Causal Inference Methods to Estimate Treatment Effects in Clinical Trials
Chair: Sudhakar Rao (Janssen) and Xiang Li (Johnson & Johnson)
Speakers: Sanne Roels, PhD (Johnson & Johnson)
Title: Causal Inference in Clinical Trials, an Application
Abstract: Causal inference methodology has historically been applied to data from obtained from epidemiological studies, which are often characterized by a lack of control on exposure (A=1, 0). In randomized controlled trials (RCT) exposure (A) is controlled via randomization. In the causal inference literature, the average treatment effect (ATE) is defined using the potential outcomes framework (Ya: Y1 and Y0, i.e. the potential outcome under A=1 and A=0, respectively): ATE = E(Y1- Y0). As one can only observe one of the two potential outcomes, identification of the ATE requires several assumptions, some of which are more easily satisfied in randomized (e.g., RCT) settings compared to non-randomized (e.g., epidemiological) settings. The causal inference framework helps to make explicit the assumptions necessary for estimating the ATE in RCTs. Here, we demonstrate its use in precisely formulating the causal question of interest in RCTs. We consider realistic settings, distinguished by the differing prognostic value of baseline covariates. For the estimation of the ATE we use TMLE and G-computation. We demonstrate and discuss the operating characteristics in these settings.
Speaker: Kelly van Lancker, PhD (Ghent University)
Title: Automated, Efficient and Model-Free Inference for Randomized Clinical Trials Via Data-Driven Covariate Adjustment
Abstract: In May 2023, the U.S. Food and Drug Administration (FDA) released guidance for industry on “Adjustment for Covariates in Randomized Clinical Trials for Drugs and Biological Products”. Covariate adjustment is a statistical analysis method for improving precision and power in clinical trials by adjusting for pre-specified, prognostic baseline variables. Though covariate adjustment is recommended by the FDA and the European Medicines Agency (EMA), many trials do not exploit the available information in baseline variables or only make use of the baseline measurement of the outcome. Specifically, practical implementation of covariate-adjusted estimators has been hindered by the regulatory mandate to pre-specify baseline covariates for adjustment, leading to challenges in determining appropriate covariates and their functional forms.
We will explore the potential of automated data-adaptive methods, such as machine learning algorithms, for covariate adjustment, addressing the challenge of pre-specification. Specifically, our approach allows the use of complex models or machine learning algorithms without compromising the interpretation or validity of the treatment effect estimate and its corresponding standard error, even in the presence of misspecified outcome working models. Our proposed estimators either hinge on achieving ultra-sparsity (which can be relaxed by limiting the number of predictors in the model) or necessitate integration with sample splitting to enhance their performance. As such, we will arrive at simple estimators and standard errors for the marginal treatment effect in randomized clinical trials, which exploit data-adaptive outcome predictions based on prognostic baseline covariates, and have low (or no) bias in finite samples even when those predictions are themselves biased.
We provide a detailed methodology overview as well as empirical study results. The findings offer a promising avenue for improving the statistical power of trial analyses through automated covariate adjustment.
Speaker: Susan Gruber, PhD (Putnam Data Sciences, LLC)
Title: Targeted Learning of Treatment Effects in Randomized Controlled Trials with Intercurrent Events
Abstract: In an ideal randomized controlled trial (RCT) baseline randomization ensures that unadjusted estimates of causal effects are unbiased. However, once underway, RCTs are subject to intercurrent events including treatment non-compliance, patient mortality, and changes in background conditions over the duration of the trial. These sometimes unavoidable events preclude accurately capturing the outcome under the treatment of interest. Planning how to appropriately address intercurrent events should be done in advance. Strategies include revising the definition of treatment, incorporating the event into a composite outcome, viewing the event as right censoring, or viewing it as a competing risk. Some options necessitate capturing additional information over time, and then using a statistical methodology appropriate for analyzing longitudinal data. We illustrate how the Targeted Learning Estimation Roadmap facilitates consideration of intercurrent events, and present results from a re-analysis of data from the 2004 Acupuncture for Chronic Headache in Primary Care pragmatic trial (Vickers, et al) using longitudinal targeted minimum loss-based estimation (L-TMLE).
Speaker: Jens Magelund Tarp, PhD (Novo Nordisk A/S)
Title: Randomized Trials Analyzed as Randomized Trials: Using Trials to Emulate Target Trials
Abstract: Estimation of treatment effects in randomized controlled trials beyond the intention-to-treat can be challenging as randomization may no longer protect against selection bias. Issues like adherence to randomized treatment, withdrawals, competing risks, and initiation of other treatments (drop-in treatments) after randomization, which may be unequally distributed in the randomization arms, will essentially make the trial a mixture of randomized and observational data. This opens for causal inference methods to analyze a randomized trial like a target randomized trial. The presentation will dive into a joint project between University of Copenhagen, University of Berkeley, and Novo Nordisk using targeted maximum likelihood estimation to target research questions originating from several ongoing long-term cardiovascular outcomes trials. An example from re-analysis of a previous outcomes trial to estimate the effect of randomized treatment while adjusting for a drop-in treatment will be given.
