Erik Youngson, MMath

Study Setting

The Canadian province of Alberta has a single vertically integrated healthcare system that is government‐funded and provides universal access to hospitals, emergency departments (EDs), and outpatient physician services for all 4.1 million Albertans as well as all prescription medications for the poor, socially disadvantaged, disabled, or those age 65 years and older. This study received approval from the University of Alberta Health Research Ethics Board with waiver of informed consent.

Data Sources

This study used deidentified linked data from 3 Alberta Health administrative databases that capture vital status and all hospital or ED visits and have previously been shown to have high accuracy for medical diagnoses.[13] The Alberta Health Care Insurance Plan Registry tracks date of death or emigration from the province. The Discharge Abstract Database includes the most responsible diagnosis identified by the hospital attending physician, up to 25 other diagnoses coded by nosologists in each hospital, the admission and discharge dates, and the admission category (elective or urgent/emergent) for all acute care hospitalizations. Of note, unlike US studies, the hospital databases are able to distinguish in‐hospital (eg, adverse events) versus premorbid diagnoses (eg, preexisting comorbidities). The Ambulatory Care Database captures all patient visits to EDs with coding for up to 10 conditions per encounter.

Study Cohort

We identified all adults with an acute care hospitalization on the GIM services at all 7 Alberta teaching hospitals (ie, defined as those with Royal College of Physicians and Surgeons of Canadaapproved residency training programs in internal medicine, the equivalent of the Association of American Medical Colleges certification in the United States) between October 1, 2009 and September 30, 2010 and between April 1, 2011 and December 1, 2011 (these 20 months covered most of the pre/post intervals for a recently reported quality improvement initiative at 1 of the teaching hospitals that had no significant impact on postdischarge outcomes).[14] Patients from out of the province or transferred from/to another inpatient service (eg, the intensive care unit, a different service in the same hospital [such as surgery], another acute care hospital, or rehabilitation hospital) or with lengths of stay greater than 30 days were excluded. We only included the first hospitalization for any patient in our study timeframe and thus excluded repeat discharges of the same patient.

Explanatory Variable of Interest

The independent variable of interest was calendar day of discharge, stratified according to weekday (Monday thru Friday) versus weekend (Saturday and Sunday). Only 1.4% of weekday discharges occurred on a statutory holiday, and for the purposes of this study, these discharges were also considered weekend discharges. At the 7 teaching hospitals in Alberta, nursing staffing ratios do not differ between weekend and weekday, but availability of all other members of the healthcare team does. Physician census decreases from 4 to 5 per ward to 1 to 2, and ward‐based social workers, occupational therapists, physiotherapists, and pharmacist educators are generally not available on weekends.

Outcomes

Our primary outcome of interest was the composite outcome of death or all‐cause nonelective readmission within 30 days of discharge (ie, not including in‐hospital events prior to discharge or elective readmissions after discharge for planned procedures such as chemotherapy); hereafter we refer to this as death or readmission. This is a patient‐relevant outcome that is highlighted in the Affordable Care Act and for which there are several validated risk adjustment models.[15] We chose a composite outcome to deal with the issue of competing risks; if weekend discharges were more likely to die then we could observe a spurious association between weekend discharge and reduced readmissions if we focused on only that outcome.

Other Measures

Comorbidities for each patient were identified using International Classification of Diseases, Ninth Revision and Tenth Revision codes from the Discharge Abstract Database for the index hospitalization and any hospitalizations in the 12 months prior to their index admission, a method previously validated in Alberta databases.[13] We also recorded health resource use during their index hospitalization and calculated each patient's LACE score at the time of discharge, which is an index for predicting unplanned readmission or early death postdischarge previously validated in Canadian administrative databases.[15] The LACE index includes length of hospital stay (L), acuity of admission (A, based on the admission category variable described earlier), comorbidity burden quantified using the Charlson Comorbidity Index (C), and emergency department visits in the 6 months prior to admission (E); patients with discharge LACE scores >10 (total possible score is 19) are defined as being at high risk of death/readmission within 30 days.[16] As detailed below, to deal with potential concerns that LOS may be a mediator in the causal pathway, we ran 2 sensitivity analyses, 1 in which we excluded LOS from the analyses and 1 in which we included expected LOS rather than the actual LOS. Expected LOS is a data‐driven estimate based on the most current 2 years of patient LOS information available in the Canadian Institute for Health Information discharge abstract database (www.cihi.ca) for all acute care hospitals in Canada, and was generated for each patient independently of our study taking into account case mix group, age, and inpatient resource intensity weights.

