Bayesian sequential design for phase II trials with applications in bariatric and metabolic surgery

• Applicant: Dr. Riko Kelter
• Subject Area: Epidemiology and Medical Biometry/Statistics ; Mathematics
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 549296018

Sequential hypothesis testing is an important statistical method to improve the efficiency of clinical trials. In contrast to trials with fixed sample size, in a sequential analysis data are evaluated as they are collected and sampling is stopped in accordance with a pre-defined stopping rule once a significant result is observed. Various sequential designs have been proposed for phase II trials, where usually the endpoint is binary and measures success or failure of a novel drug in patients. A prominent example of this setting is the treatment of obesity, where weight loss after bariatric and metabolic surgery such as gastric bypass is classified into excessive (success) and non-excessive (failure). In this project, we combine Bayesian statistics and the theory of artificial neural networks to develop novel sequential phase II clinical trial designs for binary endpoints in the context of bariatric and metabolic surgery. The goal is to invent designs which yield improved operating characteristics, such as smaller expected sample sizes and larger probabilities of early termination of a trial. This translates to significant advantages in medical research: Fewer patients are required until a conclusion is reached and ineffective treatments are identified with a larger probability. We focus on the medical context of bariatric and metabolic surgery, which plays an increasingly important role in the treatment of obesity. Here, the recent advent of non-surgical treatments raises the question whether these can be used as add-on therapies after bariatric surgery, to achieve additional weight loss. This situation calls for efficient sequential phase II trial designs to investigate their efficacy. To this end, we first develop a streamlined theory for sequential testing in the Bayesian paradigm. Then, we generalize standard Bayesian sequential trial designs by introducing calibration functions which are inspired by the theory of artificial neural networks. We construct general-purpose calibration algorithms for these novel designs, and develop a software implementation. The whole process happens in close collaboration with clinicians from a certified adipositas competence center to allow for an implementation which is guided by clinical data. The resulting software package will enable clinicians to plan, calibrate, monitor and analyze a sequential phase II trial with the novel designs. The benefits of fewer required patients and larger probabilities of early termination of a trial are relevant in, but not exclusive to the context of bariatric and metabolic surgery related trials. This extends the methodology developed in this project to other possible application contexts.

• DFG Programme: Research Grants
• Co-Investigators: Privatdozent Dr. Sebastian Dango; Professor Dr. Alexander Schnurr