The Computer-Assisted Postmortem Identification (CAPMI) system: sorting algorithm improvements.

Abstract

Refinements to the original Computer-Assisted Postmortem Identification (CAPMI) software algorithms and general data handling were suggested as a result of observations made following the Gander plane crash of 1985. The presence of highly fragmented and scattered remains following most plane crashes suggested that changes to procedure might improve CAPMI performance for use in these types of disasters. A total of 162 ante- and postmortem dental records which had been used successfully to identify victims of the Gander disaster were coded for anonymity and used for this investigation. Changes in data construction and management were made to CAPMI, according to concepts which were thought might improve system performance, and tested. Although most tested techniques improved CAPMI performance, the data suggested that replacement of "virgin" chartings with "data unknown" results in improved performance of CAPMI largely independent of other factors. Of 162 possible record matches, the original algorithm successfully listed the true record match in the top 20 possibilities 74% of the time; the tested variations on the original algorithm yielded results across a range of 38 to 83% successes, with most techniques performing better than the original algorithm. Results of this investigation have been incorporated into improved CAPMI procedures and software.