(014) Establish a Prediction Model for Treatment Response of Tumor Necrosis Factor Inhibitors in Patients with Rheumatoid Arthritis

Background: Over 30% of rheumatoid arthritis (RA) patients were non-response to tumor necrosis factor inhibitors (TNF inhibitors), which incurred considerable medical costs comparing to the responder in biologics.

Objectives: We aimed to develop two prediction models for the primary response at 6-month and secondary response at 12-month among TNF inhibitors users with disease activity score in 28 joint counts (DAS28) > 5.1.

Methods: This was a cross-sectional study using electronic health records data from 2014 to 2020 at National Cheng Kung University Hospital, Taiwan. We enrolled patients with RA who started first-line TNF inhibitors and had 2 consecutive records of DAS28 > 5.1 before the initiation of biologics. We excluded patients attending clinical trial or smaller than 18 years old. DAS28 were follow-up for primary response at 6-month and secondary response at 12-month, respectively. Input baseline predictors such as demographic data, laboratory results and medication history were reviewed from literature and selected by the rheumatologist. We used backward elimination method to select significant predictors with lowest Akaike's information criteria (AIC) value. Linear regression models were applied to predict DAS28 at 6-month and 12-month. We conducted sensitivity analyses to ensure the model’s robustness, which included removing predictors with missing rate > 50%, removing predictors with any missing rate and removing predictors to fulfil events per variable > 10.

Results: 122 patients were identified to initiate their first line TNF inhibitors with persistent high disease activity (DAS28 > 5.1). The 6-month prediction model had RMSE of 0.81751 and AIC of 78.36391. We found that erythrocyte sedimentation rate (ESR), patient global assessment (PGA), age, presence of osteoarthritis, C-reactive protein (CRP), and rheumatoid factor (RF) had positive influence on 6-month DAS28, while serum creatinine had negative influence on 6-month DAS28. The 12-month prediction model had RMSE of 0.95351 and AIC of 101.01237. We found that ESR, PGA, presence of dyslipidemia and aspartate transaminase (AST) had positive impact on 12-month DAS28, but tender joint count (TJC) had negative impact on 12-month DAS28. The results of sensitivity analyses were consistent with the main results.

Conclusions: In 6-month prediction model, ESR, PGA, age, presence of osteoarthritis, CRP, RF, and serum creatinine were associated with 6-month DAS28 (primary response). In 12-month prediction model, ESR, PGA-VAS, presence of dyslipidemia AST, and TJC were associated with 12-month DAS28 (secondary response).