OBJECTIVES: This research aims to examine the use of electronic health record (EHR) data to derive effectiveness evidence.

METHODS: EHR data collected from January 1, 2011 and February 28, 2016 (control cohort) was combined with individual patient data (IPD) from two single-arm phase II studies (intervention cohort) to estimate treatment effectiveness. The data was harmonized by applying the inclusion and exclusion criteria from the trial to the EHR database. Prognostic variables were selected a priori and three different methods (propensity score matching (PSM), inverse probability of treatment weighting (IPTW), and Genetic Matching (GenMatch)) were applied and compared to address imbalances in measured confounders. A multivariate Cox proportional hazards model was used to compare OS. The hazard ratios (HR) were evaluated for each method. A sensitivity analysis evaluating the survival of the control cohort from EHR was conducted with an indirect comparison between the control cohort and digitized trial data from the control treatment.

RESULTS: After applying the inclusion and exclusion criteria, the intervention cohort (n=183) and control cohort (n=72) were imbalanced in terms of measured confounders. The PSM method did not balance measure confounders (standardize mean differences (SMDs ≥25%) and the IPTW and GenMatch method improved imbalance (SMDs <10%) between the intervention and control cohort. The observed treatment effect on the risk of death of the PSM (HR=0.59, 95% confidence interval (CI) 0.36-1.74), IPTW (HR=0.64, 95% CI 0.48-0.88) and GenMatch adjusted analyses (HR=0.54, 95% CI=0.48-0.62) was similar favoring the intervention cohort. Median OS was similar between the EHR control cohort (15.6 months) and clinical control cohort (14.9 months).

CONCLUSIONS: Our results demonstrate the utility of EHR data to estimate comparative effectiveness in a single-arm trial setting. PSM, IPTW, and GenMatch were used to address measured confounding; however the PSM result did not perform well due to small sample size.