Abstract
Purpose
The effect of reduced hospital staffing during weekends on in-hospital mortality is not known. We compared mortality rates between patients admitted on weekends and weekdays and whether weekend-weekday variation in rates differed between patients admitted to teaching and nonteaching hospitals in California.
Methods
The sample comprised patients admitted to hospitals from the emergency department with any of 50 common diagnoses (N = 641,860). Mortality between patients admitted on weekends and those admitted on weekdays (the “weekend effect”) was compared. The magnitude of the weekend effect was also compared among patients admitted to major teaching, minor teaching, and nonteaching hospitals.
Results
The adjusted odds of death for patients admitted on weekends when compared with weekdays was 1.03 (95% confidence interval [CI]: 1.01 to 1.06; P = 0.0050). Three diagnoses (cancer of the ovary/uterus, duodenal ulcer, and cardiovascular symptoms) were associated with a statistically significant weekend effect. None of the 50 diagnoses demonstrated a statistically significant reduction in mortality for weekend admissions as compared with weekday admissions. Mortality was similar among patients admitted to major (odds ratio [OR] = 1.06; 95% CI: 0.94 to 1.19) and minor (OR = 1.03; 95% CI: 0.97 to 1.09) teaching hospitals, compared with nonteaching hospitals. However, the weekend effect was larger in major teaching hospitals compared with nonteaching hospitals (OR =1.13 vs. 1.03, P = 0.03) and minor teaching hospitals (OR = 1.05, P = 0.11).
Conclusion
Patients admitted to hospitals on weekends experienced slightly higher risk-adjusted mortality than did patients admitted on weekdays. While overall mortality was similar for patients admitted to all hospital categories, the weekend effect was larger in major teaching hospitals and is cause for concern.
Despite relatively constant demand for acute care, hospitals faced with economic constraints and problems of employee satisfaction generally reduce staffing and availability of services during weekends (
1
). While anecdotes about the potential consequences of reductions in weekend service abound, information on the effects of weekend admission on in-hospital mortality is limited.Earlier studies suggested that weekend admission was associated with increased resource utilization and slightly higher perinatal mortality when compared with weekdays (
2
, 3
, 4
, 5
). Only recently, however, have researchers begun to examine the effects of weekend admission on in-hospital mortality among more diverse patient groups. Bell et al found that Canadian patients who were admitted on weekends to acute care hospitals from emergency departments had higher risk-adjusted mortality than patients admitted on weekdays (the “weekend effect”) (6
). Barnett et al found a 9% increase in risk-adjusted mortality among patients admitted to intensive care units on weekends when compared with weekdays (7
).While these results are concerning, the generalizability of these findings to the broad array of patients admitted to hospitals in the United States remains unclear. Furthermore, prior studies have not examined whether the weekend effect might vary with hospital teaching status, even though teaching status has been associated with more rapid evaluation of unstable patients and potentially better outcomes (
8
, 9
, 10
, 11
).The current study used a database of all 1998 admissions to acute care hospitals in California to examine the effects of weekend admission on in-hospital mortality among patients with any of 50 common diagnoses, and to assess whether these effects varied by hospital teaching status. We hypothesized that the weekend effect previously demonstrated in limited patient samples would also exist among a diverse sample from California. Furthermore, we suspected that major teaching hospitals would demonstrate a smaller weekend effect.
