Bayesian Hierarchical Approach for Handling Non-ignorable Drop-out Across Multiple Clinical Trials in Schizophrenia

Clinical trials in Schizophrenia assess symptom severity using a clinician-rated scale like Positive and Negative Syndrome Scale (PANSS), measured over time. However, patients taking psychiatric medication have shown higher variability of response compared to patients taking medication related to a physical disorder. Within randomized trials, it has also been shown that the dropout rates can be quite large and vary between treatment groups, thus possibly introducing non-ignorable missingness or missing not-at-random (MNAR). If we combine such RCTs to evaluate treatment efficacy under individual patient-level (IPD) network meta-analysis (NMA) with non-ignorable dropout, we could be introducing bias in the estimation of the treatment effects. To address these challenges and maximize use of all available data, we aim to combine a popular method for addressing MNAR like pattern-mixture with Bayesian IPD NMA to improve the estimation of the treatment effects. Through simulations, we examine the impact of our approach under varying data availability conditions and complexity. We then apply our methods to clinical trials for schizophrenia treatments, demonstrating their effectiveness in handling non-ignorable dropout.

Bayesian Hierarchical Modeling to Handle Systematically Missing Outcome Data in Meta-Analysis with Individual Patient-Level Data

In the interest of conducting pairwise meta-analysis with individual patient-level data, researchers combine randomized control trials that not only compare the same treatments, but also overlap in reported outcomes. However, it is often the case that some trials may not have overlapping outcomes, which causes one outcome to be prioritized and studies not observing such outcome to be removed from the analysis. To address this, we propose a Bayesian hierarchical model that simultaneously considers all reported outcomes where at least one study includes all outcomes of interest. Through simulations, we explore the implications of our approach in scenarios with varying data availability and highlight its inherent constraints. Subsequently, we apply our proposed model to a MA of treatments for major depressive disorder, where discrepancies among reported outcomes are evident.

Using Network Meta-Analysis (NMA) in the estimation of Probability of Success (PoS)

Go/no-go decisions punctuate the drug development lifecycle. These decisions are made based on limited and often imperfect data, and often prove incorrect in hindsight: success rates of even Phase-III trials are quite low in many indications (Wong et al, 2019). To support accurate internal decision-making prior to Phase-III, Hampson et al (2021) describe a novel framework for synthesizing information from several sources (available Phase-III data, historical “benchmark” data about success rates of similar clinical programs, the likelihood of safety issues, and other potential risks) into a Bayesian model that predicts Phase-III outcomes, leading to an estimate of the overall probability of success for the program. However, in many indications, the definition of success is inherently relative to a small handful of other compounds in the competitive landscape. This internship project will explore connections between the probability-of-success (POS) framework described by Hampson et al (2021) and network meta-analysis modelling (e.g. , allowing predictions from the framework to feed into to indirect comparisons with specific competitor drugs and further support decision making.

Bayesian Bias-Adjustment Models in Network Meta-Analysis of COVID-19 trials

Clinical trials involving treatments for COVID-19 have shown varying levels of rigor and consistency, but very few studies have addressed the potential bias in estimating the treatment effect from these trials. A large body of literature have shown that including trials at risk in a network-meta analysis (NMA) could result in biased treatment effect estimates. Network meta-analysis combines multiple trials to create evidence about the comparative effectiveness of multiple treatments. In this presentation, we will introduce Bayesian bias-adjustment methods in NMA under contrast-based and arm-based frameworks to estimate bias-corrected treatment effects. The risk of bias of a trial is classified into three groups: low, some, or high concerns using the Cochrane risk-of-bias tool. Our method proposes a probabilistic model to incorporate uncertainty from studies given ‘some concerns’. We present an extensive simulation study to evaluate model performance and illustrate our methods using NMA of COVID-19 trials. The results present the impact of including studies with risk of bias in NMA and how they should be interpreted.