Methods/Statistics
Transporting a Causal Effect of Treatment Assignment from a Trial to an External Target Population When Trial Participation Impacts Adherence Rachael Ross* Rachael Ross Amy J. Pitts Elizabeth A Stuart Kara E. Rudolph
Transportability methods can extend inferences from trial samples to real-world populations (external targets); however, complications arise when trial activities such as incentives and follow-up visits impact treatment adherence. Here we consider the setting where the trial includes data on covariates, treatment assignment, adherence, and outcome; the target includes data on covariates only. When trial participation impacts adherence, we can identify the joint effect of treatment assignment and scaling up trial activities that impact adherence (e.g., incentives) in the real-world treatment setting, but scaling up such trial activities is not often realistic. We propose an alternative estimand taken from mediation analysis (an interventional mediation effect): the effect of treatment assignment when the adherence distribution is set to the distribution of adherence in the target. Since adherence in the target is unknown, we define a parameter, delta, as the target:trial adherence ratio, which allows us to leverage the trial adherence data. Delta is unknown, but we can use expert knowledge to set bounds or a reasonable distribution. In the analysis, the estimand is repeatedly estimated using random draws from the proposed delta distribution similar to quantitative bias analyses. We provide a g-computation estimator and an augmented inverse probability weighted estimator and discuss variance estimation. We illustrate our approach in a case study to transport the comparative effect of extended-release naltrexone and sublingual buprenorphine for the treatment of opioid use disorder on relapse from a trial to a real-world Medicaid population. Adherence to treatment in the trial was much higher than previously reported in real-world treatment settings and in other trials. We use the proposed approach to transport the effect under more realistic adherence.