Statistical Analysis

Baseline patient characteristics between weekend and weekday discharges were compared with t tests for continuous variables and [2] tests for binary or categorical variables. Logistic regression was used for comparison of death or readmission for weekend versus weekday discharges. Multivariable models were adjusted for age, sex, hospital, and LACE scores (as a continuous variable) at time of discharge; in sensitivity analyses we adjusted for (1) LACE score without including LOS and (2) LACE score using expected LOS rather than actual LOS. In further sensitivity analyses we (1) restricted the analysis to only those patients deemed to be at high risk for events due to LACE scores of 10 or greater and (2) included ED visits as part of the composite endpoint (ie, death, unplanned readmission, or unplanned ED visit within 30 days of discharge). Day of admission (weekend vs weekday) was also considered for the multivariable models, but was not found to be significant and thus was omitted from final models. We do not have any physician identifying variables in our dataset and thus could not investigate the potential correlation among patients discharged by the same physician. We did explore the hospital intraclass correlation coefficient, and as it was very small (0.001), we did not utilize models to account for the hierarchical nature of the data, but did include hospital as a fixed effect in the logistic models. The results were virtually identical whether we did or did not include hospital in the models. Adjusted odds ratios (aORs) are displayed with 95% confidence intervals (CI) and P values. Average LOS was calculated for weekend and weekday discharges with 95% CIs. P values for adjusted length of stay were calculated using multivariable linear regression adjusting for age, sex, day of admission, and Charlson score. All statistical analyses were done using SAS for Windows version 9.4 (SAS Institute, Inc., Cary, NC).

Patient Characteristics

Of the 7991 patients discharged during our study interval, 1146 (14.3%) were discharged on weekend or holiday days (Table 1). In contrast, 2180 of our cohort were admitted on a weekend (27.3%). The mean age of our study population was 62.1 years, 51.9% were men, mean Charlson score was 2.56, and 4591 (57.5%) had LACE scores of at least 10 at discharge.

Weekday Versus Weekend Discharge

Although patients admitted on weekdays and weekends were very similar (data available upon request), patients discharged on weekends (compared to those discharged on weekdays) were younger, more likely to be discharged home without additional support, and had fewer comorbidities (Table 1, Figure 1). Patients discharged on weekends had shorter lengths of stay than those discharged on weekdays (5.6 days vs 7.9 days, P<0.0001). In adjusted linear regression analyses, this 2.3‐day difference remained statistically significant (adjusted P value <0.0001).

Figure 1

Patients discharged on a weekend exhibited lower unadjusted 30‐day rates of death or readmission than those discharged on a weekday (10.6% vs 13.2%), but these differences disappeared after multivariable adjustment that accounted for differences in risk profile (aOR: 0.94, 95% CI: 0.771.16 (Table 2). Results were similar in sensitivity analyses adjusting for LACE scores without LOS included (aOR: 0.88, 95% CI: 0.711.08) or adjusting for LACE scores using expected LOS rather than actual LOS (aOR: 0.90, 95% CI: 0.731.10). Restricting the analysis to only those patients deemed to be at high risk for events due to LACE scores of 10 or greater confirmed that weekend and weekday discharges had similar outcomes in the first 30 days after discharge (aOR: 1.09, 95% CI: 0.851.41, Table 2). Similar patterns were seen when we included ED visits as part of the composite endpoint (ie, death, unplanned readmission, or unplanned ED visit within 30 days of discharge) (Table 2).

