PMS28 - The Effect of Medication Choice After Tumor Necrosis Factor Inhibitor (TNFI) use on the Healthcare Costs for Patients with Rheumatoid Arthritis: An Instrumental Variables Approach

Abstract

To measure differences in healthcare costs for patients with rheumatoid arthritis (RA) who initiated a targeted disease modifying anti-rheumatic drug (tDMARD), specifically tumor necrosis factor inhibitor (TNFi) therapy, and switched to another TNFi compared to a non-TNFi. Using 2010-2016 health insurance claims from IMS PharMetrics, RA patients who initiatied a TNFi were identified. We included patients who switched to another tDMARD and were located in a 2-digit ZIP code area with ≥10 patients in the data. The outcome of interest was post-switch monthly costs (total and RA-related) from inpatient, outpatient, and pharmacy claims. We first used an ordinary least squares (OLS) approach to determine costs associated with switching to a non-TNFi versus TNFi while adjusting for patient characteristics. To address the potential endogeneity of the switch-to medication selected, we applied an instrumental variable regression approach, which used variation by ZIP code in non-TNFi prescribing rates as an instrument for individual prescribing choice. Of the included 1,940 RA patients who initiated a TNFi therapy,1,467 (76%) switched to another TNFi and 473 (24%) switched to a non-TNFi. Across ZIP codes, the share of patients who switched from TNFi to non-TNFi ranged from 0 to 38.9%. Using OLS, switching to non-TNFi compared to TNFi resulted in lower total costs (-$426, p=0.035) and lower RA-related costs (-$475, p<0.001). Using the instrumental variables regression, the marginal effect of switching to non-TNFi reduced all-cause spending (-$2,532, p=0.047) and showed a larger, but not statistically significant decrease in RA-related cost (-$1,030, p=0.182). Estimates of the effect of medication choice after TNFi use on healthcare costs may be biased if providers prescribe selectively. Using an instrumental variables approach to control for this selection bias, we found that traditional methods may underestimate all-cause cost savings from switching to non-TNFi after use of a TNFi.