downstream effects on patient safety and resident education. To mitigate these effects, it is incumbent upon us to approach our training and patient care environments with a critical and creative lens, and to explore opportunities to decrease interruptions and streamline our communication systems.

Bedside calculation of early warning system (EWS) scores is standard practice in many hospitals to predict clinical deterioration. These systems were designed for periodic hand‐scoring, typically using a half‐dozen variables dominated by vital signs. Most derive from the Modified Early Warning Score (MEWS).[1, 2] Despite years of modification, EWSs have had only modest impact on outcomes.[3, 4] Major improvement is possible only by adding more information than is contained in vital signs. Thus, the next generation of EWSs must analyze electronic medical records (EMRs). Analysis would be performed by computer, displayed automatically, and updated whenever new data are entered into the EMR. Such systems could deliver timely, accurate, longitudinally trended acuity information that could aid in earlier detection of declining patient condition as well as improving sensitivity and specificity of EWS alarms.

Advancing this endeavor along with others,[5, 6] we previously published a patient acuity metric, the Rothman Index (RI), which automatically updates when asynchronous vital signs, laboratory test results, Braden Scale,[7] cardiac rhythm, and nursing assessments are entered into the EMR.[8] Our goal was to enable clinicians to visualize changes in acuity by simple line graphs personalized to each patient at any point in time across the trajectory of care. In our model validation studies,[8] we made no attempt to identify generalizable thresholds, though others[9] have defined decision cut points for RI in a nonemergent context. To examine decision support feasibility in an emergent context, and to compare RI with a general EWS standard, we compare the accuracy of the RI with the MEWS in predicting hospital death within 24 hours.

METHODS

RESULTS

A total of 1,794,910 observations during 32,472 patient visits were included; 617 patients died (1.9%). Physiological characteristics for all input variables used by RI or MEWS are shown in Table 2, comparing observations taken within 24 hours of death to all other observations.

RI versus MEWS demonstrated superior discrimination of 24‐hour mortality (AUC was 0.93 [95% confidence interval {CI}: 0.92‐0.93] vs 0.82 [95% CI: 0.82‐0.83]; difference, 0.11 [95% CI: 0.10‐0.11]; P0.0001). ROC curves for RI and MEWS are shown in Figure 1; the MEWS is subsumed by RI across the entire range. Further, paired comparisons at points of clinical importance are presented in Table 3 for LR+, LR, sensitivity, specificity, PPV, and NPV. In the first pair of columns, MEWS=4 (typical trigger point for alarms) is matched to RI using sensitivity or LR; the corresponding point is RI=16, which generates twice the LR+ and reduces false alarms by 53%. In the second pair of columns, MEWS=4 is matched to RI using PPV or LR+; the corresponding point is RI=30, which captures 54% more of those patients who will die within 24 hours.

Figure 1

DISCUSSION

We have shown that a general acuity metric (RI) computed using data routinely entered into an EMR outperforms MEWS in identifying hospitalized patients likely to die within 24 hours. At similar sensitivity, RI yields an LR+ more than 2‐fold greater, at a value often considered conclusive. MEWS is derived using 4 vital signs and a neurologic assessment. Such a focus on vital signs may limit responsiveness to changes in acuity, especially during early clinical deterioration. Indeed, threshold breach tools may inadvertently induce a false sense of an individual patient's condition and safety.[12] The present findings suggest the performance of RI over MEWS may be due to inclusion of nursing assessments, laboratory test results, and heart rhythm. Relative contributions of each category are: vital signs (35%), nursing assessments (34%), and laboratory test results (31%). We found in previous work that failed nursing assessments strongly correlate with mortality,[13] as illustrated in Table 2 by sharp differences between patients dying within 24 hours and those who did not.

Sensitivity to detect early deterioration, especially when not evidenced by compromised vital signs, is crucial for acuity vigilance and preemptive interventions. Others[14] have demonstrated that our approach to longitudinal modeling of the acuity continuum is well positioned to investigate clinical pathophysiology preceding adverse events and to identify actionable trends in patients at high risk of complications and sepsis after colorectal operations. Future research may reveal both clinical and administrative advantages to having this real‐time acuity measure available for all patients during the entire hospital visit, with efficacy in applications beyond use as a trigger for EWS alarms.

Study limitations include retrospective design, single‐center cohort, no exclusion of expected hospital deaths, and EMR requirement. For MEWS, the Glasgow Coma Scale was mapped to A/V/P/U, which does not appear to affect results, as our c‐statistic is identical to the literature.[4] Any hospital with an EMR collects the data necessary for computation of RI values. The RI algorithms are available in software compatible with systems from numerous EMR manufacturers (eg, Epic, Cerner, McKesson, Siemens, AllScripts, Phillips).

