Abstract:
Introduction: Diagnostic validation studies are used to assess the accuracy of administrative health data by testing case definitions. Many researchers use descriptive analyses to recommend a case definition.
Purpose: The purpose was to develop and assess model-based methods to select a case definition for identifying individuals with a chronic disease in administrative health data.
Methods: A simulation study was used to compare the performance of univariate and bivariate models applied to diagnostic validity measures. The models were demonstrated using analysis of 148 case definitions from a rheumatoid arthritis (RA) validation study.
Results: All models performed well based on bias and mean squared error; however, the bivariate model had poor confidence interval coverage. The RA characteristics that showed association with sensitivity or specificity were number of physician diagnoses, observation time, number of specialist diagnoses, and number of prescriptions.
Conclusion: These models provide researchers with an inferential method for recommending case definitions.