Methods
Data source
We used the 1998 California Office of Statewide Health Planning and Development Discharge Data File, which was developed as part of the Healthcare Cost and Utilization Project conducted by the Agency for Healthcare Research and Quality (AHRQ) (
12
). The dataset contains detailed administrative data for all admissions during 1998 to all licensed acute care hospitals (excluding Veterans Administration hospitals) in California. Trained abstractors working on site collected demographic and clinical data on each patient at the time of discharge from the hospital, such as age, race, sex, nature of the admission (variable “admission type” categorized as scheduled, unscheduled, infant, or unknown), source of admission (e.g., home, interhospital transfer, emergency department), name of the hospital to which the patient was admitted, the principle diagnosis deemed responsible for admission as identified at the time of discharge (according to the International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] code), and up to 24 secondary diagnoses present at admission. The California Office of Statewide Health Planning periodically reviews the data to assess accuracy and coding errors; hospitals that exceed prespecified thresholds are required to repeat their abstractions (Agency for Healthcare Research and Quality. Healthcare cost and utilization project. Available at: http://www.ahrq.gov/data/hcup/. Accessed June 2, 2003
13
).California Healthcare Cost and Utilization Project. California Patient Discharge Data Reporting Manual. 3rd ed. Available at: http://www.oshpd.state.ca.us/hid/HID/patient/discharges/patmanuals/pd3/intro.pdf. Accessed June 2, 2003
For each admission, the ICD-9-CM code was collapsed to the 3-digit code, then ranked according to the total number of in-hospital deaths that occurred for each diagnosis during 1998. The 50 ICD-9-CM codes with the greatest number of deaths were included in subsequent analyses so as to maximize the number of events (and therefore statistical power) for the study.
It was recognized that a potentially important source of bias in the study would be that scheduled admissions would be more likely to take place on weekdays and that such patients would likely have better outcomes than those admitted on weekends when scheduled admissions would be unlikely. To examine the effect of this potential bias (and identify a similar cohort of patients admitted on weekdays and weekends), analyses were conducted in three progressively restricted cohorts: all admissions, unscheduled admissions, and unscheduled admissions from the emergency department.
Weekend admission
Patients were considered weekend admissions if they were admitted between 12:00 am on Saturday and 11:59 pm on Sunday. Patients admitted on all other days were considered weekday admissions.
Teaching status
The dataset was linked to the American Hospital Association 1998 Hospital Survey, which contains detailed hospital characteristics for most acute care hospitals in the United States. The linkage was performed first by matching the names of hospitals on the California list to those in the American Hospital Association dataset. In cases where the names did not match precisely, the addresses and telephone numbers provided by the American Hospital Association were used to re-evaluate the name of the hospital located at the specified address. Once the exact name was identified, the linkage was performed, provided that the numbers of admissions listed in the two datasets were similar. We were able to match 441 (85%) of the 518 hospitals in the California dataset to the American Hospital Association Survey data. Hospitals that were members of the Council of Teaching Hospitals of the Association of American Medical Colleges were categorized as “major teaching” (
14
). Hospitals that were not members, but had one or more accredited residency training programs, were categorized as “minor teaching.” Other hospitals were categorized as “nonteaching.”Risk adjustment
Comorbidity was assessed using secondary ICD-9-CM codes to define 30 specific comorbid conditions (
15
). This approach was designed to be conservative and excludes comorbid conditions that may represent complications of care or that are related to the primary reason for hospitalization. Algorithms for defining each condition were obtained from the AHRQ website (16
). Charlson comorbidity index scores were calculated for each patient using the Dartmouth-Manitoba approach, which considers 16 comorbid conditions (Elixhauser Comorbidity Software. Available at: http://www.ahrq.gov/data/hcup/comorbid.htm. Accessed June 2, 2003
17
).Statistical analysis
Differences in demographic characteristics, comorbidity, and mortality between patients admitted on weekends and weekdays were compared using the chi-square test or t test. The proportion of patients who were admitted on weekends as compared with weekdays was calculated for each of the progressively more restricted cohorts. It was anticipated that as the restrictions were applied, the ratio of weekend to weekday admissions would approach 40%, reflecting the expectation that a similar number of “emergency” admissions should occur on each day of the week.
