Variations in case definition algorithms in diabetes-related studies using Taiwan National Health Insurance claims data

Objectives: This descriptive study examined the variations in case definition algorithms in diabetes-related studies using Taiwan National Health Insurance claims data. Methods: We searched the PubMed database to retrieve relevant papers published between 2001 and 2020. The components of a case definition algorithm included I) diagnostic codes, 2) data used, 3) minimum number of visits with diagnostic codes, and 4) time intervals required. We grouped the algorithms according to their positive predictive value (PPV) derived from a published validity study. Results: We identified 611 studies with 30 distinct case definition algorithms and classified them into 8 groups. The PPV was lowest in Group 1 and highest in Group 8. The three most frequently used algorithm appeared in Group 8, ("diagnostic code AND antidiabetic drugs prescribed," 194 papers, 32%), followed by Group 1 ("at least one outpatient diagnostic code," 119 papers, 20%) and Group 3, ("at least two outpatient diagnostic codes," 111 papers, 18%). The number of papers in Groups 7 and 8 that used antidiabetic drugs as a condition for case definition increased prominently, from 18 between 2001 and 2011 to 89 between 2018 and 2020. Furthermore, 86 papers used the more rigorous definition "diagnostic codes AND medications." However, the proportion of papers in Groups 1 and 2 decreased from 31% between 2001 and 2011 to 14% between 2018 and 2020. Conclusions: The number of diabetes-related studies using more rigorous (higher PPV) case definition algorithms increased between 2001 and 2020. Additional studies, which are requested through the reporting of studies conducted using observational routinely collected health data (RECORD), on the validity of these algorithms are required.