OBJECTIVE:  To compare the results of different methods of injury surveillance using multiple linked sources of data.

METHODS:  For the years 1995-96, cases of death and/or hospitalization resulting from motor vehicle crashes were obtained for one state from the Crash Outcome Data Evaluation System (CODES), the Fatality Analysis Reporting System (FARS), and an independent probabilistic data linkage project based upon a statewide trauma registry (TRAUMA).  Case ascertainment among the three linked databases was compared using deterministic and clerical methods.

RESULTS:  2267 cases of hospitalization were identified: 1171 in both CODES and TRAUMA, 598 only in CODES, and 498 only in TRAUMA.  388 distinct cases of death were identified, of which 329 appeared in all three databases, none in CODES only, 10 in FARS only, 31 in TRAUMA only, 3 in CODES and FARS, 1 in CODES and TRAUMA, and 14 in FARS and TRAUMA.  Discrepancies appeared to be attributable to differences in case definition rather than linkage methodologies.

CONCLUSIONS:  Case definitions for injury surveillance must be carefully specified.  Further improvements in probabilistic record linkage may make this approach easier for large population studies.