One in Thirty Hospital Deaths May Be Preventable, Landmark Review Finds

Every death that occurs inside a hospital raises an uncomfortable question: could this have been avoided? A new study from Oslo University Hospital, published in BMC Health Services Research, tackles that question with unusual rigor, and its findings carry implications for how hospitals around the world measure patient safety. By systematically reviewing the case records of more than a thousand patients who died during a single year, a team of Norwegian researchers concluded that roughly three percent of in-hospital deaths were preventable, with adverse events, most often tied to surgical treatment, acting as the direct cause of death in a substantial share of those cases.

The research team, led by Trine Sand Kaastad and Trond-Eirik Strand, set out to quantify a figure that has proven notoriously difficult to pin down. Existing literature places the proportion of preventable in-hospital deaths anywhere between 0.5 and 8.4 percent, a range so wide that it offers hospitals and health authorities little practical guidance. That uncertainty stems largely from the method at the heart of the study: retrospective case record review, or RCRR, in which trained reviewers comb through patient charts after death to determine whether harm from care, rather than the underlying disease, contributed to the outcome. The technique is considered the gold standard for assessing preventable deaths, but it has a well-documented weakness, namely that different reviewers looking at the same record can reach strikingly different conclusions.

To confront that problem, the researchers adopted a design that few studies of this kind have attempted. All 1,081 in-hospital deaths recorded at Oslo University Hospital in 2014 were reviewed by a single physician using a standardized Mortality Analysis form derived from the Institute for Healthcare Improvement’s method of reviewing 50 consecutive deaths. The same reviewer worked through the entire cohort over a period of eight years, an unusually long review window that would later prove scientifically revealing. In addition, a random subset of 20 percent of the cases was independently assessed by a patient safety specialist, allowing the team to measure how consistently two trained professionals judged the same material.

The headline result is sobering but bounded. Of the 1,078 deaths that could be formally reviewed, 32, or three percent, were judged to have had a probability of 50 percent or more of being prevented. In 19 of those cases, an adverse event was considered the direct cause of death, and such events clustered most frequently during or after surgical treatment. Perhaps just as telling is what happened after those deaths were identified: only 12 of the 32 preventable deaths, or 37.5 percent, had ever been reported through the hospital’s own incident reporting system. That gap suggests that formal reporting channels, whatever their administrative value, capture only a fraction of the lethal harm that retrospective review can uncover.

The dual-reviewer exercise exposed the method’s central vulnerability in quantitative terms. In the randomly selected subset assessed by both reviewers, one found 12 preventable deaths, a rate of six percent, while the other found 11, or 5.5 percent. Although those raw numbers appear reassuringly close, the overall interrater agreement was only fair, with a Kappa statistic of 0.4. Kappa corrects for chance agreement, and a value in that range means the two reviewers concurred far less often than the raw counts imply. In practical terms, a case one expert deemed preventable the other frequently did not, a discrepancy that reflects genuine ambiguity in judging counterfactuals: would this patient have survived with different care, or was death imminent regardless?

The most striking finding, however, emerged from the passage of time. Because the same physician reviewed the entire 2014 cohort across four consecutive review periods spanning eight years, the researchers could track whether the reviewer’s own judgments drifted. They did, and dramatically so. The proportion of deaths deemed preventable fell from 5.5 percent in the first review period to 1.0 percent in the last, a statistically significant decline with a p-value below 0.001. Nothing about the care provided in 2014 changed during those eight years; what changed was the person reading the charts. The authors interpret this as evidence that prolonged review work may alter a reviewer’s threshold for what counts as preventable, whether through growing caution, deepening familiarity with the hospital’s routines, or simple fatigue.

That drift has consequences well beyond academic debate. Hospitals and national health authorities increasingly rely on measures such as the Hospital Standardized Mortality Ratio and tools like the Global Trigger Tool to benchmark safety performance, and RCRR-based preventable death estimates feed into programs such as the UK’s Learning from Deaths initiative. If the proportion of deaths judged preventable depends heavily on who reviews the records and when, then comparisons between hospitals, or even within a single hospital over time, may reflect reviewer characteristics more than genuine differences in care quality. The Norwegian team states this conclusion plainly: using RCRR to benchmark preventable deaths may capture reviewer differences more than hospital differences.

The study’s methodology deserves attention for how it balanced rigor with feasibility. The Mortality Analysis form standardized what information the reviewer extracted and how preventability was scored, anchoring judgments to the convention that a death counts as preventable only when the probability reaches at least 50 percent. Reviewing every death rather than a sample eliminated sampling error, and the eight-year single-reviewer design, initially a practical necessity, became an accidental experiment on intra-rater reliability. The project was conducted as a quality assurance initiative under Norwegian law, approved by the hospital’s data protection officer, with data made non-identifiable before analysis and handled in accordance with GDPR and the Declaration of Helsinki.

Placed against the international literature, the three percent figure sits comfortably within the published range. Previous retrospective review studies, from the landmark Harvard Medical Practice Study onward, have produced estimates spanning half a percent to more than eight percent, shaped by differences in case mix, review instruments, reviewer training, and preventability thresholds. The Oslo findings align with the middle of that distribution, which the authors take as evidence that a standardized form can produce results consistent with comparable institutions. Yet the fair interrater agreement and the dramatic intra-rater drift temper any temptation to treat the number as a precise measurement of hospital safety.

For clinicians and policymakers, the study offers a dual message. First, preventable deaths remain a real and measurable problem: three percent of deaths at one of Europe’s largest hospitals, many linked to surgical care, and most never captured by routine incident reporting. Second, the instruments used to count those deaths are less stable than the patient safety movement has often assumed. The authors suggest that RCRR may be better suited to identifying individual cases for learning and improvement than to producing benchmarking statistics that rank institutions. As hospitals worldwide invest in mortality review programs, the Oslo experience stands as a caution: before comparing preventable death rates between hospitals, the field must first solve the harder problem of whether two reviewers, or even one reviewer over time, can agree on what preventable means.

Subject of Research: Preventable in-hospital deaths due to adverse events assessed by retrospective case record review

Article Title: Preventable in-hospital deaths due to adverse events: a retrospective case record review study

Article References: Kaastad, T. S., Tveiten, H., Riiser, T. J., Bukholm, G., & Strand, T.-E. (2026). Preventable in-hospital deaths due to adverse events: a retrospective case record review study. BMC Health Services Research. https://doi.org/10.1186/s12913-026-15787-2

Image Credits: AI Generated

DOI: 10.1186/s12913-026-15787-2

Keywords: patient safety, preventable death, in-hospital mortality, retrospective case record review, adverse events, hospital benchmarking, interrater reliability, mortality review, Oslo University Hospital, health services research, incident reporting, quality of care