Study Setting

The Canadian province of Alberta has a single vertically integrated healthcare system that is government‐funded and provides universal access to hospitals, emergency departments (EDs), and outpatient physician services for all 4.1 million Albertans as well as all prescription medications for the poor, socially disadvantaged, disabled, or those age 65 years and older. This study received approval from the University of Alberta Health Research Ethics Board with waiver of informed consent.

Data Sources

This study used deidentified linked data from 3 Alberta Health administrative databases that capture vital status and all hospital or ED visits and have previously been shown to have high accuracy for medical diagnoses.[13] The Alberta Health Care Insurance Plan Registry tracks date of death or emigration from the province. The Discharge Abstract Database includes the most responsible diagnosis identified by the hospital attending physician, up to 25 other diagnoses coded by nosologists in each hospital, the admission and discharge dates, and the admission category (elective or urgent/emergent) for all acute care hospitalizations. Of note, unlike US studies, the hospital databases are able to distinguish in‐hospital (eg, adverse events) versus premorbid diagnoses (eg, preexisting comorbidities). The Ambulatory Care Database captures all patient visits to EDs with coding for up to 10 conditions per encounter.

Study Cohort

We identified all adults with an acute care hospitalization on the GIM services at all 7 Alberta teaching hospitals (ie, defined as those with Royal College of Physicians and Surgeons of Canadaapproved residency training programs in internal medicine, the equivalent of the Association of American Medical Colleges certification in the United States) between October 1, 2009 and September 30, 2010 and between April 1, 2011 and December 1, 2011 (these 20 months covered most of the pre/post intervals for a recently reported quality improvement initiative at 1 of the teaching hospitals that had no significant impact on postdischarge outcomes).[14] Patients from out of the province or transferred from/to another inpatient service (eg, the intensive care unit, a different service in the same hospital [such as surgery], another acute care hospital, or rehabilitation hospital) or with lengths of stay greater than 30 days were excluded. We only included the first hospitalization for any patient in our study timeframe and thus excluded repeat discharges of the same patient.

Explanatory Variable of Interest

The independent variable of interest was calendar day of discharge, stratified according to weekday (Monday thru Friday) versus weekend (Saturday and Sunday). Only 1.4% of weekday discharges occurred on a statutory holiday, and for the purposes of this study, these discharges were also considered weekend discharges. At the 7 teaching hospitals in Alberta, nursing staffing ratios do not differ between weekend and weekday, but availability of all other members of the healthcare team does. Physician census decreases from 4 to 5 per ward to 1 to 2, and ward‐based social workers, occupational therapists, physiotherapists, and pharmacist educators are generally not available on weekends.

Outcomes

Our primary outcome of interest was the composite outcome of death or all‐cause nonelective readmission within 30 days of discharge (ie, not including in‐hospital events prior to discharge or elective readmissions after discharge for planned procedures such as chemotherapy); hereafter we refer to this as death or readmission. This is a patient‐relevant outcome that is highlighted in the Affordable Care Act and for which there are several validated risk adjustment models.[15] We chose a composite outcome to deal with the issue of competing risks; if weekend discharges were more likely to die then we could observe a spurious association between weekend discharge and reduced readmissions if we focused on only that outcome.

Other Measures

Comorbidities for each patient were identified using International Classification of Diseases, Ninth Revision and Tenth Revision codes from the Discharge Abstract Database for the index hospitalization and any hospitalizations in the 12 months prior to their index admission, a method previously validated in Alberta databases.[13] We also recorded health resource use during their index hospitalization and calculated each patient's LACE score at the time of discharge, which is an index for predicting unplanned readmission or early death postdischarge previously validated in Canadian administrative databases.[15] The LACE index includes length of hospital stay (L), acuity of admission (A, based on the admission category variable described earlier), comorbidity burden quantified using the Charlson Comorbidity Index (C), and emergency department visits in the 6 months prior to admission (E); patients with discharge LACE scores >10 (total possible score is 19) are defined as being at high risk of death/readmission within 30 days.[16] As detailed below, to deal with potential concerns that LOS may be a mediator in the causal pathway, we ran 2 sensitivity analyses, 1 in which we excluded LOS from the analyses and 1 in which we included expected LOS rather than the actual LOS. Expected LOS is a data‐driven estimate based on the most current 2 years of patient LOS information available in the Canadian Institute for Health Information discharge abstract database (www.cihi.ca) for all acute care hospitals in Canada, and was generated for each patient independently of our study taking into account case mix group, age, and inpatient resource intensity weights.

