Is elevated hospital mortality at weekends a selection issue?

Tuesday, June 14, 2016
Lobby (Annenberg Center)

Author(s): Rachel Meacock; Laura Anselmi; Soeren R Kristensen; Timothy Doran; Matt Sutton

Discussant: Andy Ryan

Background: The phenomenon of elevated mortality rates amongst patients admitted to hospital at the weekend has been found worldwide. This finding is often attributed to reduced availability of hospital staff and diagnostic services at the weekend, despite a lack of evidence for a causal link. Previous studies have focused only on the population of patients admitted to hospital. These patients represent a selection from the pool of patients visiting emergency departments. Little attention has been paid to how the demand for emergency care and the decision to admit patients varies by day of the week. These are potentially important sources of selection bias.

Aim: To highlight the potential importance of selection bias in the measurement of the weekend effect, and correct for this selection bias using a Heckman selection model estimated on the total population of patients visiting emergency departments in England.

Data: Individual patient-level administrative hospital data on all visits to the 140 major emergency departments in England during the financial year 2013/14 (12,900,687 visits in total), all subsequent emergency admissions to hospital (3,418,446 admissions) and in-hospital mortality within 30 days of attendance (145,019 deaths).

Methods: We estimate the weekend effects on the probability of admission to hospital given attendance at an emergency department, and the probability of mortality given admission to hospital. We then correct the mortality equation estimated only on the admitted patient population with a Heckman selection model estimated on the entire population of patients attending emergency departments. In the selection equation, we include hospital level variables describing daily fluctuations in the volume of demand and the availability of senior doctors that may affect the individual patient’s probability of admission.

Results: Average numbers of emergency department visits on weekend days are similar to week days (38,254 versus 37,812; difference = 442, 95% CI = -148 to 1031), but a lower proportion of the patient population attending emergency departments at the weekend are admitted to hospital (adjusted odds-ratio=0.946, CI 0.943 to 0.950). The mortality rate amongst the admitted population is higher for those admitted at the weekend (3.591% versus 3.418%) but the average number of deaths per day is lower (378 versus 388). After correcting for casemix, patients visiting emergency departments at the weekend are no more likely to die than those visiting during the week (adjusted odds-ratio = 1.010, CI 0.997 to 1.022). The Heckman model shows that there is significant selection bias amongst the admitted population and that correcting for this selection bias reduces or eliminates the elevated mortality rate at weekends.

Discussion: The composition of patients admitted to hospital in an emergency on weekend days differs from weekdays due to variations in patient flows and admission decisions. Variations in the demand for emergency care and the decision to admit patients may account for the previously observed weekend effect in mortality.