National improvements in resident physician-reported patient safety after limiting first-year resident physicians' extended duration work shifts: a pooled analysis of prospective cohort studies

Abstract

Background: The Accreditation Council for Graduate Medical Education (ACGME) enacted a policy in 2011 that restricted first-year resident physicians in the USA to work no more than 16 consecutive hours. This was rescinded in 2017.

Methods: We conducted a nationwide prospective cohort study of resident physicians for 5 academic years (2002-2007) before and for 3 academic years (2014-2017) after implementation of the 16 hours 2011 ACGME work-hour limit. Our analyses compare trends in resident physician-reported medical errors between the two cohorts to evaluate the impact of this policy change.

Results: 14 796 residents provided data describing 78 101 months of direct patient care. After adjustment for potential confounders, the work-hour policy was associated with a 32% reduced risk of resident physician-reported significant medical errors (rate ratio (RR) 0.68; 95% CI 0.64 to 0.72), a 34% reduced risk of reported preventable adverse events (RR 0.66; 95% CI 0.59 to 0.74) and a 63% reduced risk of reported medical errors resulting in patient death (RR 0.37; 95% CI 0.28 to 0.49).

Conclusions: These findings have broad relevance for those who work in and receive care from academic hospitals in the USA. The decision to lift this work hour policy in 2017 may expose patients to preventable harm.

Keywords: duty hours/work hours; medical error, measurement/epidemiology; patient safety.

Figure 1

The risk of resident physician-reported adverse patient safety outcomes for first-year resident physicians who contributed data after implementation of the 16 hours 2011 Accreditation Council for Graduate Medical Education work-hour limit relative to first-year resident physicians who contributed data from 2002 to 2007. Basic models are adjusted for hours in patient care. Fully adjusted models are adjusted for age, gender, specialty and hours in patient care. All models are generalised linear mixed models with a binomial distribution and logit link function. Created by the authors.