For each of the 50 conditions, mortality among weekend and weekday admissions was compared using logistic regression fitted by generalized estimating equations that account for clustering of patients within hospitals and for variations in the weekend effect across hospitals. This approach provides identical parameter estimates to ordinary logistic regression, but more robust confidence intervals (
18
). Unadjusted analyses included only an indicator variable for weekend admission; the coefficient associated with this variable was used to estimate the odds of death for weekend admissions as compared with weekday admissions (the weekend effect odds ratio). Separate analyses were conducted by admission cohort.Additional analyses adjusted for demographic characteristics (age, sex, and race) and comorbidity using indicator variables to represent each of the individual conditions considered by Elixhauser et al (
15
). These models also included an indicator variable for weekend admission.To estimate the weekend effect for all 50 diagnoses in aggregate, we employed a two-stage modeling strategy. The first stage involved determining a predicted risk of death for each patient from disease-specific logistic regression models that included variables for demographic factors and comorbidity, but not a variable for weekend admission. The second-stage model included all patients as well as the logit of the predicted risk of death for each patient and an indicator variable for weekend admission. We chose this two-stage approach (in lieu of developing an aggregate model with individual risk variables) to allow the effect of the risk variables to vary across diagnoses. Lastly, we determined the magnitude of the weekend effect in teaching and nonteaching hospitals by adding indicator variables for major and minor teaching hospitals and interaction terms between teaching status and weekend admission.
P values <0.05 (two-tailed) were considered statistically significant. All statistical analyses were performed using Stata 8.0 (Stata Corp., College Station, Texas) and SAS 9.0 (SAS Institute Inc., Cary, North Carolina).
Results
During 1998 there were 3,725,373 admissions to acute care hospitals in California and 86,557 deaths. The 50 study diagnoses were responsible for 1,100,984 admissions (30% of all admissions) and 64,840 deaths (75% of all deaths). Exclusion of scheduled admissions (n = 191,153) and patients who had not been admitted from the emergency department (n = 267,971) resulted in a final cohort of 641,860 patients and 41,702 deaths. An additional 71,700 patients were excluded from the teaching status analysis because they had been admitted to facilities that could not be linked to the American Hospital Association Survey data.
The mean age of patients in the final cohort was 67 years and 52% were women (Table 1). Overall, 461,102 patients (72%) were admitted on weekdays and 180,758 (28%) were admitted on weekends. Differences between the 570,160 admissions who were successfully matched to the American Hospital Association Survey data and the 71,700 patients who could not be matched, as well as for patients admitted on weekends versus weekdays, were generally minimal. However, patients who were not successfully matched were slightly older, more likely to be black, and had higher in-hospital mortality compared with those who were matched.
Table 1Demographic Data for Unscheduled Patients Admitted with 50 Selected Diagnoses on Weekdays and Weekends
| Characteristic | Matched to American Hospital Association Data | Not Matched to American Hospital Association Data | ||
|---|---|---|---|---|
| Weekend (n = 160,615) | Weekday (n = 409,545) | Weekend (n = 20,143) | Weekday (n = 51,557) | |
| Number (%) or Mean ± SD | ||||
| Age (years) | 66.8 ± 19.4 | 66.8 ± 19.1 | 68.0 ± 18.6 | 67.5 ± 18.6 |
| Women | 83,588 (52) | 212,650 (52) | 10,770 (53) | 27,278 (53) |
| White | 125,034 (78) | 316,569 (77) | 14,278 (71) | 36,498 (71) |
| Black | 13,955 (9) | 37,677 (9) | 3314 (17) | 8718 (17) |
| Asian | 9659 (6) | 24,579 (6) | 1388 (7) | 3358 (7) |
| Mortality | 10,654 (6.6) | 26,175 (6.4) | 1388 (6.9) | 3485 (6.8) |
| Charlson comorbidity score | 1.11 ± 1.54 | 1.11 ± 1.54 | 1.15 ± 1.55 | 1.14 ± 1.51 |
Mortality among all diagnoses
As the cohort was progressively restricted, the ratio of weekend to weekday admissions increased from 0.26 among all admissions to 0.39 among admissions from the emergency department (Table 2). This increase in the weekend/weekday ratio corresponded to a reduction in the unadjusted weekend effect odds ratio from 1.20 among all admissions to 1.03 among admissions from the emergency department; a similar trend was observed in the adjusted analysis. After adjustment for differences in demographic characteristics and comorbidity, mortality was 3% higher (odds ratio [OR] = 1.03; 95% confidence interval [CI]: 1.01 to 1.06; P = 0.005) with weekend admissions, compared with weekday admissions, among patients admitted from the emergency department.