The advent of the EMR in hospitals marries well with an EWS that leverages from additional data more information than is contained in vital signs, permitting complex numeric computations of acuity scores, a process simply not possible with paper systems. Further, the automatic recalculation of the score reduces the burden on clinicians, and broadens potential use over a wide range, from minute‐by‐minute recalculations when attached to sensors in the ICU, to comparative metrics of hospital performance, to nonclinical financial resource applications. This new information technology is guiding methods to achieve a significant performance increment over current EWS and may assist earlier detection of deterioration, providing a chance to avoid medical crises.[15]

Bedside calculation of early warning system (EWS) scores is standard practice in many hospitals to predict clinical deterioration. These systems were designed for periodic hand‐scoring, typically using a half‐dozen variables dominated by vital signs. Most derive from the Modified Early Warning Score (MEWS).[1, 2] Despite years of modification, EWSs have had only modest impact on outcomes.[3, 4] Major improvement is possible only by adding more information than is contained in vital signs. Thus, the next generation of EWSs must analyze electronic medical records (EMRs). Analysis would be performed by computer, displayed automatically, and updated whenever new data are entered into the EMR. Such systems could deliver timely, accurate, longitudinally trended acuity information that could aid in earlier detection of declining patient condition as well as improving sensitivity and specificity of EWS alarms.

Advancing this endeavor along with others,[5, 6] we previously published a patient acuity metric, the Rothman Index (RI), which automatically updates when asynchronous vital signs, laboratory test results, Braden Scale,[7] cardiac rhythm, and nursing assessments are entered into the EMR.[8] Our goal was to enable clinicians to visualize changes in acuity by simple line graphs personalized to each patient at any point in time across the trajectory of care. In our model validation studies,[8] we made no attempt to identify generalizable thresholds, though others[9] have defined decision cut points for RI in a nonemergent context. To examine decision support feasibility in an emergent context, and to compare RI with a general EWS standard, we compare the accuracy of the RI with the MEWS in predicting hospital death within 24 hours.

METHODS

RESULTS

A total of 1,794,910 observations during 32,472 patient visits were included; 617 patients died (1.9%). Physiological characteristics for all input variables used by RI or MEWS are shown in Table 2, comparing observations taken within 24 hours of death to all other observations.

RI versus MEWS demonstrated superior discrimination of 24‐hour mortality (AUC was 0.93 [95% confidence interval {CI}: 0.92‐0.93] vs 0.82 [95% CI: 0.82‐0.83]; difference, 0.11 [95% CI: 0.10‐0.11]; P0.0001). ROC curves for RI and MEWS are shown in Figure 1; the MEWS is subsumed by RI across the entire range. Further, paired comparisons at points of clinical importance are presented in Table 3 for LR+, LR, sensitivity, specificity, PPV, and NPV. In the first pair of columns, MEWS=4 (typical trigger point for alarms) is matched to RI using sensitivity or LR; the corresponding point is RI=16, which generates twice the LR+ and reduces false alarms by 53%. In the second pair of columns, MEWS=4 is matched to RI using PPV or LR+; the corresponding point is RI=30, which captures 54% more of those patients who will die within 24 hours.

Figure 1

DISCUSSION

We have shown that a general acuity metric (RI) computed using data routinely entered into an EMR outperforms MEWS in identifying hospitalized patients likely to die within 24 hours. At similar sensitivity, RI yields an LR+ more than 2‐fold greater, at a value often considered conclusive. MEWS is derived using 4 vital signs and a neurologic assessment. Such a focus on vital signs may limit responsiveness to changes in acuity, especially during early clinical deterioration. Indeed, threshold breach tools may inadvertently induce a false sense of an individual patient's condition and safety.[12] The present findings suggest the performance of RI over MEWS may be due to inclusion of nursing assessments, laboratory test results, and heart rhythm. Relative contributions of each category are: vital signs (35%), nursing assessments (34%), and laboratory test results (31%). We found in previous work that failed nursing assessments strongly correlate with mortality,[13] as illustrated in Table 2 by sharp differences between patients dying within 24 hours and those who did not.

Sensitivity to detect early deterioration, especially when not evidenced by compromised vital signs, is crucial for acuity vigilance and preemptive interventions. Others[14] have demonstrated that our approach to longitudinal modeling of the acuity continuum is well positioned to investigate clinical pathophysiology preceding adverse events and to identify actionable trends in patients at high risk of complications and sepsis after colorectal operations. Future research may reveal both clinical and administrative advantages to having this real‐time acuity measure available for all patients during the entire hospital visit, with efficacy in applications beyond use as a trigger for EWS alarms.

Study limitations include retrospective design, single‐center cohort, no exclusion of expected hospital deaths, and EMR requirement. For MEWS, the Glasgow Coma Scale was mapped to A/V/P/U, which does not appear to affect results, as our c‐statistic is identical to the literature.[4] Any hospital with an EMR collects the data necessary for computation of RI values. The RI algorithms are available in software compatible with systems from numerous EMR manufacturers (eg, Epic, Cerner, McKesson, Siemens, AllScripts, Phillips).

