Aim 2: Impute DMTs with incomplete RX data

Impute interventions with incomplete codified DMT data

Focusing on existing validated EHR-derived endpoints, we will develop strategies to correct for imperfect medication codes of DMTs. We propose :

• A surrogate assisted unsupervised algorithm to improve the time-varying DMT status definition by combining information from codified and narrative EHR data;

• And a calibrated causal treatment effect estimation that corrects for error-prone treatment arm definitions via semi-supervised learning.

• We will perform real-world studies to compare the effects of DMTs for RA on controlling inflammation and maintaining remission using EHR data at two healthcare sites, correcting for imperfect prescription data.