Table 2Effect of Progressive Cohort Restriction on Ratio of Weekend/Weekday Admissions and Weekend Effect
| Type of Admission | No. of Admissions | Ratio of Weekend to Weekday Admissions | Mortality (%) | Weekend Effect | ||
|---|---|---|---|---|---|---|
| Weekend | Weekday | Unadjusted | Adjusted | |||
| Odds Ratio (95% Confidence Interval) | ||||||
| All admissions | 1,100,984 | 0.26 | 6.7 | 5.7 | 1.20 (1.17–1.23) | 1.12 (1.09–1.14) |
| Unscheduled admissions | 909,832 | 0.31 | 6.7 | 6.0 | 1.14 (1.11–1.16) | 1.10 (1.07–1.12) |
| Emergency department admissions | 641,861 | 0.39 | 6.7 | 6.4 | 1.03 (1.01–1.06) | 1.03 (1.01–1.06) |
Mortality among individual diagnoses
Among the 50 study diagnoses, the total number of emergency department admissions available for analysis ranged from 390 for ovarian cancer to 58,807 for pneumonia (Table 3). As the cohort was progressively restricted from all admissions to emergency department admissions only, the number of diagnoses with a weekend effect odds ratio indicating decreased mortality among weekend admissions as compared with weekday admissions (OR <1.0) increased from 8 to 15. Among patients admitted from the emergency department, only three diagnoses—cancer of the ovary/uterus, duodenal ulcer, and cardiovascular symptoms—had adjusted weekend effect odds ratios that were associated with significantly increased mortality among weekend admissions.
Table 3Number of Admissions and the Risk-Adjusted Weekend Effect for Each of the Selected Diagnoses as the Cohort was Progressively Restricted
| Diagnosis (ICD-9-CM Code) | Number of Admissions Through Emergency Department | Weekend Effect | ||
|---|---|---|---|---|
| All Admissions | Unscheduled Admissions | Unscheduled Admissions Through Emergency Department | ||
| Odds Ratio (95% Confidence Interval) | ||||
| Cancer of ovary, uterus (183) | 390 | 3.12 (1.99–4.00) | 2.20 (1.35–3.59) | 2.12 (1.08–4.16) |
| Duodenal ulcer (532) | 6059 | 1.64 (1.22–2.20) | 1.56 (1.16–2.10) | 1.47 (1.07–2.01) |
| Symptoms involving cardiovascular system (785) | 1027 | 1.80 (1.25–2.56) | 1.54 (1.06–2.24) | 1.58 (1.01–2.50) |
| Acute myeloid leukemia (205) | 807 | 1.20 (0.95–1.51) | 1.15 (0.87–1.52) | 0.92 (0.63–1.35) |
| Acute myocardial infarction (410) | 42,974 | 1.09 (1.00–1.17) | 1.04 (0.97–1.12) | 1.01 (0.93–1.10) |
| Acute renal failure (584) | 6164 | 1.07 (0.93–1.24) | 1.06 (0.91–1.24) | 0.98 (0.82–1.18) |
| Alteration in level of consciousness (780) | 29,251 | 0.81 (0.59–1.11) | 0.96 (0.67–1.37) | 1.11 (0.77–1.60) |
| Aortic aneurysm (441) | 1682 | 2.13 (1.77–2.58) | 1.38 (1.13–1.69) | 1.13 (0.90–1.41) |
| Aspiration pneumonia (507) | 12,481 | 1.10 (1.00–1.20) | 1.10 (1.01–1.20) | 1.08 (0.98–1.20) |
