Utility of CDC record-linkage software for injury epidemiology

Background/Purpose: Record linkage methods are potentially useful for injury research. We therefore evaluated LinkPlus, a probabilistic linkage program available free from CDC. Methods: We obtained 2007 traffic crash data from NHTSA. We used LinkPlus and the dated but industry-standard Automatch to match motorist records from NASS-GES to those from FARS within geographic region using month, day of week, hour/minute of crash, age, sex, seating position, Vehicle Identification Number (truncated), vehicle make/model, and driver ZIP-code. Both programs assigned a weight to each record pair using the likelihood-ratio methods of Fellegi and Sunter. Using histograms, we selected high cutoff weights to accept pairs as probable matches, and lower weights to assign pairs for clerical review. Results/Outcomes: Both programs were reasonably easy to use, although LinkPlus was not documented as completely. Motorist fatalities were identified by 1,103 NASS-GES records and 35,701 FARS records. Of 1,103 possible matches, LinkPlus identified 1,013 and Automatch identified 1,010; of these, 1,004 were identical. A total of 1,035 record pairs were linked or assigned for clerical review by one or both programs; inconsistent outcomes mostly involved difficulty identifying situations where more than one GES record was a potential match to a FARS record using LinkPlus. Conclusions: LinkPlus is generally comparable to standard software for probabilistic record linkage, is not difficult to use, and currently costs nothing. Assuming that it is periodically upgraded, technical support is maintained, and the price does not increase excessively, LinkPlus may be valuable for public health research beyond its original purposes.