Association Between DCBN and LOS

Site and Subjects

Our study was based at the University of California, San Francisco (UCSF) Medical Center, a 400‐bed academic hospital located in San Francisco, California. We examined adult medical and surgical discharges from July 2012 through April 2015. Patients who stayed less than 24 hours or more than 20 days were excluded. Discharges from the hospital medicine service and the following surgical services were included in the analysis: cardiac surgery, colorectal surgery, cardiothoracic surgery, general surgery, gynecologic oncology, gynecology, neurosurgery, orthopedics, otolaryngology, head and neck surgery, plastic surgery, thoracic surgery, urology, and vascular surgery. No exclusions were made based on patient status (eg, observation vs inpatient). UCSF's institutional review board approved our study.

During the time of our study, discharges before noon time became an institutional priority. To this end, rates of DCBN were tracked using retrospective data, and various units undertook efforts such as informal afternoon meetings to prompt planning for the next morning's discharges. These efforts did not differentially affect medical or surgical units or emergent or nonemergent admissions, and no financial incentives or other changes in workflow were in place to increase DCBN rates.

Data Sources

We used the cost accounting system at UCSF (Enterprise Performance System Inc. [EPSI], Chicago, IL) to collect demographic information about each patient, including age, sex, primary race, and primary ethnicity. This system was also used to collect characteristics of each hospitalization including LOS (calculated from admission date time and discharge date time), hospital service at discharge, the discharge attending, discharge disposition of the patient, and the CMI, a marker of the severity of illness of the patient during that hospitalization. EPSI was also used to collect data on the admission type of all patients, either emergent, urgent, or routine, and the insurance status of the patient during that hospitalization.

Data on time of discharge were entered by the discharging nurse or unit assistant to reflect the time the patient left the hospital. Using these data, we defined a before‐noon discharge as one taking place between 8:00 _am and 12:00 _pm.

Statistical Analysis

Wilcoxon rank sum test and ² statistics were used to compare baseline characteristics of hospitalizations of patients discharged before and after noon.

We used generalized linear models to assess the association of a discharge before noon on the LOS with gamma models. We accounted for clustering of discharge attendings using generalized estimating equations with exchangeable working correlation and robust standard errors. After the initial unadjusted analyses, covariates were included in the adjusted analysis if they were associated with an LOS at P < 0.05 or reasons of face validity. These variables are shown in Table 1. Because an effort to increase the discharges before noon was started in the 2014 academic year, we added an interaction term between the date of discharge and whether a discharge occurred before noon. The interaction term was included by dividing the study period into time periods corresponding to sequential 6‐month intervals. A new variable was defined by a categorical variable that indicated in which of these time periods a discharge occurred.

We conducted a sensitivity analysis using propensity scores. The propensity score was based on demographic and clinical variables (as listed in Table 1) that exhibited P < 0.2 in bivariate analysis between the variable and being discharged before noon. We then used the propensity score as a covariate in a generalized linear model of the LOS with a gamma distribution and with generalized estimating equations as described above.

Finally, we performed prespecified secondary subset analyses of patients admitted emergently and nonemergently.

Statistical modeling and analysis was completed using Stata version 13 (StataCorp, College Station, TX).

Patient Demographics and Discharge Before Noon

Our study population comprised 27,983 patients for a total of 38,365 hospitalizations with a median LOS of 3.7 days. We observed 6484 discharges before noon (16.9%) and 31,881 discharges after noon (83.1%). The characteristics of the hospitalizations are shown in Table 1.

Patients who were discharged before noon tended to be younger, white, and discharged with a disposition to home without home health. The median CMI was slightly higher in discharges before noon (1.81, P = 0.006), and elective admissions were more likely than emergent to be discharged before noon (53.9% vs 31.6%, P < 0.0005).

Multivariable Analysis

A discharge before noon was associated with a 4.3% increase in LOS (adjusted odds ratio [OR]: 1.043, 95% confidence interval [CI]: 1.003‐1.086), adjusting for CMI, the service type, discharge on the weekend, discharge disposition, age, sex, ethnicity, race, urgency of admission, payor class, and a full interaction with the date of discharge (in 6‐month intervals). In preplanned subset analyses, the association between longer LOS and DCBN was more pronounced in patients admitted emergently (adjusted OR: 1.14, 95% CI: 1.033‐1.249) and less pronounced for patients not admitted emergently (adjusted OR: 1.03, 95% CI: 0.988‐1.074), although the latter did not meet statistical significance. In patients admitted emergently, this corresponds to approximately a 12‐hour increase in LOS. The interaction term of discharge date and DCBN was significant in the model. In further subset analyses, the association between longer LOS and DCBN was more pronounced in medicine patients (adjusted OR: 1.116, 95% CI: 1.014‐1.228) than in surgical patients (adjusted OR: 1.030, 95% CI: 0.989‐1.074), although the relationship in surgical patients did not meet statistical significance.

We also undertook sensitivity analyses utilizing propensity scores as a covariate in our base multivariable models. Results from these analyses did not differ from the base models and are not presented here. Results also did not differ when comparing discharges before and after the initiation of an attending only service.

