The “weekend effect” refers to the association between weekend hospital admissions and poorer outcomes, such as higher mortality rates. Analysis of National Health Service claims data from the United Kingdom suggested a 10% increase in 30-day mortality in patients admitted on Saturdays and 15% in patients admitted on Sundays,1 leading to the push for a 7-day work week and invoking controversial changes in their junior doctor (residency) working contract. Studies in the United States highlighting differences in outcomes for patients admitted on weekends compared to weekdays have mostly focused on specific diagnoses and results have been variable. Few have gone on to look at the association of weekend hospital admissions on cost2,3 and length of stay3 but results are overall inconclusive. Some have suggested that such poorer outcomes for patients admitted on weekends are due to reduced staffing and delayed procedures on weekends compared to weekdays, although this has been debated.4 The lack of consensus has made it difficult for hospitals to plan if and how to expand weekend manpower or services.
In the United States, increase in mortality rate for patients admitted on weekends has been demonstrated for a range of diagnoses, including pulmonary embolism,5 intracerebral hemorrhage,6 upper gastrointestinal hemorrhage,7,8 ruptured aortic aneurysm,9 heart failure,10 and acute kidney injury.11 However, other diagnoses such as atrial flutter or fibrillation,2 hip fractures,12 ischemic stroke,13 and esophageal variceal hemorrhage,14 show no difference in mortality between weekday and weekend admissions. Yet, other conditions such as myocardial infarction15,16 and subarachnoid hemorrhage17,18 have multiple studies with conflicting results. None of these studies have comprehensively looked at the effect of weekend admissions across all diagnoses nor compared the effect size between common diagnoses in the United States using the same risk adjustment. Reporting of differences in length of stay and cost is also rare.
We postulated that the weekend admissions are associated with increased mortality and length of stay, but that the effect would be heterogeneous between different diagnosis groups. Using a large nationally representative inpatient database, we investigated the association between weekend versus weekday admissions on in-hospital mortality, length of stay, and cost for acute hospitalizations in the United States. We performed subgroup analyses of the top 20 diagnoses to determine which diagnoses, if any, should be targeted for expanded weekend manpower or services.
We used information from the National Inpatient Sample (NIS) database for this study,19 which is the largest all-payer inpatient healthcare database in the United States. It contains administrative claims information on a 20% stratified sample of discharges from all hospitals participating in the Healthcare Cost and Utilization Project (HCUP), which includes over 90% of hospitals and 95% of discharges in the country. The NIS contains clinical and nonclinical data elements, including diagnoses, severity and comorbidity measures, demographics, admission characteristics, and charges.
The study included all patients who were 18 years or older and were admitted to hospitals participating in HCUP from 2012 to 2014. Elective or planned admissions were excluded from this study because of the anticipated degree of unmeasured confounding that would be present between patients electively admitted on weekends compared to weekdays.
The primary exposure variable was admission on weekends (defined as Friday midnight to Sunday midnight) compared to the rest of the week. The primary outcome variable was in-hospital mortality. The secondary outcome variables were length of stay (measured in integer days) and cost. Length of stay was compared only using only patients who survived the hospital admission to eliminate the effect of death in shortening the length of stay. Cost was calculated by using charges available in the NIS and multiplied by the accompanying cost-to-charge ratios. Charges reflect total amount that hospitals billed for services but do not reflect how much these services actually cost. The HCUP cost-to-charge ratios are hospital-specific data based on hospital accounting reports collected by the Centers for Medicare & Medicaid Services.19
Covariates included age, sex, race, income, payer, presence or absence of comorbidities as defined by the Elixhauser comorbidity index,20 risk of mortality, and severity of illness scores as defined by the 3M Health Information Systems.21 Mortality risk and severity of illness groups are defined by using a proprietary iterative process developed by 3M Health Information Systems using International Classification of Diseases, 9th Revision-Clinical Modification (ICD-9-CM) principal and secondary diagnosis codes and procedure codes, age, sex, and discharge disposition, evaluated with historical data.21 Severity of illness refers to the extent of physiologic decompensation or loss of function of an organ system, whereas risk of mortality refers to the likelihood of dying.
