In outcome analyses of clinical trials and mortality follow-up studies, the underlying cause of death (UCOD) is commonly assigned either by official coders or by a panel of physicians. We evaluated the validity of UCOD assigned by official coders by comparison with the assignments of a panel of physicians who reviewed the available medical records of the deceased. The comparisons focused on deaths occurring from October 1995 through June 1998 in a series of residents in a veterans home. Because

more »

... of limited resources, only the first 104 deaths that occurred during the study period were included. Agreement rate, sensitivity, specificity, and kappa statistics were calculated to assess the consistency of coder versus physician panel assignment of UCOD by selected main causes of death. For 32 of the 104 deaths, the panel concluded that the information obtained from medical records was insufficient to determine the UCOD, and the following analyses were confined to the other 72 deaths. For the 72 deaths considered by the panel to have sufficient information to determine UCOD, all four physicians agreed on a single UCOD in 50 (69%) cases, while three or four agreed in 66 (92%) cases. A consensus was reached in cases with disagreement. The two procedures completely agreed in 40 (56%) of the deaths. For general category UCOD, the kappa value was high for cancer (0.83) and cardiovascular disease (CVD, 0.73) but only moderate for pulmonary disease (PD, 0.60). When the UCOD assigned by the panel was used as the gold standard, official coders showed relatively low sensitivity for correct determination of UCOD in cases of CVD (0.76) compared with cancer (0.86) and PD (0.80). Given the high inter-physician consistency and the relatively low sensitivity of official coders in assigning CVD as the UCOD, we conclude that the use of clinical review panels would provide more accurate UCOD assignments for use in outcome analyses in mortality follow-up studies and clinical trials in Taiwan.