Site and Subjects

Our study was based at the University of California, San Francisco (UCSF) Medical Center, a 400‐bed academic hospital located in San Francisco, California. We examined adult medical and surgical discharges from July 2012 through April 2015. Patients who stayed less than 24 hours or more than 20 days were excluded. Discharges from the hospital medicine service and the following surgical services were included in the analysis: cardiac surgery, colorectal surgery, cardiothoracic surgery, general surgery, gynecologic oncology, gynecology, neurosurgery, orthopedics, otolaryngology, head and neck surgery, plastic surgery, thoracic surgery, urology, and vascular surgery. No exclusions were made based on patient status (eg, observation vs inpatient). UCSF's institutional review board approved our study.

During the time of our study, discharges before noon time became an institutional priority. To this end, rates of DCBN were tracked using retrospective data, and various units undertook efforts such as informal afternoon meetings to prompt planning for the next morning's discharges. These efforts did not differentially affect medical or surgical units or emergent or nonemergent admissions, and no financial incentives or other changes in workflow were in place to increase DCBN rates.

Data Sources

We used the cost accounting system at UCSF (Enterprise Performance System Inc. [EPSI], Chicago, IL) to collect demographic information about each patient, including age, sex, primary race, and primary ethnicity. This system was also used to collect characteristics of each hospitalization including LOS (calculated from admission date time and discharge date time), hospital service at discharge, the discharge attending, discharge disposition of the patient, and the CMI, a marker of the severity of illness of the patient during that hospitalization. EPSI was also used to collect data on the admission type of all patients, either emergent, urgent, or routine, and the insurance status of the patient during that hospitalization.

Data on time of discharge were entered by the discharging nurse or unit assistant to reflect the time the patient left the hospital. Using these data, we defined a before‐noon discharge as one taking place between 8:00 _am and 12:00 _pm.

Statistical Analysis

Wilcoxon rank sum test and ² statistics were used to compare baseline characteristics of hospitalizations of patients discharged before and after noon.

We used generalized linear models to assess the association of a discharge before noon on the LOS with gamma models. We accounted for clustering of discharge attendings using generalized estimating equations with exchangeable working correlation and robust standard errors. After the initial unadjusted analyses, covariates were included in the adjusted analysis if they were associated with an LOS at P < 0.05 or reasons of face validity. These variables are shown in Table 1. Because an effort to increase the discharges before noon was started in the 2014 academic year, we added an interaction term between the date of discharge and whether a discharge occurred before noon. The interaction term was included by dividing the study period into time periods corresponding to sequential 6‐month intervals. A new variable was defined by a categorical variable that indicated in which of these time periods a discharge occurred.

We conducted a sensitivity analysis using propensity scores. The propensity score was based on demographic and clinical variables (as listed in Table 1) that exhibited P < 0.2 in bivariate analysis between the variable and being discharged before noon. We then used the propensity score as a covariate in a generalized linear model of the LOS with a gamma distribution and with generalized estimating equations as described above.

Finally, we performed prespecified secondary subset analyses of patients admitted emergently and nonemergently.

Statistical modeling and analysis was completed using Stata version 13 (StataCorp, College Station, TX).

Patient Demographics and Discharge Before Noon

Our study population comprised 27,983 patients for a total of 38,365 hospitalizations with a median LOS of 3.7 days. We observed 6484 discharges before noon (16.9%) and 31,881 discharges after noon (83.1%). The characteristics of the hospitalizations are shown in Table 1.

Patients who were discharged before noon tended to be younger, white, and discharged with a disposition to home without home health. The median CMI was slightly higher in discharges before noon (1.81, P = 0.006), and elective admissions were more likely than emergent to be discharged before noon (53.9% vs 31.6%, P < 0.0005).

Multivariable Analysis

A discharge before noon was associated with a 4.3% increase in LOS (adjusted odds ratio [OR]: 1.043, 95% confidence interval [CI]: 1.003‐1.086), adjusting for CMI, the service type, discharge on the weekend, discharge disposition, age, sex, ethnicity, race, urgency of admission, payor class, and a full interaction with the date of discharge (in 6‐month intervals). In preplanned subset analyses, the association between longer LOS and DCBN was more pronounced in patients admitted emergently (adjusted OR: 1.14, 95% CI: 1.033‐1.249) and less pronounced for patients not admitted emergently (adjusted OR: 1.03, 95% CI: 0.988‐1.074), although the latter did not meet statistical significance. In patients admitted emergently, this corresponds to approximately a 12‐hour increase in LOS. The interaction term of discharge date and DCBN was significant in the model. In further subset analyses, the association between longer LOS and DCBN was more pronounced in medicine patients (adjusted OR: 1.116, 95% CI: 1.014‐1.228) than in surgical patients (adjusted OR: 1.030, 95% CI: 0.989‐1.074), although the relationship in surgical patients did not meet statistical significance.

We also undertook sensitivity analyses utilizing propensity scores as a covariate in our base multivariable models. Results from these analyses did not differ from the base models and are not presented here. Results also did not differ when comparing discharges before and after the initiation of an attending only service.