We compared patient characteristics and other covariates between patients emergently admitted on weekends and weekdays. Continuous variables that were not normally distributed were either categorized (age, risk of mortality, and severity of illness scores) or log-transformed if right skewed (length of stay and cost). Categorical data were reported as percentages and continuous data as medians (interquartile range). We compared the inpatient mortality rate between weekend and weekday admissions by using χ2 tests. Multivariable logistic regression was used to adjust for covariates of age, gender, race, payer, income, risk of mortality and severity of illness scores, number of comorbidities, and the presence or absence of each of the 29 comorbidities available in the database to determine an adjusted odds ratio (OR), P values, and confidence intervals (CIs).
We also compared the length of stay amongst survivors and costs between weekend and weekday admissions. Multivariable linear regression was applied to the natural log of these outcome variables and the coefficients exponentiated to determine the difference in length of stay and cost of weekend admissions as compared to weekday. Covariates in the model were the same as those used for the primary outcome.
To determine if particular diagnoses had a pronounced weekend effect, the above analyses were repeated in subgroups of the top 20 most prevalent diagnoses on weekends by using the Clinical Classifications Software for ICD-9-CM diagnosis groups. For subgroup analyses, a Bonferroni correction was used, so P values of <.0025 were considered significant.
Statistical analyses were performed by using SAS version 9.4 (SAS Institute Inc, Cary, NC). All regression models were run using PROC SURVEYREG for continuous outcomes and PROC SURVEYLOGISTIC for binary outcomes to account for the sampling structure of NIS. Two-sided P values of .05 were considered significant, apart from the Bonferroni correction applied to the subgroup analysis. As this study involved publicly available deidentified data, our study was exempt from institutional board review.
The crude in-hospital mortality rate was 2.8% for patients admitted on weekends and 2.5% for patients admitted on weekdays (unadjusted OR, 1.110; 95% CI, 1.105-1.113; P < .0001). This relationship was attenuated after adjustment for demographics, severity, and comorbidities, but remained statistically significant (OR 1.029; 95% CI, 1.020-1.039; P < .0001; Table 2), which corresponds to an adjusted risk difference of 0.07% increase in mortality of weekend admissions. The OR for mortality on weekends compared to weekdays was further calculated for each of the top 20 diagnoses (Table 3). Out of all the diagnosis groups, only 1 (urinary tract infection) had a statistically significant P value after Bonferroni correction. We also looked separately at patients who were electively admitted—there was a highly significant OR of mortality of 1.67 (95% CI, 1.60-1.74). Patients classified as elective admissions were excluded for subsequent analyses.
Length of Stay
The magnitude of association between weekend admissions and mortality in this large administrative database contradicts existing literature, which some believe conclusively proves the international phenomenon of the weekend effect.22,23 However, our results support a minimal increase in odds of death of 2.9%, with no consistent effect amongst the top 20 diagnoses. Only 1 diagnosis group (urinary tract infection) showed a statistically significant increase in mortality, which could be due to chance. In contrast, the policy-influencing paper in the United Kingdom reports that patients admitted on Saturdays and Sundays have an increased risk of death of 10% and 15%, respectively, compared to patients admitted on Wednesdays.24 They also repeated their measurements on a United Health Care Systems database, comprising 254 leading managed care hospitals in the US, over a time period of 3 months in 2010, and found a hazard ratio of 1.18 (95% CI, 1.11-1.26). Ruiz et al.22 combined almost 3 million medical records from 28 metropolitan hospitals in 5 different countries in the Global Comparators Project, including 5 in the United States, and showed increased mortality on weekends in all countries, concluding that the weekend effect is a systematic phenomenon.
