Lead Investigator: Arman Sabbaghi, Purdue University
Title of Proposal Research: Tripartite Estimands for Adherence Causal Inference in Clinical Trials, With Applications for the Development of COVID-19 Treatments
Vivli Data Request: 7658
Funding Source: None
Potential Conflicts of Interest: Run Zhuang is currently working on funded research projects with Eli Lilly and Company and Janssen Pharmaceuticals as part of his doctoral research. He will not share any of the real-life data provided by Sanofi with his research collaborators from Eli Lilly and Company or Janssen Pharmaceuticals.
Summary of the Proposed Research:
In most clinical trials, patients are randomized to an experimental treatment and a control treatment in order to assess the efficacy of the former. Clinical trials generally have a prescriptive set of procedures for the conduct of the trial, which is called a protocol. For ethical and medical reasons, patients may not be able to follow the defined protocol and dosing.
Specifically:
1. patients may have adverse events during the trial,
2. patients or their physicians may decide to discontinue their study medication, regardless of whether they are experiencing any benefit from the treatment,
3. patients or their physicians may feel that there is insufficient improvement in their condition (or even worsening) and likewise decide to terminate further use of their study medication,
4.personal circumstances that may keep a patient from continuing to participate in a clinical trial (e.g. moving out of town, family matters that take precedence such as the illness of a family member), or
5.patients may not be able to adhere to the protocol schedule – they miss visits, they forget to take their medication, they withdraw consent to take a battery of tests during the trial.
Each of the disruptions to the planned clinical trial protocol as described above (as well as others that can occur in clinical trials) can obscure or confound the effect of both experimental and control treatments being studied.
Estimating the so-called direct treatment effects is quite difficult and not as direct due to the biases that must be accounted for from the above disruptions. The proposed Tripartite Estimands framework circumvents the biases incurred by incorporating non-adherence to the above disruptions into the model for the outcomes of the different treatments, yielding results that capture a more comprehensive view of the treatment effects.
While the Tripartite Approach can be useful in many clinical trial settings, it may have particular utility for the thousands of ongoing clinical trials of new medications NOT for COVID-19 that have experienced substantial disruptions due to the COVID-19 epidemic. For example, many trials were ongoing at the beginning of 2020 (in diabetes, pain, cancer, rheumatoid arthritis, asthma, etc.), and when the pandemic hit, many of the patients in these studies either
1. could not get to the study site for treatment due to quarantine or the site being closed
2. could not get their study medication because supply chains were severely disrupted or
3. actually contracted COVID-19 and had to discontinue their study medication or drop-out of the study in which they were enrolled.
Statistical Analysis Plan:
Briefly, the Tripartite Framework will follow the analysis methodology as published in [Qu, Y., Fu, H., Luo, J., Ruberg, S.J. (2020) A General Framework for Treatment Effect Estimators Considering Patient Adherence. Stat Biomedical Research (in press) – Online https://doi.org/10.1080/19466315.2019.1700157]. In that research, a causal framework and an inferential model are used to estimate treatment effects – the former using counterfactual models and the latter using inverse probability weighting. Computer code has been written to perform such analyses, and an example is presented briefly in that publication. This research effort is to replicate that or similar causal inference models to additional real clinical trial datasets.
Furthermore, additional statistical methods will be explored by incorporating mixture models in a Bayesian Framework. We hope to be able to extend the Frangakis and Rubin (2002) approach to the Tripartite estimands to assess the causal effects of treatments in randomized clinical trials.
This study was selected due to it drug-adherence of around 85%. Specifically, 84.5% of main meal arm and 89.4% of the breakfast arm completed the study. Our proposed research involving the Tripartite Estimands approach evaluates treatment differences in the proportion of patients with adverse events, the proportion of patients due to lack of efficacy, and the primary efficacy outcome for those who can adhere to study treatment under the causal inference framework. The goal is to compare and contrast the results of standard ITT estimates to those obtained from the Tripartite Framework, and to evaluate the circumstances when one might be better suited for describing a treatment effect than the other.
We will use Principal Stratification to estimate the relevant proportions. Measures of interest would be the treatment differences in the proportion of patients with adverse events, the proportion of patients due to lack of efficacy and the primary efficacy outcome for those who can adhere to study treatment under the causal inference framework. Covariates will be standardized before proceeding with analysis. The statistical approaches will be both Bayesian and frequentist.
Requested Studies:
A 24-week, Open-label, Randomized, 2-arm Parallel Group, Multinational, Multi-center Clinical Trial to Compare the Efficacy and Safety of Lixisenatide Injected Prior to the Main Meal of the Day Versus Lixisenatide Injected Prior to Breakfast in Type 2 Diabetic Patients Not Adequately Controlled on Metformin
Data Contributor: Sanofi
Study ID: NCT01517412
Sponsor ID: EFC12261
Public Disclosures:
Jiang S., Liu B., Nei L., Sabbaghi A., Abdul Wahab A.H., Qu Y., Lyu T., Zhuang R. Is good enough good enough: A comprehensive evaluation of ICH E9 (R1) Strategies. ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop. PS5C. 2023.