Magnitude and modifiers of the weekend effect in hospital admissions: a systematic review and meta-analysis

Abstract

Objective: To examine the magnitude of the weekend effect, defined as differences in patient outcomes between weekend and weekday hospital admissions, and factors influencing it.

Design: A systematic review incorporating Bayesian meta-analyses and meta-regression.

Data sources: We searched seven databases including MEDLINE and EMBASE from January 2000 to April 2015, and updated the MEDLINE search up to November 2017. Eligibility criteria: primary research studies published in peer-reviewed journals of unselected admissions (not focusing on specific conditions) investigating the weekend effect on mortality, adverse events, length of hospital stay (LoS) or patient satisfaction.

Results: For the systematic review, we included 68 studies (70 articles) covering over 640 million admissions. Of these, two-thirds were conducted in the UK (n=24) or USA (n=22). The pooled odds ratio (OR) for weekend mortality effect across admission types was 1.16 (95% credible interval 1.10 to 1.23). The weekend effect appeared greater for elective (1.70, 1.08 to 2.52) than emergency (1.11, 1.06 to 1.16) or maternity (1.06, 0.89 to 1.29) admissions. Further examination of the literature shows that these estimates are influenced by methodological, clinical and service factors: at weekends, fewer patients are admitted to hospital, those who are admitted are more severely ill and there are differences in care pathways before and after admission. Evidence regarding the weekend effect on adverse events and LoS is weak and inconsistent, and that on patient satisfaction is sparse. The overall quality of evidence for inferring weekend/weekday difference in hospital care quality from the observed weekend effect was rated as 'very low' based on the Grading of Recommendations, Assessment, Development and Evaluations framework.

Conclusions: The weekend effect is unlikely to have a single cause, or to be a reliable indicator of care quality at weekends. Further work should focus on underlying mechanisms and examine care processes in both hospital and community.

Prospero registration number: CRD42016036487.

Keywords: Bayesian meta-analysis; adverse events; healthcare databases; mortality; systematic review; weekend effect.

Figure 1

Bayesian meta-analysis covering all types of admissions for the weekend effect on mortality (sorted by country). Note: Mohammed 2012 and Ruiz 2015 contributed to two estimates for each country as the weekend effect was estimated separately for different sub-populations (eg, emergency and elective admissions). ‘Posterior predictive’ indicates the predictive interval (see main text) obtained from the Bayesian meta-analysis. I²=16% (95% CrI for I² 0% to 62%). The I² represents the ratio of between-study variance to total variance in this three-level model. The apparently low I² could be attributed to the between-study variance being relatively small compared with the between-estimate variance within individual studies. As the wide CrI indicates, the I² was estimated with substantial uncertainty. Several studies included in the review were not included in this meta-analysis due to substantial overlap of data between studies; in this case, studies that were judged to have adopted the most comprehensive statistical adjustment were selected. CrI, credible interval.

Figure 2

Factors that may contribute to or modify the weekend effect.