The advent of the EMR in hospitals marries well with an EWS that leverages from additional data more information than is contained in vital signs, permitting complex numeric computations of acuity scores, a process simply not possible with paper systems. Further, the automatic recalculation of the score reduces the burden on clinicians, and broadens potential use over a wide range, from minute‐by‐minute recalculations when attached to sensors in the ICU, to comparative metrics of hospital performance, to nonclinical financial resource applications. This new information technology is guiding methods to achieve a significant performance increment over current EWS and may assist earlier detection of deterioration, providing a chance to avoid medical crises.[15]

Bedside calculation of early warning system (EWS) scores is standard practice in many hospitals to predict clinical deterioration. These systems were designed for periodic hand‐scoring, typically using a half‐dozen variables dominated by vital signs. Most derive from the Modified Early Warning Score (MEWS).[1, 2] Despite years of modification, EWSs have had only modest impact on outcomes.[3, 4] Major improvement is possible only by adding more information than is contained in vital signs. Thus, the next generation of EWSs must analyze electronic medical records (EMRs). Analysis would be performed by computer, displayed automatically, and updated whenever new data are entered into the EMR. Such systems could deliver timely, accurate, longitudinally trended acuity information that could aid in earlier detection of declining patient condition as well as improving sensitivity and specificity of EWS alarms.

Advancing this endeavor along with others,[5, 6] we previously published a patient acuity metric, the Rothman Index (RI), which automatically updates when asynchronous vital signs, laboratory test results, Braden Scale,[7] cardiac rhythm, and nursing assessments are entered into the EMR.[8] Our goal was to enable clinicians to visualize changes in acuity by simple line graphs personalized to each patient at any point in time across the trajectory of care. In our model validation studies,[8] we made no attempt to identify generalizable thresholds, though others[9] have defined decision cut points for RI in a nonemergent context. To examine decision support feasibility in an emergent context, and to compare RI with a general EWS standard, we compare the accuracy of the RI with the MEWS in predicting hospital death within 24 hours.

METHODS

RESULTS

A total of 1,794,910 observations during 32,472 patient visits were included; 617 patients died (1.9%). Physiological characteristics for all input variables used by RI or MEWS are shown in Table 2, comparing observations taken within 24 hours of death to all other observations.

RI versus MEWS demonstrated superior discrimination of 24‐hour mortality (AUC was 0.93 [95% confidence interval {CI}: 0.92‐0.93] vs 0.82 [95% CI: 0.82‐0.83]; difference, 0.11 [95% CI: 0.10‐0.11]; P0.0001). ROC curves for RI and MEWS are shown in Figure 1; the MEWS is subsumed by RI across the entire range. Further, paired comparisons at points of clinical importance are presented in Table 3 for LR+, LR, sensitivity, specificity, PPV, and NPV. In the first pair of columns, MEWS=4 (typical trigger point for alarms) is matched to RI using sensitivity or LR; the corresponding point is RI=16, which generates twice the LR+ and reduces false alarms by 53%. In the second pair of columns, MEWS=4 is matched to RI using PPV or LR+; the corresponding point is RI=30, which captures 54% more of those patients who will die within 24 hours.

Figure 1

DISCUSSION

We have shown that a general acuity metric (RI) computed using data routinely entered into an EMR outperforms MEWS in identifying hospitalized patients likely to die within 24 hours. At similar sensitivity, RI yields an LR+ more than 2‐fold greater, at a value often considered conclusive. MEWS is derived using 4 vital signs and a neurologic assessment. Such a focus on vital signs may limit responsiveness to changes in acuity, especially during early clinical deterioration. Indeed, threshold breach tools may inadvertently induce a false sense of an individual patient's condition and safety.[12] The present findings suggest the performance of RI over MEWS may be due to inclusion of nursing assessments, laboratory test results, and heart rhythm. Relative contributions of each category are: vital signs (35%), nursing assessments (34%), and laboratory test results (31%). We found in previous work that failed nursing assessments strongly correlate with mortality,[13] as illustrated in Table 2 by sharp differences between patients dying within 24 hours and those who did not.

Sensitivity to detect early deterioration, especially when not evidenced by compromised vital signs, is crucial for acuity vigilance and preemptive interventions. Others[14] have demonstrated that our approach to longitudinal modeling of the acuity continuum is well positioned to investigate clinical pathophysiology preceding adverse events and to identify actionable trends in patients at high risk of complications and sepsis after colorectal operations. Future research may reveal both clinical and administrative advantages to having this real‐time acuity measure available for all patients during the entire hospital visit, with efficacy in applications beyond use as a trigger for EWS alarms.

Study limitations include retrospective design, single‐center cohort, no exclusion of expected hospital deaths, and EMR requirement. For MEWS, the Glasgow Coma Scale was mapped to A/V/P/U, which does not appear to affect results, as our c‐statistic is identical to the literature.[4] Any hospital with an EMR collects the data necessary for computation of RI values. The RI algorithms are available in software compatible with systems from numerous EMR manufacturers (eg, Epic, Cerner, McKesson, Siemens, AllScripts, Phillips).

The advent of the EMR in hospitals marries well with an EWS that leverages from additional data more information than is contained in vital signs, permitting complex numeric computations of acuity scores, a process simply not possible with paper systems. Further, the automatic recalculation of the score reduces the burden on clinicians, and broadens potential use over a wide range, from minute‐by‐minute recalculations when attached to sensors in the ICU, to comparative metrics of hospital performance, to nonclinical financial resource applications. This new information technology is guiding methods to achieve a significant performance increment over current EWS and may assist earlier detection of deterioration, providing a chance to avoid medical crises.[15]

Increasingly, there is a focus on prevention of hospital‐acquired conditions including venous thromboembolism (VTE). Many studies have evaluated pulmonary embolism (PE) and lower extremity deep vein thrombosis (LEDVT), but despite increasing recognition of upper extremity deep vein thrombosis (UEDVT),[1, 2, 3, 4] less is known about this condition in hospitalized patients.