| Atherosclerosis (440) | 1731 | 1.52 (1.04–2.03) | 1.42 (1.05–1.91) | 1.07 (0.70–1.63) |
| Breast cancer (174) | 350 | 2.86 (1.59–5.17) | 1.69 (1.07–2.68) | 1.14 (0.60–2.20) |
| Cancer of trachea, bronchus, lung (162) | 4784 | 1.51 (1.31–1.73) | 1.20 (1.04–1.39) | 1.11 (0.93–1.32) |
| Cardiac arrhythmia (427) | 29,995 | 1.28 (1.08–1.50) | 1.24 (1.09–1.42) | 1.10 (0.95–1.28) |
| Cerebral infarction (434) | 24,565 | 1.04 (0.93–1.15) | 1.01 (0.90–1.13) | 0.99 (0.88–1.11) |
| Cerebrovascular disease not otherwise specified (436) | 7932 | 1.04 (0.86–1.25) | 1.08 (0.86–1.37) | 1.12 (0.91–1.37) |
| Chronic airway obstruction (496) | 1947 | 0.88 (.063–1.24) | 1.02 (0.67–1.56) | 1.07 (0.61–1.86) |
| Chronic bronchitis (491) | 28,914 | 1.06 (0.91–1.23) | 1.13 (0.97–1.32) | 1.06 (0.89–1.26) |
| Chronic ischemic heart disease (414) | 38,999 | 0.97 (0.80–1.17) | 1.06 (0.85–1.32) | 1.10 (0.82–1.47) |
| Chronic liver disease (571) | 9303 | 1.21 (1.06–1.40) | 1.23 (1.07–1.40) | 1.14 (0.99–1.32) |
| Colon cancer (153) | 2329 | 1.80 (1.44–2.24) | 1.39 (1.09–1.77) | 1.27 (0.96–1.69) |
| Diabetes (250) | 24,284 | 1.09 (0.87–1.36) | 1.07 (0.85–1.34) | 1.03 (0.78–1.34) |
| Diverticulosis (562) | 11,850 | 1.14 (0.84–1.54) | 1.02 (0.74–1.40) | 0.86 (0.61–1.22) |
| Dysfunction of cardiac device/graft (996) | 11,645 | 1.24 (1.02–1.52) | 1.14 (0.93–1.39) | 1.02 (0.78–1.32) |
| Electrolyte disorders (276) | 26,149 | 0.94 (0.81–1.08) | 0.93 (0.79–1.08) | 0.90 (0.75–1.09) |
| Gastric ulcer (531) | 7434 | 0.80 (0.58–1.12) | 0.76 (0.55–1.09) | 0.71 (0.50–1.01) |
| Gastrointestinal bleed (578) | 9057 | 1.28 (1.04–1.57) | 1.27 (1.04–1.57) | 1.14 (0.92–1.42) |
| Heart failure (428) | 55,835 | 1.03 (0.95–1.12) | 1.06 (0.98–1.16) | 1.05 (0.96–1.16) |
| Hip fracture (820) | 22,001 | 1.13 (0.97–1.32) | 1.14 (0.97–1.34) | 1.13 (0.93–1.36) |
| Hypertensive heart disease (402) | 8934 | 1.09 (0.84–1.43) | 1.21 (0.92–1.60) | 1.15 (0.85–1.57) |
| Hypertensive heart and kidney disease (404) | 4264 | 1.06 (0.77–1.46) | 1.14 (0.81–1.61) | 1.07 (0.72–1.58) |
| Intercerebral hemorrhage (431) | 6210 | 1.20 (1.08–1.34) | 1.11 (1.00–1.23) | 0.98 (0.87–1.10) |
| Intestinal disease not otherwise specified (569) | 3685 | 1.23 (0.91–1.66) | 1.06 (0.78–1.43) | 0.96 (0.69–1.34) |
| Intestinal obstruction (560) | 17,979 | 1.14 (0.95–1.35) | 1.18 (0.98–1.42) | 1.17 (0.96–1.42) |
| Liver abscess or portal hypertension (572) | 3044 | 0.82 (0.63–1.07) | 0.86 (0.67–1.13) | 0.91 (0.67–1.23) |
| Lung disease not otherwise specified (518) | 15,806 | 1.01 (0.92–1.10) | 1.05 (0.96–1.15) | 1.02 (0.93–1.12) |
| Lymphoma (202) | 1303 | 1.07 (0.82–1.40) | 1.01 (0.76–1.35) | 0.90 (0.62–1.31) |
| Mesenteric ischemia (557) | 2791 | 1.20 (0.99–1.46) | 1.11 (0.91–1.35) | 0.92 (0.72–1.16) |