Statistical Analysis

Baseline patient characteristics between weekend and weekday discharges were compared with t tests for continuous variables and [2] tests for binary or categorical variables. Logistic regression was used for comparison of death or readmission for weekend versus weekday discharges. Multivariable models were adjusted for age, sex, hospital, and LACE scores (as a continuous variable) at time of discharge; in sensitivity analyses we adjusted for (1) LACE score without including LOS and (2) LACE score using expected LOS rather than actual LOS. In further sensitivity analyses we (1) restricted the analysis to only those patients deemed to be at high risk for events due to LACE scores of 10 or greater and (2) included ED visits as part of the composite endpoint (ie, death, unplanned readmission, or unplanned ED visit within 30 days of discharge). Day of admission (weekend vs weekday) was also considered for the multivariable models, but was not found to be significant and thus was omitted from final models. We do not have any physician identifying variables in our dataset and thus could not investigate the potential correlation among patients discharged by the same physician. We did explore the hospital intraclass correlation coefficient, and as it was very small (0.001), we did not utilize models to account for the hierarchical nature of the data, but did include hospital as a fixed effect in the logistic models. The results were virtually identical whether we did or did not include hospital in the models. Adjusted odds ratios (aORs) are displayed with 95% confidence intervals (CI) and P values. Average LOS was calculated for weekend and weekday discharges with 95% CIs. P values for adjusted length of stay were calculated using multivariable linear regression adjusting for age, sex, day of admission, and Charlson score. All statistical analyses were done using SAS for Windows version 9.4 (SAS Institute, Inc., Cary, NC).

Patient Characteristics

Of the 7991 patients discharged during our study interval, 1146 (14.3%) were discharged on weekend or holiday days (Table 1). In contrast, 2180 of our cohort were admitted on a weekend (27.3%). The mean age of our study population was 62.1 years, 51.9% were men, mean Charlson score was 2.56, and 4591 (57.5%) had LACE scores of at least 10 at discharge.

Weekday Versus Weekend Discharge

Although patients admitted on weekdays and weekends were very similar (data available upon request), patients discharged on weekends (compared to those discharged on weekdays) were younger, more likely to be discharged home without additional support, and had fewer comorbidities (Table 1, Figure 1). Patients discharged on weekends had shorter lengths of stay than those discharged on weekdays (5.6 days vs 7.9 days, P<0.0001). In adjusted linear regression analyses, this 2.3‐day difference remained statistically significant (adjusted P value <0.0001).

Figure 1

Patients discharged on a weekend exhibited lower unadjusted 30‐day rates of death or readmission than those discharged on a weekday (10.6% vs 13.2%), but these differences disappeared after multivariable adjustment that accounted for differences in risk profile (aOR: 0.94, 95% CI: 0.771.16 (Table 2). Results were similar in sensitivity analyses adjusting for LACE scores without LOS included (aOR: 0.88, 95% CI: 0.711.08) or adjusting for LACE scores using expected LOS rather than actual LOS (aOR: 0.90, 95% CI: 0.731.10). Restricting the analysis to only those patients deemed to be at high risk for events due to LACE scores of 10 or greater confirmed that weekend and weekday discharges had similar outcomes in the first 30 days after discharge (aOR: 1.09, 95% CI: 0.851.41, Table 2). Similar patterns were seen when we included ED visits as part of the composite endpoint (ie, death, unplanned readmission, or unplanned ED visit within 30 days of discharge) (Table 2).