UEDVTs may be classified as primary, including disorders such as Paget‐Schroetter syndrome or other structural abnormality, or may be idiopathic; the majority are secondary clots.[5] Conventional risk factors for LEDVT including older age and obesity have been found to be less commonly associated,[1, 2, 5, 6, 7] and patients with UEDVT are generally younger, leaner, and a higher proportion are men. They are more likely to have malignancy or history of VTE and have undergone recent surgery or intensive care unit stay.[1, 2, 6] Central venous catheters (CVCs), often used in hospitalized patients, remain among the biggest known risks for UEDVT[1, 2, 3, 7, 8, 9, 10]; concomitant malignancy, VTE history, severe infection, surgery lasting >1 hour, and length of stay (LOS) >10 days confer additional risks with CVCs.[6, 7, 8, 11]

UEDVTs, once thought to be relatively benign, are now recognized to result in complications including PE, progression, recurrence, and post‐thrombotic syndrome.[2, 4, 12, 13] Despite extensive efforts to increase appropriate VTE prophylaxis in inpatients,[14] the role of chemoprophylaxis to prevent UEDVT remains undefined. Current guidelines recommend anticoagulation for treatment and complication prevention,[13, 15] but to date the evidence derives largely from observational studies or is extrapolated from the LEDVT literature.[2, 13]

To improve understanding of UEDVT at our institution, we set out to (1) determine UEDVT incidence in hospitalized patients, (2) describe associated risks and outcomes, and (3) assess management during hospitalization and at discharge.

METHODS

We identified all consecutive adult patients diagnosed with Doppler ultrasound‐confirmed UEDVT during hospitalization at Harborview Medical Center between September 2011 and November 2012. For patients who were readmitted during the study period, the first of their hospitalizations was used to describe associated factors, management, and outcomes. We present characteristics of all other hospitalizations during this time period for comparison. Harborview is a 413‐bed academic tertiary referral center and the only level 1 trauma center in a 5‐state area. Patients with UEDVT were identified using an information technology (IT) tool (the Harborview VTE tool) (Figure 1), which captures VTE events from vascular laboratory and radiology studies using natural language processing. Doppler ultrasound to assess for deep vein thrombosis (DVT) and computed tomographic scans to diagnose PE were ordered by inpatient physicians for symptomatic patients. The reason for obtaining the study is included in the ultrasound reports. We do not routinely screen for UEDVT at our institution. UEDVT included clots in the deep veins of the upper extremities including internal jugular, subclavian, axillary, and brachial veins. Superficial thrombosis and thrombophlebitis were excluded. We previously compared VTE events captured by this tool with administrative billing data and found that all VTE events that were coded were captured with the tool.

Figure 1

The VTE tool (Figure 1) displays imaging results together with demographic, clinical, and medication data and links this information with admission, discharge, and death summaries as well as CVC insertion procedure notes from the electronic health record (EHR). Additional data, including comorbid conditions, primary reason for hospitalization, past medical history such as prior VTE events, and cause of death (if not available in the admission note or discharge/death summaries), were obtained from EHR abstraction by 1 of the investigators. A 10% random sample of charts was rereviewed by another investigator with complete concordance. Supplementary data about date of CVC insertion if placed at an outside facility, date of CVC removal if applicable, clinical assessments regarding whether a clot was CVC‐associated, and contraindications to therapeutic anticoagulation were also abstracted directly from the EHR. Administrative data were used to identify the case mix index, an indicator of severity of illness.

Pharmacologic VTE prophylaxis included all chemical prophylaxis specified on our institutional guideline, most commonly subcutaneous unfractionated heparin 5000 units every 8 hours or low molecular weight heparin (LMWH), either enoxaparin 40 mg every 12 or 24 hours or dalteparin 5000 units every 24 hours. Mechanical prophylaxis was defined as use of sequential compression devices (SCDs) when pharmacologic prophylaxis was contraindicated. Prophylaxis was considered to be appropriate if it was applied according to our guideline for >90% of hospital days prior to UEDVT diagnosis. Therapeutic anticoagulation included heparin bridging (most commonly continuous heparin infusion, LMWH 1 mg/kg or dalteparin) as well as oral vitamin K antagonists. The VTE tool (Figure 1) allows identification of pharmacologic prophylaxis and therapy that is actually administered (not just ordered) directly from our pharmacy IT system. SCD application (not just ordered SCDs) is electronically integrated into the tool from nursing documentation.

CVCs included internal jugular or subclavian triple lumen catheters, tunneled dialysis catheters, or peripherally inserted central catheters (PICCs), single or double lumen. Criteria used to identify that a UEDVT was CVC‐associated included temporal relationship (CVC was placed prior to clot diagnosis), plausibility (ipsilateral clot), evidence of clot surrounding CVC on ultrasound, and physician designation of association (as documented in progress notes or discharge summary).

