Even for those interested in patient safety and quality improvement, incident-reporting (IR) systems often represent a source of frustration, rather than a useful tool for capturing important patient-safety and quality-of-care problems. IR systems suffer from well-known limitations.1 They detect only a small percentage of target problems,2 and the incidents that users do choose to report often include a large percentage of mundane events. Underuse of IR systems is particularly marked among physicians. In the survey reported by Farley et al(see page 416) in this issue, 86% of hospitals responded that physicians submitted “few or no” incident reports.3 This poor showing among physicians may reflect the misperception that incident reports fall under the jurisdiction of nurses and pharmacists. However, other reasons undoubtedly include the same factors that affect the use of IR systems by other healthcare professionals, including the time to fill out reports and the perceived utility of doing so.4

A more fundamental problem that bedevils the use of IR systems is that they generate numerators without denominators: X patients bled while receiving anticoagulants, and Y patients fell out of bed, without any indication of the total numbers of patients at risk for these events. In principle, hospitals could follow trends in these numerators on the assumption that the unknown denominators remain relatively constant over time. However, IR systems typically detect such small numbers of the targeted events that even small changes in reporting practices can produce large changes in the apparent incidence of events.

An incident during a recent rotation as the attending physician on an inpatient teaching unit illustrates the problem. Frustration with the nurses’ inattention to one of our patients elicited from me a grumble about the quality …