Study Setting

The Canadian province of Alberta has a single vertically integrated healthcare system that is government‐funded and provides universal access to hospitals, emergency departments (EDs), and outpatient physician services for all 4.1 million Albertans as well as all prescription medications for the poor, socially disadvantaged, disabled, or those age 65 years and older. This study received approval from the University of Alberta Health Research Ethics Board with waiver of informed consent.

Data Sources

This study used deidentified linked data from 3 Alberta Health administrative databases that capture vital status and all hospital or ED visits and have previously been shown to have high accuracy for medical diagnoses.[13] The Alberta Health Care Insurance Plan Registry tracks date of death or emigration from the province. The Discharge Abstract Database includes the most responsible diagnosis identified by the hospital attending physician, up to 25 other diagnoses coded by nosologists in each hospital, the admission and discharge dates, and the admission category (elective or urgent/emergent) for all acute care hospitalizations. Of note, unlike US studies, the hospital databases are able to distinguish in‐hospital (eg, adverse events) versus premorbid diagnoses (eg, preexisting comorbidities). The Ambulatory Care Database captures all patient visits to EDs with coding for up to 10 conditions per encounter.

Study Cohort

We identified all adults with an acute care hospitalization on the GIM services at all 7 Alberta teaching hospitals (ie, defined as those with Royal College of Physicians and Surgeons of Canadaapproved residency training programs in internal medicine, the equivalent of the Association of American Medical Colleges certification in the United States) between October 1, 2009 and September 30, 2010 and between April 1, 2011 and December 1, 2011 (these 20 months covered most of the pre/post intervals for a recently reported quality improvement initiative at 1 of the teaching hospitals that had no significant impact on postdischarge outcomes).[14] Patients from out of the province or transferred from/to another inpatient service (eg, the intensive care unit, a different service in the same hospital [such as surgery], another acute care hospital, or rehabilitation hospital) or with lengths of stay greater than 30 days were excluded. We only included the first hospitalization for any patient in our study timeframe and thus excluded repeat discharges of the same patient.

Explanatory Variable of Interest

The independent variable of interest was calendar day of discharge, stratified according to weekday (Monday thru Friday) versus weekend (Saturday and Sunday). Only 1.4% of weekday discharges occurred on a statutory holiday, and for the purposes of this study, these discharges were also considered weekend discharges. At the 7 teaching hospitals in Alberta, nursing staffing ratios do not differ between weekend and weekday, but availability of all other members of the healthcare team does. Physician census decreases from 4 to 5 per ward to 1 to 2, and ward‐based social workers, occupational therapists, physiotherapists, and pharmacist educators are generally not available on weekends.

Outcomes

Our primary outcome of interest was the composite outcome of death or all‐cause nonelective readmission within 30 days of discharge (ie, not including in‐hospital events prior to discharge or elective readmissions after discharge for planned procedures such as chemotherapy); hereafter we refer to this as death or readmission. This is a patient‐relevant outcome that is highlighted in the Affordable Care Act and for which there are several validated risk adjustment models.[15] We chose a composite outcome to deal with the issue of competing risks; if weekend discharges were more likely to die then we could observe a spurious association between weekend discharge and reduced readmissions if we focused on only that outcome.

Other Measures

Comorbidities for each patient were identified using International Classification of Diseases, Ninth Revision and Tenth Revision codes from the Discharge Abstract Database for the index hospitalization and any hospitalizations in the 12 months prior to their index admission, a method previously validated in Alberta databases.[13] We also recorded health resource use during their index hospitalization and calculated each patient's LACE score at the time of discharge, which is an index for predicting unplanned readmission or early death postdischarge previously validated in Canadian administrative databases.[15] The LACE index includes length of hospital stay (L), acuity of admission (A, based on the admission category variable described earlier), comorbidity burden quantified using the Charlson Comorbidity Index (C), and emergency department visits in the 6 months prior to admission (E); patients with discharge LACE scores >10 (total possible score is 19) are defined as being at high risk of death/readmission within 30 days.[16] As detailed below, to deal with potential concerns that LOS may be a mediator in the causal pathway, we ran 2 sensitivity analyses, 1 in which we excluded LOS from the analyses and 1 in which we included expected LOS rather than the actual LOS. Expected LOS is a data‐driven estimate based on the most current 2 years of patient LOS information available in the Canadian Institute for Health Information discharge abstract database (www.cihi.ca) for all acute care hospitals in Canada, and was generated for each patient independently of our study taking into account case mix group, age, and inpatient resource intensity weights.

