Supplementing Claims Data with Electronic Medical Records to Improve Estimation and Classification of Rheumatoid...

Area of Science:

• Rheumatology
• Health Informatics
• Data Science

Background:

• Estimating rheumatoid arthritis (RA) disease activity solely from insurance claims data has shown limited performance.
• Electronic Medical Record (EMR) data offers a potentially valuable, readily accessible data source for enhancing such estimations.

Purpose of the Study:

• To evaluate if incorporating EMR data into claims-based models improves the accuracy of rheumatoid arthritis disease activity estimation.
• To compare the performance of "Claims-Only," "Claims + Medications," and "Claims + Medications + Labs" models.

Main Methods:

• Utilized linked Medicare claims and EMR data from a subset of the Brigham and Women's Hospital Rheumatoid Arthritis Sequential Study (BRASS).
• Employed adaptive LASSO for variable selection.
• Assessed model performance using adjusted R2 for continuous Disease Activity Score in 28 joints with C-reactive protein (DAS28-CRP) and C-statistics for binary classification (high/moderate vs. low disease activity/remission).

Main Results:

• The "Claims + Medications + Labs" model achieved an adjusted R2 of 0.18 for continuous DAS28-CRP, a significant improvement over "Claims-Only" (0.02) and "Claims + Medications" (0.09).
• The "Claims + Medications + Labs" model yielded a C-statistic of 0.76 for binary classification, outperforming "Claims-Only" (0.61) and "Claims + Medications" (0.68).

Conclusions:

• Integrating EMR-derived variables (medications and laboratory data) with claims data offers a modest improvement in estimating rheumatoid arthritis disease activity.
• While continuous DAS28-CRP estimation remains challenging, the combined claims-EMR approach demonstrates reasonable accuracy in discriminating between binary categories of disease activity.