The vanishing nonforensic autopsy.

W e've all heard about cases in which a patient presumed to have died from acute myocar-dial infarction was discovered at autopsy to have had an aortic dis-section, or a patient who presented with decompensated liver failure from presumed alcoholic cirrho-sis but proved at autopsy to have widely metastatic hepatocellular carcinoma. Indeed, an extensive literature documents the frequency with which autopsy reveals clinically significant diagnoses that were missed before death. 1 Autopsies also generate more accurate vital statistics, provide pathological descriptions of new diseases, and offer powerful tools for education and quality assurance (see table). Yet despite these benefits, autopsies are performed after less than 10% of all U.S. deaths (see graph). 2 Moreover, national averages reflect high autopsy rates at a small proportion of hospitals; at the majority of nonacademic institutions, few or no autopsies are ever performed. The scientific, educational, and public health benefits of the autopsy , though generally acknowledged , remain difficult to quantify. However, autopsy plays a demon-strably important role in confirming or overturning diagnoses entertained by treating physicians. One classic study examined 100 autopsies performed in each of three decades at a Boston teaching hospital to determine the frequency with which autopsy identified missed diagnoses (e.g., the missed aortic dissection) that would have resulted in a change in therapy and might have prolonged survival if identified earlier. 3 These errors — categorized as class I errors — were found in approximately 10% of autopsies in all three decades. The authors also found class II errors — missed diagnoses that didn't affect survival but were nonetheless clinically important (e.g., the advanced hepatocellular cancer) — in an additional 12% in each decade. Physicians have generally attributed such results to selection bias, arguing that the cases physicians select for autopsy are those with the greatest diagnostic uncertainty. Such selection would leave the rates of diagnostic errors detected at autopsy relatively unchanged , despite true gains in diagnostic accuracy. But evidence suggests that physicians actually have little ability to accurately select cases for which autopsy will have the greatest diagnostic yield. Indeed, if selection bias were the main factor, one would expect studies from institutions with high autopsy rates to report substantially lower rates of errors in ante-mortem diagnosis. We found only modest evidence of such an inverse relationship in 40 years' worth of autopsy studies. 1 Controlling for autopsy rate, case mix, year, and country, we estimated …