Does a Structured Handover Actually Reduce Errors?

In 10,740 patient admissions across nine hospitals, the medical-error rate was 24.5 per 100 admissions before the handoff program and 18.8 per 100 admissions after it, a relative reduction of 23% (P<0.001) (Starmer et al., 2014). Preventable adverse events, the harms a better handoff would plausibly head off, went from 4.7 to 3.3 events per 100 admissions, a 30% relative reduction (P<0.001).

Read the scale carefully before quoting either figure. These are rates per 100 admissions, so a single admission can contribute more than one event, and neither number is the share of patients who came to harm. The 23% is a relative change between two observed rates, not an absolute risk difference and not a probability that any given handover goes wrong.

The most informative number in the paper is the one that did not move. Nonpreventable adverse events, which no communication change should be expected to touch, stayed flat at 3.0 and 2.8 events per 100 admissions (P=0.79). That internal control is what separates this study from a plain before-and-after chart, because a broad measurement drift or a general safety trend across the study period would likely have shifted both categories together.

The design is a prospective multicenter pre-post intervention study: nine hospitals, resident physicians, active surveillance for errors, and no randomization or concurrent control group. Secular improvement in patient safety over the study period, observation effects on the clinicians being watched, and quality initiatives running alongside the program cannot be ruled out. The authors' own conclusion says implementation of the program was associated with reductions in medical errors and in preventable adverse events, and that verb choice is deliberate.

Four things shipped at once: a mnemonic to standardize oral and written handoffs, handoff and communication training, a faculty development and observation program, and a sustainability campaign. Nothing in the study separates their contributions, so a claim that a standardized template alone produced the result is not supported. Anyone planning to adopt one component and expect the published effect is reading more into the paper than it reports.

Results also varied by site. Significant error reductions appeared at six of nine sites, which means three sites in a coordinated multicenter program with shared training and shared measurement did not show one. The published abstract reports rates and p-values rather than confidence intervals, so the precision around the 23% figure is not something this review can state, and no interval should be attached to it.

The program moved the behavior it targeted. Across sites, significant increases were observed in the inclusion of all prespecified key elements in written documents and in oral communication during handoff, nine written elements and five oral elements, with P<0.001 for all 14 comparisons (Starmer et al., 2014). That is a coherent mechanism story: the content of handovers measurably improved in the same period the error rates fell.

The common operational objection is that structure costs time at shift change. This study did not find it. Oral handoff duration was 2.4 and 2.5 minutes per patient before and after (P=0.55), with no significant changes in resident workflow, including patient-family contact and computer time. Time-motion observation, not self-report, produced those workflow figures, which is a stronger basis than asking clinicians whether the new format felt slower.

One label belongs on this result before it travels anywhere. The senior author on Starmer et al. (2014) is Landrigan, the same investigator behind the resident duty-hour and extended-shift research cited elsewhere in this library, so the handoff finding and the duty-hour findings come from one Boston research group rather than independent teams. Shared surveillance procedures, shared method choices, and a shared working definition of what counts as a preventable error travel across all of that work.

This page also rests on a single pre-post evaluation. Later implementations of the same handoff bundle have been published, and none of them were checked for this review, so nothing here should be read as replication by a separate research group. Read the 23% as one program's result in nine hospitals during one study period, and read the six-of-nine site split as the best available warning that the same package does not land the same way everywhere.

The setting is narrow in ways that matter to anyone reading across from it. The population is resident physicians in nine hospitals handing over pediatric inpatients, every rate is expressed per 100 admissions, and each handover has a named receiver who assumes responsibility for a specific patient, a written record that already existed, and active surveillance able to find an error after the fact. A handover where the thing passed on is a queue, an open ticket, or a machine state, with no single receiver and no way to detect afterward what went wrong, is a different exchange wearing the same name.

The paper was published in 2014 with data collected earlier, before EHR-integrated handoff tools became common, so the written-handoff baseline it improved on may not match what a unit runs today. It remains the landmark citation in this literature, which is a reason to quote it accurately rather than a reason to treat its effect size as a forecast for a modern implementation.