There are several possible explanations for differences in our findings. Freemantle’s study differed to ours by comparing outcomes of weekends to an index of Wednesday; they also found an increased mortality on Mondays and Fridays, which could suggest the presence of residual confounding and doubt as to whether Wednesday is the ideal control group. A further difference is the definition of mortality—we looked at in-hospital mortality, as compared to 30-day mortality. In addition, Freemantle’s study included elective admissions. When we looked at the effect of weekend admissions on mortality, we found a highly significant OR of 1.67, compared to 1.03 in emergency admissions. We attributed this discrepancy to unmeasured confounding, such as preference of physicians or difference in classification of elective admissions in different hospitals. Because of significant effect modification of elective compared to emergency admissions, we decided to restrict our analysis to emergency admissions only. This also enabled direct associations with potential policy recommendations on whether to expand weekend clinical care, which is most relevant to emergency admissions. Finally, the Global Comparators Project only samples a small proportion of hospitals in each country, leading to limited generalizability; in addition, international comparisons are difficult to interpret due to differing health systems.
The overall and diagnosis-specific difference in length of stay was small and of doubtful clinical significance. With an adjusted decrease in length of stay in patients admitted on weekends of 2.24%, when applied to a median length of stay of 3 days, it translates into a 1.7-hour difference in length of stay. However, there was striking heterogeneity noted between diagnoses, with a difference ranging from 8.91% decrease in length of stay (mood disorders) to 7.14% increase in length of stay (nonspecific chest pain), which is likely to explain the overall small magnitude of effect. We noted that the diagnoses associated with increased length of stay for weekend admissions tended to be those requiring inpatient procedures or investigations, such as acute myocardial infarction (3.90% increase), acute cerebrovascular disease (2.15% increase), cardiac dysrhythmias (1.39% increase), nonspecific chest pain (7.14% increase), and biliary tract disease (4.88% increase). As hospitals often do not provide certain nonemergent procedures or investigations on weekends, delay in procedures or investigations may explain the increase in length of stay. These include percutaneous coronary intervention or stress testing for evaluation of cardiac ischemia and endoscopic procedures for biliary tract disease and gastrointestinal hemorrhage. It must, however, be noted in conjunction that numerous studies have established higher complication rates when nonemergent surgeries are performed out of hours or on weekends.25-28 Therefore, we suggest further studies to compare the effect of weekends on increased procedural complications as to any morbidity caused by increased length of stay, which the present dataset was unable to capture. Another potential explanation for the heterogeneity in length of stay could be the greater availability of caregivers to assist with discharge on weekends, such as for patients admitted for mood disorders.
Surprisingly, weekend admissions appeared to be less costly than weekday admissions overall. Because of the large sample size, very minor differences in cost are likely to be statistically significant. Indeed, for the absolute difference of 0.45%, given a median cost of $6562 on weekends, this only represents a cost saving of approximately $30 per patient admission. There was also heterogeneity observed amongst the different diagnosis groups, and cerebrovascular disease, biliary tract disease and gastrointestinal hemorrhage, which were also associated with increase length of stay, were associated with an increased cost. However, our study is unable to establish causation, and differences in staffing numbers and reimbursement on weekends may confound cost estimates. We propose that further studies using hospital databases with greater granularity in data are necessary to determine the etiology of cost differences between weekends and weekdays.
Our study’s key strengths are the large sample size and generalizability to the US. As a large administrative database, we recognize the likelihood of inconsistencies in hospital coding for covariates, diagnoses, and charges, which may lead to misclassification bias. The NIS definition of weekend (Friday midnight to Sunday midnight) may differ from other definitions of weekend; ideally Friday 5 pm to Monday 8 am may be more clinically representative. This cohort of hospital admissions also does not account for the day of presentation to the emergency department, but rather only the day that ward admission was documented. The variable delays in emergency department, for example if emergency departments are busier on weekends, leading to delays in ward admission, may confound our results. Our exclusion of elective admissions was dependent on the administrative coding of elective versus emergency admissions, of which the definition may differ between hospitals. Finally, despite adjustment on clinical and sociodemographic covariates, there is a possibility of residual confounding in this retrospective comparison between weekend and weekday admissions.
Our study does not suggest that system-wide policies to increase weekend service coverage will impact mortality, although effects on length of stay and cost are inconclusive. Hospitals wishing to improve coverage may consider focusing on procedural diagnoses as listed above which may shorten length of stay, although the out-of-hours complication rate should be carefully monitored.
The authors declare no conflicts of interest.