Background: The data used by the U.S. Institute of Medicine to estimate deaths from medical errors come from a study that relied on nurse and physician reviews of medical records to detect the errors.
Objective: To measure the reliability of medical record review for detecting adverse events and negligent adverse events.
Design: Medical record review.
Setting: Hospitalizations in Utah and Colorado in 1992.
Measurements: After three independent reviews of 500 medical records, the following were measured: reliability and the effect of varying criteria for reviewer confidence in and reviewer agreement about the presence of adverse events.
Results: For agreements in judgments of adverse events among the three sets of reviews, the Îº statistics ranged from 0.40 to 0.41 (95% CIs ranged from 0.30 to 0.51) for adverse events and from 0.19 to 0.23 (CIs, 0.05 to 0.37) for negligent adverse events. Rates for adverse events and for negligent adverse events varied substantially depending on the degree of agreement and the level of confidence that was required among reviewers.
Conclusion: Estimates of adverse event rates from medical record review, including those reported by the Institute of Medicine in its 2000 report on medical errors, are highly sensitive to the degree of consensus and confidence among reviewers.