OBJECTIVE: To measure interobserver agreement on diagnoses classified and coded by family physicians using manual or computerized input modes. METHOD: Used increasingly in a variety of information management systems, the International Classification of Primary Care is the system best adapted to primary care. Ten physicians independently viewed 44 taped medical visits. Five physicians were randomly assigned to manual coding and five to computer coding. The study of reproducibility explored three aspects: written diagnoses, manually coded diagnoses, and diagnoses coded using a software program. The K statistic was calculated in order to compare interobserver agreement. RESULTS: Descriptive analysis of interobserver agreement in the written diagnoses revealed an agreement rate of 70.5% (+/- 6.3). Among physicians using manual coding, the agreement rate was 70.2% (+/- 7.2). In the group using the software program, the agreement rate was 75.0% (+/- 8.7). The K coefficients were low, but three were significant with critical ratios (z) above 1.96. CONCLUSION: Results suggest that input method has no bearing on interobserver agreement and that agreement is more a function of clinical presentation of health problems than of coding process.
RCT Entities:
OBJECTIVE: To measure interobserver agreement on diagnoses classified and coded by family physicians using manual or computerized input modes. METHOD: Used increasingly in a variety of information management systems, the International Classification of Primary Care is the system best adapted to primary care. Ten physicians independently viewed 44 taped medical visits. Five physicians were randomly assigned to manual coding and five to computer coding. The study of reproducibility explored three aspects: written diagnoses, manually coded diagnoses, and diagnoses coded using a software program. The K statistic was calculated in order to compare interobserver agreement. RESULTS: Descriptive analysis of interobserver agreement in the written diagnoses revealed an agreement rate of 70.5% (+/- 6.3). Among physicians using manual coding, the agreement rate was 70.2% (+/- 7.2). In the group using the software program, the agreement rate was 75.0% (+/- 8.7). The K coefficients were low, but three were significant with critical ratios (z) above 1.96. CONCLUSION: Results suggest that input method has no bearing on interobserver agreement and that agreement is more a function of clinical presentation of health problems than of coding process.