Model-based methods for case definitions from administrative health data: application to rheumatoid arthritis

Area of Science:

• Health Informatics
• Biostatistics
• Epidemiology

Background:

• Selecting accurate disease case definitions from administrative health data is crucial for epidemiological research.
• Validation studies are essential to assess the performance of different case definitions.
• Rheumatoid arthritis (RA) case definition selection from administrative data presents unique challenges.

Purpose of the Study:

• To propose and demonstrate a model-based method for selecting optimal disease case definitions for administrative health data.
• To empirically evaluate various case definition attributes for rheumatoid arthritis (RA) ascertainment.
• To compare the performance of univariate and bivariate models in case definition selection.

Main Methods:

• Utilized data from 148 rheumatoid arthritis (RA) case definitions across hospital, physician, and prescription medication administrative data.
• Employed separate univariate models for sensitivity and specificity, a univariate model for Youden's index, and a bivariate mixed-effects model.
• Investigated covariates including diagnosis counts, observation times, and medication records.

Main Results:

• Key case definition attributes improving sensitivity and/or specificity included multiple physician diagnoses and specialist diagnoses.
• The presence of RA-related prescription medication records significantly enhanced case ascertainment.
• Both univariate and bivariate models identified similar significant case definition criteria, with minor differences for Youden's index.

Conclusions:

• A model-based method offers empirical evidence to guide the selection of disease case definitions from administrative health data.
• The choice between univariate and bivariate models should align with the specific goals and scope of the validation study.
• This approach enhances the reliability of disease ascertainment in large administrative datasets.