Simple percentages of patient characteristics, associated factors, management, and outcomes were calculated using counts as the numerator and number of patients as the denominator. For information about UEDVTs, we used total number of UEDVTs as the denominator. Line days were day counts from insertion until removal if applicable. The CVC placement date was available in our mandated central line placement procedure notes (directly accessed from the VTE tool) for all lines placed at our institution; date of removal (if applicable) was determined from chart abstraction. For the vast majority of patients whose CVCs were placed at outside facilities, date of placement was available in the EHR (often in the admission note or in the ultrasound report/reason for study). If date of line placement at an outside facility was not known, date of admission was used. The University of Washington Human Subjects Board approved this review.

RESULTS

DISCUSSION

UEDVT is increasingly recognized in hospitalized patients.[3, 9] At our medical center, 0.2% of symptomatic inpatients were diagnosed with UEDVT over 14 months. These patients were predominantly men with high rates of CVCs, malignancy, VTE history, severe infection, and renal disease. Interestingly, although the literature suggests that some proportion of patients with UEDVT have anatomic abnormalities, such as Paget‐Schroetter syndrome,[15] none of the patients in our study were found to have these anomalies. In our review, hospitalized patients with UEDVT were critically ill, with a long LOS and high morbidity and mortality, suggesting that in addition to just being a complication of hospitalization,[1, 6] UEDVT may be a marker of severe illness.

In our institution, clinical presentation was consistent with what has been described with the majority of patients presenting with upper extremity swelling.[1, 3] The internal jugular veins were the most common anatomic UEDVT site, followed by brachial then axillary veins. In other series including both in‐ and outpatients, subclavian clots were most commonly diagnosed, reflecting in part higher rates of CVC association and CVC location in those studies.[3, 9] Concurrent DVT and PE rates were similar to those reported.[1, 3, 10]

Although many studies have focused on prevention of LEDVT and PE, few trials have specifically targeted UEDVT. Among our patients with UEDVTs that were not present on admission, VTE prophylaxis rates were considerably higher than what has been reported,[1, 6] suggesting that in these critically ill patients' prophylaxis may not prevent symptomatic UEDVT. It is unknown how many UEDVTs were prevented with prophylaxis, as only patients with symptomatic UEDVT were included. Adequacy of prophylaxis at outside hospitals for patients transferred in could not be assessed. Nonetheless, low numbers of UEDVT at a trauma referral center with many high‐risk patients raise the question of whether prophylaxis makes a difference. Additional study is needed to further define the role of chemoprophylaxis to prevent UEDVT in hospitalized patients.

In our inpatient group, 84% required CVCs; 44% of patients were thought to have CVC‐associated UEDVTs. Careful patient selection and attention to potentially modifiable risks, such as insertion site, catheter type, and tip position, may need further examination in this population.[3, 11, 16] Catheter duration was long; focus on removing CVCs when no longer necessary is important. Interestingly, almost 10% in our study underwent diagnostic ultrasound because a new CVC could not be successfully placed suggesting that UEDVT may develop in critically ill patients regardless of CVCs.

In our study, there were high rates of guideline‐recommended pharmacologic treatment; surprisingly the majority of CVCs with associated clot were removed. Guidelines currently support 3 months of anticoagulation for treatment of UEDVT[2, 13, 17]; evidence derives from observational trials or is largely extrapolated from LEDVT literature.[2, 13] Routine CVC removal is not specifically recommended for CVC‐associated UEDVT, particularly if lines remain functional and medically necessary; systemic anticoagulation should be provided.[13]

In our review, no hospitalized patients with UEDVT developed complications or were readmitted to our medical center within 30 days for clot progression, new PE, or post‐thrombotic syndrome, which is lower than rates reported over longer time periods.[2, 6, 10, 12] Ten percent died during hospitalization, all from their primary disease rather than from complications of VTE or VTE treatment, and no additional patients died at our institution within 30 days. Although these rates are lower than have been otherwise reported,[2, 10] the inpatient mortality rate is similar to a recent study that included inpatients; however, all patients who died in that study had cancer and CVCs.[3] In the latter study, 6.4% died within 30 days of discharge.

CONCLUSIONS

Among hospitalized patients, UEDVT is increasingly recognized. In our medical center, hospitalized patients diagnosed with UEDVT were more likely to have CVCs, malignancy, renal disease, and severe infection. Many of these patients were transferred critically ill, had prolonged LOS, and had high in‐hospital mortality. Most developed UEDVT despite prophylaxis, and the majority of UEDVTs were treated even in the absence of concurrent LEDVT or PE. As we move toward an era of increasing accountability, with a focus on preventing hospital‐acquired conditions including VTE, additional research is needed to identify modifiable risks, explore opportunities for effective prevention, and optimize outcomes such as prevention of complications or readmissions, particularly in critically ill patients with UEDVT.

Increasingly, there is a focus on prevention of hospital‐acquired conditions including venous thromboembolism (VTE). Many studies have evaluated pulmonary embolism (PE) and lower extremity deep vein thrombosis (LEDVT), but despite increasing recognition of upper extremity deep vein thrombosis (UEDVT),[1, 2, 3, 4] less is known about this condition in hospitalized patients.

