All Times EDT
Keywords: stochastic recruitment model, Bayesian hierarchical model, data science, collaboration, agile
Contemporary clinical trials often have complex logistics: they are run across multiple centers/countries and involve a lot of uncertainties about disease prevalence rates, patient characteristics and regulatory requirements, all of which can vary across countries and centers within a country. As a result, planning and delivering such studies on-time has long been recognized as a challenge in the pharmaceutical industry. The key to the accurate prediction is properly accounting for all the sources of uncertainty mentioned above. One well-known approach to modeling recruitment in complex trials which has been gaining traction steadily is Poisson-Gamma stochastic model. It's one of the applications of co-called predictive modelling approaches -the term that has been firmly embedded into the field of statistical innovation over the course of the past few years. While the methodology based on Poi-Gamma stochastic process is complex on its own, it's represents only a tip of the "iceberg" of challenges posed by contemporary clinical trial recruitment planning . The more formidable challenge is getting the right data and organizing it into a form suitable for building prior distributions for the parameters the Poi-gamma model ( e.g. predicting individual site performance from past performance data) . In this talk we will describe how we tackled this challenge by building a data pipeline from scratch and combining it with sophisticated modelling to create an internal decision-making platform that enabled users (ClinOps professionals) to calculate probability of successfully recruiting their study on time). This innovative project clearly demonstrated increased need for collaboration across different quantitative spaces (Statistics, Data science , ML , Tech) and resulting need for statisticians to acquire new skills essential for success of such collaboration.