The recent application of machine learning on biomedical images and signals and biomarkers in general has opened up avenues to explore and develop non-invasive diagnostic devices. The development of these classifier algorithms are mostly taken up in an early phase proof-of-concept study (POC). A proper validation of these devices must still be carried out in a pivotal trial based on a completely independent (of POC) data set before the device can be registered for market. In this talk, we explore statistical aspects of trial design for these pivotal trials such as primary endpoint selection, sample size and operating characteristics and adaptive strategies. In particular, we look at seamless and Bayesian strategies in order to minimize the logistical burden of such pivotal trials.
Rajat Mukherjee, Principal Consultant at Cytel, has over 20 years of professional experience as a statistician both in industry and academia. He has expertise in several areas of statistics including design and analysis of clinical trials, Bayesian clinical trials, adaptive designs, design and analysis of complex epidemiological studies, statistical computing, survival and longitudinal analysis, nonparametric and semiparametric inference, statistical classification, machine-learning and high-dimensional data.
At Cytel, Rajat leads a data science team working in the areas of diagnostic devices, biomarker discovery and real world evidence. The team also develops custom Bayesian software for design and analysis of trial and real world data. Rajat provides statistical consulting for adaptive strategies in trial design and machine learning experiments.