UEDVTs may be classified as primary, including disorders such as Paget‐Schroetter syndrome or other structural abnormality, or may be idiopathic; the majority are secondary clots.[5] Conventional risk factors for LEDVT including older age and obesity have been found to be less commonly associated,[1, 2, 5, 6, 7] and patients with UEDVT are generally younger, leaner, and a higher proportion are men. They are more likely to have malignancy or history of VTE and have undergone recent surgery or intensive care unit stay.[1, 2, 6] Central venous catheters (CVCs), often used in hospitalized patients, remain among the biggest known risks for UEDVT[1, 2, 3, 7, 8, 9, 10]; concomitant malignancy, VTE history, severe infection, surgery lasting >1 hour, and length of stay (LOS) >10 days confer additional risks with CVCs.[6, 7, 8, 11]

UEDVTs, once thought to be relatively benign, are now recognized to result in complications including PE, progression, recurrence, and post‐thrombotic syndrome.[2, 4, 12, 13] Despite extensive efforts to increase appropriate VTE prophylaxis in inpatients,[14] the role of chemoprophylaxis to prevent UEDVT remains undefined. Current guidelines recommend anticoagulation for treatment and complication prevention,[13, 15] but to date the evidence derives largely from observational studies or is extrapolated from the LEDVT literature.[2, 13]

To improve understanding of UEDVT at our institution, we set out to (1) determine UEDVT incidence in hospitalized patients, (2) describe associated risks and outcomes, and (3) assess management during hospitalization and at discharge.

METHODS

We identified all consecutive adult patients diagnosed with Doppler ultrasound‐confirmed UEDVT during hospitalization at Harborview Medical Center between September 2011 and November 2012. For patients who were readmitted during the study period, the first of their hospitalizations was used to describe associated factors, management, and outcomes. We present characteristics of all other hospitalizations during this time period for comparison. Harborview is a 413‐bed academic tertiary referral center and the only level 1 trauma center in a 5‐state area. Patients with UEDVT were identified using an information technology (IT) tool (the Harborview VTE tool) (Figure 1), which captures VTE events from vascular laboratory and radiology studies using natural language processing. Doppler ultrasound to assess for deep vein thrombosis (DVT) and computed tomographic scans to diagnose PE were ordered by inpatient physicians for symptomatic patients. The reason for obtaining the study is included in the ultrasound reports. We do not routinely screen for UEDVT at our institution. UEDVT included clots in the deep veins of the upper extremities including internal jugular, subclavian, axillary, and brachial veins. Superficial thrombosis and thrombophlebitis were excluded. We previously compared VTE events captured by this tool with administrative billing data and found that all VTE events that were coded were captured with the tool.

Figure 1

The VTE tool (Figure 1) displays imaging results together with demographic, clinical, and medication data and links this information with admission, discharge, and death summaries as well as CVC insertion procedure notes from the electronic health record (EHR). Additional data, including comorbid conditions, primary reason for hospitalization, past medical history such as prior VTE events, and cause of death (if not available in the admission note or discharge/death summaries), were obtained from EHR abstraction by 1 of the investigators. A 10% random sample of charts was rereviewed by another investigator with complete concordance. Supplementary data about date of CVC insertion if placed at an outside facility, date of CVC removal if applicable, clinical assessments regarding whether a clot was CVC‐associated, and contraindications to therapeutic anticoagulation were also abstracted directly from the EHR. Administrative data were used to identify the case mix index, an indicator of severity of illness.

Pharmacologic VTE prophylaxis included all chemical prophylaxis specified on our institutional guideline, most commonly subcutaneous unfractionated heparin 5000 units every 8 hours or low molecular weight heparin (LMWH), either enoxaparin 40 mg every 12 or 24 hours or dalteparin 5000 units every 24 hours. Mechanical prophylaxis was defined as use of sequential compression devices (SCDs) when pharmacologic prophylaxis was contraindicated. Prophylaxis was considered to be appropriate if it was applied according to our guideline for >90% of hospital days prior to UEDVT diagnosis. Therapeutic anticoagulation included heparin bridging (most commonly continuous heparin infusion, LMWH 1 mg/kg or dalteparin) as well as oral vitamin K antagonists. The VTE tool (Figure 1) allows identification of pharmacologic prophylaxis and therapy that is actually administered (not just ordered) directly from our pharmacy IT system. SCD application (not just ordered SCDs) is electronically integrated into the tool from nursing documentation.

CVCs included internal jugular or subclavian triple lumen catheters, tunneled dialysis catheters, or peripherally inserted central catheters (PICCs), single or double lumen. Criteria used to identify that a UEDVT was CVC‐associated included temporal relationship (CVC was placed prior to clot diagnosis), plausibility (ipsilateral clot), evidence of clot surrounding CVC on ultrasound, and physician designation of association (as documented in progress notes or discharge summary).

Simple percentages of patient characteristics, associated factors, management, and outcomes were calculated using counts as the numerator and number of patients as the denominator. For information about UEDVTs, we used total number of UEDVTs as the denominator. Line days were day counts from insertion until removal if applicable. The CVC placement date was available in our mandated central line placement procedure notes (directly accessed from the VTE tool) for all lines placed at our institution; date of removal (if applicable) was determined from chart abstraction. For the vast majority of patients whose CVCs were placed at outside facilities, date of placement was available in the EHR (often in the admission note or in the ultrasound report/reason for study). If date of line placement at an outside facility was not known, date of admission was used. The University of Washington Human Subjects Board approved this review.