| Other brain conditions (348) | 1283 | 1.35 (1.04–1.74) | 1.20 (0.91–1.58) | 1.13 (0.81–1.58) |
| Pancreatic cancer (157) | 969 | 1.27 (0.96–1.67) | 1.11 (0.82–1.50) | 0.91 (0.61–1.34) |
| Pancreatic disease (577) | 16,229 | 1.02 (0.80–1.32) | 1.01 (0.78–1.31) | 1.01 (0.75–1.36) |
| Pneumonia (482) | 14,199 | 1.00 (0.91–1.10) | 1.03 (0.93–1.14) | 1.01 (0.91–1.13) |
| Pneumonia, organism not specified (486) | 58,807 | 0.99 (0.92–1.06) | 1.00 (0.94–1.08) | 0.94 (0.87–1.02) |
| Pulmonary embolism (415) | 4341 | 1.42 (1.15–1.76) | 1.36 (1.09–1.70) | 1.22 (0.59–1.60) |
| Secondary cancer of lung or gastrointestinal tract (197) | 5070 | 1.46 (1.28–1.66) | 1.31 (1.16–1.47) | 1.09 (0.94–1.27) |
| Secondary cancer of other site (198) | 3925 | 1.35 (1.13–1.61) | 1.25 (1.03–1.51) | 1.19 (0.94–1.52) |
| Sepsis (038) | 25,690 | 1.10 (1.03–1.17) | 1.09 (1.02–1.17) | 1.04 (0.96–1.12) |
| Subarachnoid hemorrhage (430) | 1819 | 1.17 (0.99–1.39) | 1.08 (0.92–1.28) | 0.98 (0.79–1.21) |
| Traumatic brain injury (852) | 2510 | 1.14 (0.95–1.37) | 1.11 (0.92–1.34) | 1.01 (0.81–1.26) |
| Unspecified intercranial hemorrhage (432) | 1459 | 1.44 (1.11–1.85) | 1.34 (1.04–1.73) | 1.15 (0.89–1.50) |
| Urinary tract disease (599) | 19,911 | 0.91 (0.73–1.13) | 0.94 (0.75–1.19) | 0.92 (0.72–1.19) |
ICD-9-CM = International Classification of Diseases, Ninth Revision, Clinical Modification.
Influence of hospital teaching status
Patients admitted from the emergency department were admitted to 333 different hospitals: 17 major teaching hospitals, 61 minor teaching hospitals, and 255 nonteaching hospitals. Overall, 10% of patients (n = 60,775) were admitted to major teaching hospitals, 25% (n = 140,418) to minor teaching hospitals, and 65% (n = 368,967) to nonteaching hospitals (Table 4). In analyses of admissions from the emergency department, unadjusted mortality rates were lower for patients admitted to major teaching hospitals compared with nonteaching hospitals on weekdays (5.7% vs. 6.7%, P = 0.01) but were similar for patients admitted on weekends (6.6% vs. 6.8%, P = 0.71). After adjusting for demographic factors and comorbidity, the odds of death were similar in major (OR = 1.06; 95% CI: 0.94 to 1.19; P = 0.36) and minor (OR = 1.03; 95% CI: 0.97 to 1.09; P = 0.31) teaching hospitals, as compared with nonteaching hospitals. In analyses that included interaction effects between weekend admission and teaching status, the magnitude of the weekend effect was greater in major teaching hospitals (OR = 1.13) than in nonteaching (OR = 1.03, P = 0.03) and minor teaching (OR = 1.05, P = 0.11) hospitals, but similar between minor teaching and nonteaching hospitals (P = 0.42).
