Methodological Innovations

Randomized controlled trials (RCTs) are the gold standard for estimating the effect of an intervention on an outcome but are resource intensive, requiring a substantial amount of time, money, and participants to complete. In many cases, there are external data sources that could be leveraged to improve the efficiency of RCTs. While leveraging external data sources can improve the efficiency of RCTs, combining RCT and external data sources can cause bias if the external data are systematically different than the RCT data. We have developed Bayesian methods for dynamic borrowing using commensurate priors, multisource exchangeability models, and other approaches that evaluate the consistency between the RCT and external data sources and only borrows from the external data if they are consistent with the RCT. This results in more efficient estimation, while limiting bias due to heterogeneity between sources. Current research in this area focuses on statistical methods for combining RCT and real world data.

Selected Publications:

Murray, T.A., Hobbs, B.P., Lystig, T.C., Carlin, B.P. “Semiparametric Bayesian Commensurate Survival Model for Post-Market Device Surveillance with Non-Exchangeable Historical Data.” Biometrics, 70, 185 - 191, 2014.

Kaizer, A.M., Koopmeiners, J.S., Hobbs, B.P. ``Bayesian Hierarchical Modeling based on Multi-Source Exchangeability.'' Biostatistics, 19, 169 - 184, 2018.

Kotalik, A., Vock, D.M., Donny, E.C., Hatsukami, D.K., Koopmeiners, J.S. ``Dynamic Borrowing in the Presence of Treatment Effect Heterogeneity.'' Biostatistics, 22, 789-904, 2021.

Not all subjects respond to the same intervention or treatment in the same way. That is, the anticipated effect of a treatment may differ based on patient characteristics which is often referred to as treatment effect heterogeneity. A secondary goal of many clinical trials is to identify (a) whether or not the treatment effect is heterogeneous and (b) identify subgroups for which the effect of treatment varies. Many approaches to do the former do not preserve type I error. The result may lead to development of personalized or individual treatment rules which are worse than one-size-fits all approaches. Our group has developed novel permutation-based approaches to preserve type I error control when identifying treatment effect heterogeneity and ensure that individual treatment rules improve upon one-size-fits all approaches with high probability.

Selected Publications:

Wolf, J.M., Koopmeiners, J.S., Vock, D.M. “A Permutation Procedure to Detect Heterogeneous Treatment Effects in Randomized Clinical Trials while Controlling the Type I Error rate.” Clinical Trials, 19, 512 - 521, 2022

Cain CH, Murray TA, Rudser KD, Rothman AJ, Melzer AC, Joseph AM, Vock DM. Design considerations and analytical framework for reliably identifying a beneficial individualized treatment rule. Contemp Clin Trials. 2022 Oct 12;123:106951. doi: 10.1016/j.cct.2022.106951. Epub ahead of print. PMID: 36241146..

A key decision in the drug development process is to determine the correct dose and dosing schedule for a new treatment. Determining the correct dose involves a tradeoff between efficacy and toxicity and is usually the focus of early phase clinical trials that have substantially smaller sample sizes than large, confirmatory Phase 3 clinical trials. Furthermore, while randomization is the standard for Phase 2 and 3 clinical trials, Phase 1 clinical trials, which may represent the first time that a treatment is evaluated in humans, often utilize dose escalation designs to protect patient safety. In this case, the decision to escalate or de-escalate is made adaptively, and depends on the observed outcomes for patients treated at lower dose-levels. Our investigators have developed novel approaches to early phase dose escalation clinical trials that account for the unique features of the data, including considering the trade-off between efficacy and toxicity in Phase 1-2 clinical trials, delayed outcomes, flexible models for dose-finding, and dose-finding for targeted cancer treatments. Current research in this area focuses on dose-finding in the presence of heterogeneous subgroups for personalized medicine.

Selected Publications:

Koopmeiners, J.S., Modiano, J. “A Bayesian Adaptive Phase I-II Clinical Trial for Evaluating Efficacy and Toxicity with Delayed Outcomes.” Clinical Trials, 11, 38 – 48, 2014.

Cunanan, K.M., Koopmeiners, J.S. ``Efficacy/Toxicity Dose-Finding using Hierarchical Modeling for Multiple Populations.'' Contemporary Clinical Trials, 71, 162 - 172, 2018.

Murray, T.A. “Logistic Retainment Interval Dose Exploration Design for Phase I Clinical Trials of Cytotoxic Agents.” Pharmaceutical Statistics, 20, 850 - 863, 2021

Zhou, H., Murray, T.A., Pan, H., Yuan, Y. Comparative review of novel model-assisted designs for phase I clinical trials. Statistics in Medicine. 37, 2208– 2222, 2018

Designs that adapt the randomization ratio to favor better performing arms based on interim data have intuitive appeal. However, their implementation must be done with care to control type I error and ensure an adequately powered trial. Practical limitations need to be accounted for as well, such as the feasibility for adapting in a fully versus group sequential manner. Because of these limitations, intensive consideration should be given to their actual benefit over a more conventional fixed randomization design. In a two-arm trial, the advantage is modest and thus may only be appealing in acute settings or where recruitment is likely to substantially improve. In multi-arm settings, these designs may engender greater advantages.

Selected Publications:

Proper, J., Connett, J., & Murray, T. (2021). Alternative models and randomization techniques for Bayesian response-adaptive randomization with binary outcomes. Clinical trials (London, England), 18(4), 417–426. https://doi.org/10.1177/17407745211010139

Proper, J., Murray, T.A. (2022). An alternative metric for evaluating the potential patient benefit of response-adaptive randomization procedures. Biometrics, 1– 13. https://doi.org/10.1111/biom.13673

Medical decision-making involves an inherent risk-benefit analysis between the potential for adverse effects and the potential for benefit. Utility-based designs for randomized clinical trials provide a framework to formally quantify and compare treatments based on these tradeoffs. These designs require elicitation of a consensus utility function from clinical investigators that quantifies the desirability of all joint realizations of the key clinical outcomes, typically a safety outcome and an efficacy outcome. Coupled with a flexible Bayesian probability model for the joint distribution of these outcomes, treatment comparisons may be carried out based on the posterior distribution for the expected utility under each treatment. Utility-based designs have been developed for bivariate categorical outcomes and semi-competing risks, and adaptively with pre-specified prognostic subgroups.

Selected Publications:

Murray, T.A., Thall, P.F., and Yuan, Y. (2016) Utility-based designs for randomized comparative trials with categorical outcomes. Statistics in Medicine, 35: 4285– 4305. https://doi.org/10.1002/sim.6989

Murray, T.A., Thall, P.F., Yuan, Y., McAvoy, S., & Gomez, D.R. (2017). Robust treatment comparison based on utilities of semi-competing risks in non-small-cell lung cancer. Journal of the American Statistical Association, 112, 11–23. https://doi.org/10.1080/01621459.2016.1176926

Murray, T.A., Yuan, Y., Thall, P.F., Elizondo, J.H. and Hofstetter, W.L. (2018), A utility-based design for randomized comparative trials with ordinal outcomes and prognostic subgroups. Biometrics, 74: 1095-1103. https://doi.org/10.1111/biom.12842