Site and Subjects

Our study was based at the University of California, San Francisco (UCSF) Medical Center, a 400‐bed academic hospital located in San Francisco, California. We examined adult medical and surgical discharges from July 2012 through April 2015. Patients who stayed less than 24 hours or more than 20 days were excluded. Discharges from the hospital medicine service and the following surgical services were included in the analysis: cardiac surgery, colorectal surgery, cardiothoracic surgery, general surgery, gynecologic oncology, gynecology, neurosurgery, orthopedics, otolaryngology, head and neck surgery, plastic surgery, thoracic surgery, urology, and vascular surgery. No exclusions were made based on patient status (eg, observation vs inpatient). UCSF's institutional review board approved our study.

During the time of our study, discharges before noon time became an institutional priority. To this end, rates of DCBN were tracked using retrospective data, and various units undertook efforts such as informal afternoon meetings to prompt planning for the next morning's discharges. These efforts did not differentially affect medical or surgical units or emergent or nonemergent admissions, and no financial incentives or other changes in workflow were in place to increase DCBN rates.

Data Sources

We used the cost accounting system at UCSF (Enterprise Performance System Inc. [EPSI], Chicago, IL) to collect demographic information about each patient, including age, sex, primary race, and primary ethnicity. This system was also used to collect characteristics of each hospitalization including LOS (calculated from admission date time and discharge date time), hospital service at discharge, the discharge attending, discharge disposition of the patient, and the CMI, a marker of the severity of illness of the patient during that hospitalization. EPSI was also used to collect data on the admission type of all patients, either emergent, urgent, or routine, and the insurance status of the patient during that hospitalization.

Data on time of discharge were entered by the discharging nurse or unit assistant to reflect the time the patient left the hospital. Using these data, we defined a before‐noon discharge as one taking place between 8:00 _am and 12:00 _pm.

Statistical Analysis

Wilcoxon rank sum test and ² statistics were used to compare baseline characteristics of hospitalizations of patients discharged before and after noon.

We used generalized linear models to assess the association of a discharge before noon on the LOS with gamma models. We accounted for clustering of discharge attendings using generalized estimating equations with exchangeable working correlation and robust standard errors. After the initial unadjusted analyses, covariates were included in the adjusted analysis if they were associated with an LOS at P < 0.05 or reasons of face validity. These variables are shown in Table 1. Because an effort to increase the discharges before noon was started in the 2014 academic year, we added an interaction term between the date of discharge and whether a discharge occurred before noon. The interaction term was included by dividing the study period into time periods corresponding to sequential 6‐month intervals. A new variable was defined by a categorical variable that indicated in which of these time periods a discharge occurred.

We conducted a sensitivity analysis using propensity scores. The propensity score was based on demographic and clinical variables (as listed in Table 1) that exhibited P < 0.2 in bivariate analysis between the variable and being discharged before noon. We then used the propensity score as a covariate in a generalized linear model of the LOS with a gamma distribution and with generalized estimating equations as described above.

Finally, we performed prespecified secondary subset analyses of patients admitted emergently and nonemergently.

Statistical modeling and analysis was completed using Stata version 13 (StataCorp, College Station, TX).

Patient Demographics and Discharge Before Noon

Our study population comprised 27,983 patients for a total of 38,365 hospitalizations with a median LOS of 3.7 days. We observed 6484 discharges before noon (16.9%) and 31,881 discharges after noon (83.1%). The characteristics of the hospitalizations are shown in Table 1.

Patients who were discharged before noon tended to be younger, white, and discharged with a disposition to home without home health. The median CMI was slightly higher in discharges before noon (1.81, P = 0.006), and elective admissions were more likely than emergent to be discharged before noon (53.9% vs 31.6%, P < 0.0005).

Multivariable Analysis

A discharge before noon was associated with a 4.3% increase in LOS (adjusted odds ratio [OR]: 1.043, 95% confidence interval [CI]: 1.003‐1.086), adjusting for CMI, the service type, discharge on the weekend, discharge disposition, age, sex, ethnicity, race, urgency of admission, payor class, and a full interaction with the date of discharge (in 6‐month intervals). In preplanned subset analyses, the association between longer LOS and DCBN was more pronounced in patients admitted emergently (adjusted OR: 1.14, 95% CI: 1.033‐1.249) and less pronounced for patients not admitted emergently (adjusted OR: 1.03, 95% CI: 0.988‐1.074), although the latter did not meet statistical significance. In patients admitted emergently, this corresponds to approximately a 12‐hour increase in LOS. The interaction term of discharge date and DCBN was significant in the model. In further subset analyses, the association between longer LOS and DCBN was more pronounced in medicine patients (adjusted OR: 1.116, 95% CI: 1.014‐1.228) than in surgical patients (adjusted OR: 1.030, 95% CI: 0.989‐1.074), although the relationship in surgical patients did not meet statistical significance.

We also undertook sensitivity analyses utilizing propensity scores as a covariate in our base multivariable models. Results from these analyses did not differ from the base models and are not presented here. Results also did not differ when comparing discharges before and after the initiation of an attending only service.