Table 4Number of Admissions and Risk-Adjusted In-Hospital Mortality among Patients Admitted to Major Teaching, Minor Teaching, and Nonteaching Hospitals
| Type of Hospital | Number of Admissions (n = 570,160) | Mortality (%) | Mortality | Weekend Effect | |
|---|---|---|---|---|---|
| Weekday | Weekend | Odds Ratio (95% Confidence Interval) | |||
| Major teaching | 60,775 | 5.7 | 6.6 | 1.06 (0.94–1.19) | 1.13 (1.04–1.22) |
| Minor teaching | 140,418 | 6.0 | 6.3 | 1.03 (0.97–1.09) | 1.05 (1.00–1.09) |
| Nonteaching | 368,967 | 6.7 | 6.8 | 1.00 | 1.03 (0.99–1.06) |
* Adjusted for patient age, race, sex, and comorbid conditions.
Discussion
This study demonstrates that patients admitted to hospitals in California from the emergency department on weekends had a minimally increased odds of dying during hospitalization when compared with patients admitted on weekdays. The study also demonstrates the potential influence of selection bias on studies measuring the weekend effect. Analyzing three progressively restrictive cohorts, the weekend effect declined as selection criteria increased, and, importantly, as the proportion of weekend admissions increased. We also found that major teaching hospitals had greater weekend-weekday variations in mortality (a larger weekend effect) than did minor teaching or nonteaching hospitals, which is in contrast to our initial hypothesis.
Our finding of a 3% increase in risk-adjusted mortality among the aggregate data and that 6% (3/50) of specific diagnoses demonstrated a statistically significant weekend effect suggests that the weekend effect is smaller than reported in previous studies. Barnett et al reported a 9% increase in mortality for patients admitted on weekends, but this difference may have been due to unmeasured differences in severity of illness (
7
). Bell and Redelmeier (6
), in their study of patients in Canada, found that weekend admission was associated with a 4% increase in mortality among their aggregate dataset and that 23% of diagnoses were associated with a statistically significant weekend effect. The smaller weekend effect observed in our study is likely due to several factors. The proportion of all patients admitted on weekends versus weekdays was 37% in the Canadian study compared with 39% in ours, suggesting that the weekend effect in the former study may have been magnified by unmeasured differences between patients admitted on weekends and weekdays. Further, there may be actual differences in care between Canada and the United States (19
, 20
, 21
).The progressive reduction in the weekend effect as the cohort was progressively restricted from all admissions to admissions from the emergency department highlights both the limitations of risk adjustment and the importance of selecting an appropriate cohort for observational studies. Despite the use of a well-validated risk-adjustment methodology for administrative data, conducting the analysis on all unscheduled admissions would give the appearance of a clinically important 10% increase in mortality for patients admitted on the weekends; however, the fact that the ratio of weekend to weekday admissions in this cohort was 31% rather than 40% as expected, hints at the possibility of unadjusted differences between the weekend and weekday admissions (e.g., many weekday admissions referred from providers' offices, but not weekend admissions because offices are closed). Such an assumption is supported by the subsequent finding that when the cohort was further restricted to patients admitted from the emergency department, the ratio of weekend to weekday admissions increased to 39% while the weekend effect was reduced to 3%.
Our finding that the magnitude of the weekend effect in major teaching hospitals was larger than in nonteaching hospitals was unexpected. We had expected that the relatively constant level of staffing provided by residents and fellows would result in a smaller weekend effect in major teaching hospitals. In retrospect, it is likely that our hypothesis was overly simplistic. The weekend effect reflects more than physician availability on weekends. The observed weekend effect more likely represents a confluence of factors that ultimately result in reduced quality of care, such as discontinuity of care, reduced levels of staffing, reduced availability of certain procedures, and an overall reduction in patient supervision when hospital staffing decreases during the weekend.