RESULTS

DISCUSSION

UEDVT is increasingly recognized in hospitalized patients.[3, 9] At our medical center, 0.2% of symptomatic inpatients were diagnosed with UEDVT over 14 months. These patients were predominantly men with high rates of CVCs, malignancy, VTE history, severe infection, and renal disease. Interestingly, although the literature suggests that some proportion of patients with UEDVT have anatomic abnormalities, such as Paget‐Schroetter syndrome,[15] none of the patients in our study were found to have these anomalies. In our review, hospitalized patients with UEDVT were critically ill, with a long LOS and high morbidity and mortality, suggesting that in addition to just being a complication of hospitalization,[1, 6] UEDVT may be a marker of severe illness.

In our institution, clinical presentation was consistent with what has been described with the majority of patients presenting with upper extremity swelling.[1, 3] The internal jugular veins were the most common anatomic UEDVT site, followed by brachial then axillary veins. In other series including both in‐ and outpatients, subclavian clots were most commonly diagnosed, reflecting in part higher rates of CVC association and CVC location in those studies.[3, 9] Concurrent DVT and PE rates were similar to those reported.[1, 3, 10]

Although many studies have focused on prevention of LEDVT and PE, few trials have specifically targeted UEDVT. Among our patients with UEDVTs that were not present on admission, VTE prophylaxis rates were considerably higher than what has been reported,[1, 6] suggesting that in these critically ill patients' prophylaxis may not prevent symptomatic UEDVT. It is unknown how many UEDVTs were prevented with prophylaxis, as only patients with symptomatic UEDVT were included. Adequacy of prophylaxis at outside hospitals for patients transferred in could not be assessed. Nonetheless, low numbers of UEDVT at a trauma referral center with many high‐risk patients raise the question of whether prophylaxis makes a difference. Additional study is needed to further define the role of chemoprophylaxis to prevent UEDVT in hospitalized patients.

In our inpatient group, 84% required CVCs; 44% of patients were thought to have CVC‐associated UEDVTs. Careful patient selection and attention to potentially modifiable risks, such as insertion site, catheter type, and tip position, may need further examination in this population.[3, 11, 16] Catheter duration was long; focus on removing CVCs when no longer necessary is important. Interestingly, almost 10% in our study underwent diagnostic ultrasound because a new CVC could not be successfully placed suggesting that UEDVT may develop in critically ill patients regardless of CVCs.

In our study, there were high rates of guideline‐recommended pharmacologic treatment; surprisingly the majority of CVCs with associated clot were removed. Guidelines currently support 3 months of anticoagulation for treatment of UEDVT[2, 13, 17]; evidence derives from observational trials or is largely extrapolated from LEDVT literature.[2, 13] Routine CVC removal is not specifically recommended for CVC‐associated UEDVT, particularly if lines remain functional and medically necessary; systemic anticoagulation should be provided.[13]

In our review, no hospitalized patients with UEDVT developed complications or were readmitted to our medical center within 30 days for clot progression, new PE, or post‐thrombotic syndrome, which is lower than rates reported over longer time periods.[2, 6, 10, 12] Ten percent died during hospitalization, all from their primary disease rather than from complications of VTE or VTE treatment, and no additional patients died at our institution within 30 days. Although these rates are lower than have been otherwise reported,[2, 10] the inpatient mortality rate is similar to a recent study that included inpatients; however, all patients who died in that study had cancer and CVCs.[3] In the latter study, 6.4% died within 30 days of discharge.

CONCLUSIONS

Among hospitalized patients, UEDVT is increasingly recognized. In our medical center, hospitalized patients diagnosed with UEDVT were more likely to have CVCs, malignancy, renal disease, and severe infection. Many of these patients were transferred critically ill, had prolonged LOS, and had high in‐hospital mortality. Most developed UEDVT despite prophylaxis, and the majority of UEDVTs were treated even in the absence of concurrent LEDVT or PE. As we move toward an era of increasing accountability, with a focus on preventing hospital‐acquired conditions including VTE, additional research is needed to identify modifiable risks, explore opportunities for effective prevention, and optimize outcomes such as prevention of complications or readmissions, particularly in critically ill patients with UEDVT.

Increasingly, there is a focus on prevention of hospital‐acquired conditions including venous thromboembolism (VTE). Many studies have evaluated pulmonary embolism (PE) and lower extremity deep vein thrombosis (LEDVT), but despite increasing recognition of upper extremity deep vein thrombosis (UEDVT),[1, 2, 3, 4] less is known about this condition in hospitalized patients.

