OBJECTIVES: Developing a control group of a clinical trial using real world data is desirable and ethically sound despite challenges pertaining to internal validity. To examine the internal validity, we reproduced the control group in a Randomized Controlled Trial (RCT) using Electric Health Record (EHR) data and evaluated the difference between the outcome of the original trial and that of the reproduced analysis.

METHODS: We selected an RCT, REMDACTA trial, that was performed to evaluate the efficacy of tocilizumab plus remdesivir against placebo plus remdesivir in patients with severe COVID-19 pneumonia. We reproduced its control group (patients with severe COVID-19 pneumonia taking only remdesivir), using Japanese EHR data, Millennial Medical Record provided by Life Data Initiative. Target patients for the main analysis were those prescribed remdesivir within 2 days after admission and diagnosed with COVID-19 (defined by ICD-10 code, U07.1) and/or with COVID-19 pneumonia (defined by diagnosis name). Patients in the sub analysis included only those with COVID-19 pneumonia diagnosis. Among the target patients, those undergoing image inspection, oxygen administration, and not taking any medicines for pneumonia before the first remdesivir prescription were eligible for the analyses. Median length of stay was compared in the reproduced group and in REMDACTA trial.

RESULTS: The database included 676 and 110 target patients for the main and sub analyses, respectively. However, only 57 and 14 patients met the eligibility criteria for the main and sub analyses, respectively. The reduction in the number of patients is attributed to the criteria of image inspection and oxygen administration. Median length of stay in the reproduced group were 13 and 11 days in the main and sub analyses, respectively. In REMDACTA trial, 95% CI of median time was 11.0–16.0.

CONCLUSIONS: We successfully reproduced the median time of the control group by EHR data.