Our study has several limitations. While we used an established method for adjusting for comorbidity, it is possible that patients admitted on weekends had more severe disease and inherently worse prognoses than did those admitted on weekdays. We attempted to limit this potential bias by limiting our analysis to the cohort of emergency admissions, but confounding by unadjusted severity remains a possibility. Furthermore, the study was conducted using administrative data. While quality control measures are in place to monitor the data abstraction process, errors in data abstraction remain a possibility. However, there would be no obvious reason why sporadic abstraction errors would systematically vary between weekend and weekday admissions. Finally, it is possible that our measurement of the influence of weekend admission on patient mortality underestimates the adverse effect of weekend care because patients admitted on weekends may “crossover” to receive weekday care and vice versa.
In light of prior research, our findings confirm the presence of the weekend effect but suggest that it may be smaller than previously determined. The finding that the weekend effect may be larger in major teaching hospitals is cause for concern and requires confirmation.
Acknowledgements
The authors would like to express their thanks to Sanjay Saint, Tim Hofer, and A. Mark Fendrick for their assistance with a closely related project.
References
- Effect of weekend physical therapy treatment on postoperative length of stay following total hip and total knee arthroplasty.Physiother Can. 1993; 45: 245-249
- Variations in number of births and perinatal mortality by day of week in England and Wales.BMJ. 1978; 2: 1670-1673
- The weekend—a dangerous time to be born?.Br J Ob Gyn. 1981; 88: 1200-1203
- Induction of labor, birthweight and perinatal mortality by day of the week.Br J Ob Gyn. 1987; 94: 413-419
- Neonatal mortality in weekend vs weekday births.JAMA. 2003; 289: 2958-2962
- Mortality among patients admitted to hospitals on weekends as compared with weekdays.N Engl J Med. 2001; 345: 663-668
- Day of the week of intensive care admission and patient outcomes.Med Care. 2002; 40: 530-539
- Effects of admission to a teaching hospital on the cost and quality of care for Medicare beneficiaries.N Engl J Med. 1999; 340: 293-299
- Teaching hospitals and quality of care.Milbank Q. 2002; 80: 569-593
- Inpatient transfers to the intensive care unit; delays are associated with increased mortality and morbidity.J Gen Intern Med. 2003; 18: 77-83
- Patient outcomes in academic medical centers.Arch Surg. 2003; 138: 47-51
Agency for Healthcare Research and Quality. Healthcare cost and utilization project. Available at: http://www.ahrq.gov/data/hcup/. Accessed June 2, 2003
California Healthcare Cost and Utilization Project. California Patient Discharge Data Reporting Manual. 3rd ed. Available at: http://www.oshpd.state.ca.us/hid/HID/patient/discharges/patmanuals/pd3/intro.pdf. Accessed June 2, 2003
- Severity-adjusted mortality and length of stay in teaching and nonteaching hospitals.JAMA. 1997; 278: 485-490
- Comorbidity measures for use with administrative data.Med Care. 1998; 36: 8-27
Elixhauser Comorbidity Software. Available at: http://www.ahrq.gov/data/hcup/comorbid.htm. Accessed June 2, 2003
- Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases.J Clin Epidemiol. 1992; 45: 613-619
- Longitudinal data analysis using generalized linear models.Biometrika. 1986; 73: 13-22
- A comparison of cardiovascular procedure use between the United States and Canada.Health Serv Res. 1998; 33: 467-487
- Hospital utilization in Ontario and the United States.Can J Public Health. 1996; 87: 253-256
- A comparison of management patterns after acute myocardial infarction in Canada and the United States.N Engl J Med. 1993; 328: 779-784
Article info
Publication history
Accepted:
February 3,
2004
Received:
June 18,
2003
Footnotes
☆Funded in part by a Research Enhancement Award (01-094) from the Health Services Research and Development Service, Veterans Health Administration, Department of Veterans Affairs. Dr. Rosenthal is a Senior Quality Scholar, Office of Academic Affiliations, Veterans Health Administration.
Identification
Copyright
© 2004 Elsevier Inc. Published by Elsevier Inc. All rights reserved.
