All Times EDT
Keywords: Probabilistic Programming, AI, Bayesian data analysis
Under today’s industry standard, statisticians analyze clinical trial data only in late phases of the trials themselves. Because of this, avoidable issues are detected late, leading to failed trials and exploding costs. What if such analysis was available from the earliest stages of a clinical trial, allowing for course-correction and swift, successful completion of a trial - all without having to hire an army of statisticians?
Introducing automated Bayesian methods in clinical trials makes that possible. Recently, researchers have developed automated Bayesian data modeling systems that can give Bayesian answers to data exploration, data cleaning, statistical inference, and predictive modeling problems (Mansinghka et al., 2016; Vergari et al., 2018; Saad et al., 2019). Researchers in probabilistic programming have added easy-to-use languages for querying these models (Mansinghka et al., 2015; Saad & Mansinghka, 2016). Programs in those languages allow users to detect issues within clinical trials with a fraction of the time of hand-coding. The modeling is data-driven, yet qualitative and quantitative constraints can be supplied by the domain experts if needed.
How does it work? Automated data modeling fits Bayesian models onto incoming trial data in real time. Data engineers and statisticians can write short programs in an English-like programming language, to query the automatically learned models. This is more versatile than the Bayesian methods used by experts today, in that one can detect different issues without having to re-train the model. Within minutes, issues can be identified with statistical models, including site performance problems, data entry errors and violations of inclusion/exclusion criteria. This round table, featuring experts from industry and academia, will discuss the progress to date and hurdles that need to be overcome for these techniques to become routinely applied to reduce the cost and improve the quality of clinical trial oversight.