(P-339) Methods and Practical Considerations in Use of an External Control to Assess Potential Drug Risk

Background: Increasingly, external comparators for clinical studies derive from synthetic control methods (Thorlund, 2020). Valid safety and efficacy comparisons require understanding the practical challenges and potential pitfalls in operationalizing these methods.

Objectives: Describe the approach taken to ensure comparability of actively treated synthetic controls from a rheumatoid arthritis (RA) disease registry (Corrona) and the tofacitinib (RCT) arm of a placebo controlled clinical trial for safety analyses. Results to be reported separately upon study completion.

Methods: RCT and Corrona teams determined compatibility of data elements, definitions, and validation methods.

Setting:
RCT: The Phase 1-3 and long term extension (LTE) RA clinical trial program.
Corrona: Corrona prospective cohort observational study of RA patients.

Exposure:
RCT: Subjects who received at least 1 dose of tofacitinib during the trials.
Corrona: Participants who initiated bDMARDs within RCT time frame and met trial eligibility criteria.

Duration of follow up: Follow-up was while on drug. Differential follow-up due to differences in switching and retention rates in the data sources was assessed.

Main outcome measures:
Incident malignancy excluding non-melanoma skin cancer and incident major adverse cardiovascular events; operationalized differences were assessed and validated between RCT and Corrona.

Analysis:
Propensity score was estimated based on a priori chosen factors and patient characteristics that differed by a standardized difference of ≥0.1. The risk analysis used the score in two ways:
To select patients on common support and adjust risk estimate using a Cox regression model (primary analysis).
To assess risk among 1-1 propensity matched patients.
Sensitivity analyses using multiple imputation and inverse probability of censoring weighting were carried out.

Results: This approach resulted in comparable populations for valid and robust analysis of outcome risk. The primary and multiple secondary analyses allowed for assessment of the consistency of the risk estimates.

Conclusions: The viability of synthetic controls in drug safety evaluations is contingent on 1) data quality, 2) data alignment and 3) appropriate analytic methods. Collaboration between clinical and observational researchers is critical to ensure data validity. Further, analytic approaches must consider the differences in the study populations and the datasets themselves.

Reference: Thorlund K, et al. Synthetic and External Controls in Clinical Trials- A Primer for Researchers. Clinical Epidemiology, 2020; 12: 457-467.