Statistical Analysis

Baseline patient characteristics between weekend and weekday discharges were compared with t tests for continuous variables and [2] tests for binary or categorical variables. Logistic regression was used for comparison of death or readmission for weekend versus weekday discharges. Multivariable models were adjusted for age, sex, hospital, and LACE scores (as a continuous variable) at time of discharge; in sensitivity analyses we adjusted for (1) LACE score without including LOS and (2) LACE score using expected LOS rather than actual LOS. In further sensitivity analyses we (1) restricted the analysis to only those patients deemed to be at high risk for events due to LACE scores of 10 or greater and (2) included ED visits as part of the composite endpoint (ie, death, unplanned readmission, or unplanned ED visit within 30 days of discharge). Day of admission (weekend vs weekday) was also considered for the multivariable models, but was not found to be significant and thus was omitted from final models. We do not have any physician identifying variables in our dataset and thus could not investigate the potential correlation among patients discharged by the same physician. We did explore the hospital intraclass correlation coefficient, and as it was very small (0.001), we did not utilize models to account for the hierarchical nature of the data, but did include hospital as a fixed effect in the logistic models. The results were virtually identical whether we did or did not include hospital in the models. Adjusted odds ratios (aORs) are displayed with 95% confidence intervals (CI) and P values. Average LOS was calculated for weekend and weekday discharges with 95% CIs. P values for adjusted length of stay were calculated using multivariable linear regression adjusting for age, sex, day of admission, and Charlson score. All statistical analyses were done using SAS for Windows version 9.4 (SAS Institute, Inc., Cary, NC).

Patient Characteristics

Of the 7991 patients discharged during our study interval, 1146 (14.3%) were discharged on weekend or holiday days (Table 1). In contrast, 2180 of our cohort were admitted on a weekend (27.3%). The mean age of our study population was 62.1 years, 51.9% were men, mean Charlson score was 2.56, and 4591 (57.5%) had LACE scores of at least 10 at discharge.

Weekday Versus Weekend Discharge

Although patients admitted on weekdays and weekends were very similar (data available upon request), patients discharged on weekends (compared to those discharged on weekdays) were younger, more likely to be discharged home without additional support, and had fewer comorbidities (Table 1, Figure 1). Patients discharged on weekends had shorter lengths of stay than those discharged on weekdays (5.6 days vs 7.9 days, P<0.0001). In adjusted linear regression analyses, this 2.3‐day difference remained statistically significant (adjusted P value <0.0001).

Figure 1

Patients discharged on a weekend exhibited lower unadjusted 30‐day rates of death or readmission than those discharged on a weekday (10.6% vs 13.2%), but these differences disappeared after multivariable adjustment that accounted for differences in risk profile (aOR: 0.94, 95% CI: 0.771.16 (Table 2). Results were similar in sensitivity analyses adjusting for LACE scores without LOS included (aOR: 0.88, 95% CI: 0.711.08) or adjusting for LACE scores using expected LOS rather than actual LOS (aOR: 0.90, 95% CI: 0.731.10). Restricting the analysis to only those patients deemed to be at high risk for events due to LACE scores of 10 or greater confirmed that weekend and weekday discharges had similar outcomes in the first 30 days after discharge (aOR: 1.09, 95% CI: 0.851.41, Table 2). Similar patterns were seen when we included ED visits as part of the composite endpoint (ie, death, unplanned readmission, or unplanned ED visit within 30 days of discharge) (Table 2).