Impact of Extended-Duration Shifts on Medical Errors, Adverse Events, and Attentional Failures

Three hundred percent. Hold that number for a second. Not a three percent increase, not thirty — three hundred. That is how much more likely interns working five or more extended shifts per month were to report a fatigue-related preventable adverse event that resulted in a patient's death compared to months when they worked none. Picture the intern: first year out of medical school, third extended shift in a row, standing at a bedside at three a.m., making a decision that will follow that patient — and maybe that doctor — for the rest of their lives. The question Barger and colleagues set out to answer is whether that picture is as dangerous as it looks. The concern wasn't new. The Accreditation Council for Graduate Medical Education, known as the ACGME, had already placed limits on resident work hours. A 1999 report from the Institute of Medicine estimated that between 48,000 and 98,000 deaths each year in the United States result from medical error. What was already known from a randomized controlled trial in critical care units was that eliminating shifts of 24 hours or more reduced both significant medical errors and attentional failures measured by polysomnography, or sleep monitoring. But that trial was too small to say anything firm about preventable adverse events or fatalities, and it was conducted in a single setting. What happened across the full breadth of American residency training, across every specialty and hospital type, was still an open question. To answer it at scale, Barger and colleagues recruited 2,737 interns — residents in their first postgraduate year — and collected 17,003 monthly web-based reports over the 2002 to 2003 academic year. The mean number of monthly surveys completed per participant was 7.2, and 682 interns, about a quarter of the cohort, completed all twelve monthly surveys. The key methodological decision was the case-crossover design. Instead of comparing interns who worked many extended shifts to interns who worked few, the analysis compared each intern to themselves across months with different exposures. Months were sorted into three categories: no extended shifts, one to four, or five or more. Extended-duration was defined as a shift of at least 24 continuous hours. Because each intern served as their own control, a whole category of confounders — age, specialty, hospital environment, individual risk tolerance — dropped out of the equation automatically. The Mantel-Haenszel test was used to calculate pooled odds ratios across those categories. Barger and colleagues also ran a validation check. A random seven percent of participants kept daily work and sleep diaries, and those diaries correlated strongly with the monthly self-reports: Pearson correlations of 0.76 for average work hours and 0.94 for the number of extended shifts. The data were not just what tired interns vaguely remembered; they tracked closely with day-by-day records. Now for what the data actually show. In months with no extended shifts, interns reported at least one fatigue-related significant medical error in 3.8 percent of those months. In months with one to four extended shifts, that figure jumped to 9.8 percent — an odds ratio of 3.5. In months with five or more extended shifts, it reached 16 percent, an odds ratio of 7.5. Those are not marginal effects. A 7.5-fold increase in the odds of reporting a serious medical error is a large signal, and it's dose-dependent: the more extended shifts, the worse the outcome. Preventable adverse events — actual patient harm from a non-intercepted management error — showed a similar pattern. The odds ratio for months with one to four extended shifts was 8.7. For five or more extended shifts, it was 7.0. In absolute terms, reported rates of fatigue-related preventable adverse outcomes rose from 0.2 percent of months with no extended shifts to 1.6 percent in the highest-exposure category. For the most severe outcome, fatal adverse events, months with five or more extended shifts had an odds ratio of 4.1. That is the source of the three hundred percent figure. Attentional failures — defined in the study as nodding off or falling asleep — showed parallel effects. Falling asleep during rounds had an odds ratio of 5.5 for interns in the highest-exposure category. Falling asleep during lectures had an odds ratio of 4.3. Even falling asleep during surgery was elevated. The mean weekly work hours tell the underlying story clearly: 55.5 hours per week in months with no extended shifts, rising to 79.3 hours per week in months with five or more. These aren't small scheduling differences. They represent a fundamentally different physiological state. The findings held up to stress testing. Among the 682 interns who completed every monthly survey, the odds ratios for fatigue-related errors were 4.0 for one to four shifts and 6.7 for five or more — close to the full cohort numbers. A secondary analysis restricted only to months spent on hospital wards returned essentially the same odds ratios: 3.5 and 7.5. That ward-restricted analysis was important because it controlled for the possibility that low extended shift months were simply quieter rotations like radiology or outpatient clinics. Importantly, 81 percent of intern months were compliant with ACGME frequency guidelines, meaning these effects emerged even within the officially acceptable range of scheduling. The study also found that 83.6 percent of interns reported working more hours than the ACGME rules allowed after those rules came into effect, which tells you something about how the rules translated into practice. There are real limitations here, and Barger and colleagues are straightforward about them. All outcome data are self-reported. Interns were asked whether they believed fatigue or sleep deprivation caused a given error, and that wording could have inflated the measured association. Prior work also shows that self-report systems catch only a fraction of actual medical errors, meaning the true rates of harm were almost certainly higher than what shows up in these surveys — not lower. If anything, that undercounting would push against finding strong associations, so the magnitude of what was found is notable. The within-person design eliminates many confounders, but rotation-level confounding is harder to rule out completely. A month with no extended shifts might simply be a month on a less clinically intense service. The ward-restricted analysis addresses this directly, and the results held. The authors note a possible self-selection dynamic: interns in more demanding programs may both work more extended shifts and function in environments with different error detection and reporting cultures. What this study establishes is a quantitatively large, dose-dependent, nationally representative association between extended-duration shifts and the outcomes that matter most — errors, patient harm, and death. The prior randomized trial showed the signal in a controlled intensive care unit setting. Barger and colleagues showed it generalizes across the full range of American medical training, in a cohort large enough to detect effects on rare events like fatal adverse events. The policy question the paper raises but deliberately does not answer is what to do about it. Eliminating extended-duration shifts entirely changes continuity of care, alters the structure of medical education, and creates its own logistical challenges. Those trade-offs are real. But the data make it harder to treat the 24-hour shift as a neutral feature of training. When interns averaging around sixty-five hours a week but working even a handful of extended shifts show eightfold greater odds of a preventable adverse event, the cost of those shifts is no longer hypothetical. It is measurable. The intern is simultaneously a learner and a practicing clinician, and the patients on the other side of those three a.m. decisions are not abstractions either. The balance between those two facts is where the hardest work in medical education policy still needs to happen. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.