Causal inference methods to assess safety upper bounds in randomized trials with noncompliance
*Yiting Wang, Janssen Research & Development, LLC 


Premature discontinuation or other forms of noncompliance with treatment assignment can complicate causal inference of treatment effects in randomized trials. The intent-to-treat (ITT) analysis gives unbiased estimates for causal effects of treatment assignment on outcome, but can understate potential benefit or harm of actual treatment (administered as intended). The corresponding upper confidence limit can be also underestimated. We used simulations to compare estimates of hazard ratio and upper bound of the 2-sided 95% confidence interval from causal inference methods that account for noncompliance in the form of treatment discontinuation and cross-over with those from ITT, per protocol and as-treated analyses. The simulation parameters were chosen to reflect interest in cardiovascular safety trials of diabetes drugs, with a focus on upper bound estimates relative to 1.3. We will present the main results, and discuss practical implications and extension to observational studies.