UEDVTs may be classified as primary, including disorders such as Paget‐Schroetter syndrome or other structural abnormality, or may be idiopathic; the majority are secondary clots.[5] Conventional risk factors for LEDVT including older age and obesity have been found to be less commonly associated,[1, 2, 5, 6, 7] and patients with UEDVT are generally younger, leaner, and a higher proportion are men. They are more likely to have malignancy or history of VTE and have undergone recent surgery or intensive care unit stay.[1, 2, 6] Central venous catheters (CVCs), often used in hospitalized patients, remain among the biggest known risks for UEDVT[1, 2, 3, 7, 8, 9, 10]; concomitant malignancy, VTE history, severe infection, surgery lasting >1 hour, and length of stay (LOS) >10 days confer additional risks with CVCs.[6, 7, 8, 11]

UEDVTs, once thought to be relatively benign, are now recognized to result in complications including PE, progression, recurrence, and post‐thrombotic syndrome.[2, 4, 12, 13] Despite extensive efforts to increase appropriate VTE prophylaxis in inpatients,[14] the role of chemoprophylaxis to prevent UEDVT remains undefined. Current guidelines recommend anticoagulation for treatment and complication prevention,[13, 15] but to date the evidence derives largely from observational studies or is extrapolated from the LEDVT literature.[2, 13]

To improve understanding of UEDVT at our institution, we set out to (1) determine UEDVT incidence in hospitalized patients, (2) describe associated risks and outcomes, and (3) assess management during hospitalization and at discharge.

METHODS

We identified all consecutive adult patients diagnosed with Doppler ultrasound‐confirmed UEDVT during hospitalization at Harborview Medical Center between September 2011 and November 2012. For patients who were readmitted during the study period, the first of their hospitalizations was used to describe associated factors, management, and outcomes. We present characteristics of all other hospitalizations during this time period for comparison. Harborview is a 413‐bed academic tertiary referral center and the only level 1 trauma center in a 5‐state area. Patients with UEDVT were identified using an information technology (IT) tool (the Harborview VTE tool) (Figure 1), which captures VTE events from vascular laboratory and radiology studies using natural language processing. Doppler ultrasound to assess for deep vein thrombosis (DVT) and computed tomographic scans to diagnose PE were ordered by inpatient physicians for symptomatic patients. The reason for obtaining the study is included in the ultrasound reports. We do not routinely screen for UEDVT at our institution. UEDVT included clots in the deep veins of the upper extremities including internal jugular, subclavian, axillary, and brachial veins. Superficial thrombosis and thrombophlebitis were excluded. We previously compared VTE events captured by this tool with administrative billing data and found that all VTE events that were coded were captured with the tool.

Figure 1

The VTE tool (Figure 1) displays imaging results together with demographic, clinical, and medication data and links this information with admission, discharge, and death summaries as well as CVC insertion procedure notes from the electronic health record (EHR). Additional data, including comorbid conditions, primary reason for hospitalization, past medical history such as prior VTE events, and cause of death (if not available in the admission note or discharge/death summaries), were obtained from EHR abstraction by 1 of the investigators. A 10% random sample of charts was rereviewed by another investigator with complete concordance. Supplementary data about date of CVC insertion if placed at an outside facility, date of CVC removal if applicable, clinical assessments regarding whether a clot was CVC‐associated, and contraindications to therapeutic anticoagulation were also abstracted directly from the EHR. Administrative data were used to identify the case mix index, an indicator of severity of illness.

Pharmacologic VTE prophylaxis included all chemical prophylaxis specified on our institutional guideline, most commonly subcutaneous unfractionated heparin 5000 units every 8 hours or low molecular weight heparin (LMWH), either enoxaparin 40 mg every 12 or 24 hours or dalteparin 5000 units every 24 hours. Mechanical prophylaxis was defined as use of sequential compression devices (SCDs) when pharmacologic prophylaxis was contraindicated. Prophylaxis was considered to be appropriate if it was applied according to our guideline for >90% of hospital days prior to UEDVT diagnosis. Therapeutic anticoagulation included heparin bridging (most commonly continuous heparin infusion, LMWH 1 mg/kg or dalteparin) as well as oral vitamin K antagonists. The VTE tool (Figure 1) allows identification of pharmacologic prophylaxis and therapy that is actually administered (not just ordered) directly from our pharmacy IT system. SCD application (not just ordered SCDs) is electronically integrated into the tool from nursing documentation.

CVCs included internal jugular or subclavian triple lumen catheters, tunneled dialysis catheters, or peripherally inserted central catheters (PICCs), single or double lumen. Criteria used to identify that a UEDVT was CVC‐associated included temporal relationship (CVC was placed prior to clot diagnosis), plausibility (ipsilateral clot), evidence of clot surrounding CVC on ultrasound, and physician designation of association (as documented in progress notes or discharge summary).

Simple percentages of patient characteristics, associated factors, management, and outcomes were calculated using counts as the numerator and number of patients as the denominator. For information about UEDVTs, we used total number of UEDVTs as the denominator. Line days were day counts from insertion until removal if applicable. The CVC placement date was available in our mandated central line placement procedure notes (directly accessed from the VTE tool) for all lines placed at our institution; date of removal (if applicable) was determined from chart abstraction. For the vast majority of patients whose CVCs were placed at outside facilities, date of placement was available in the EHR (often in the admission note or in the ultrasound report/reason for study). If date of line placement at an outside facility was not known, date of admission was used. The University of Washington Human Subjects Board approved this review.

RESULTS