All Times EDT
Keywords: health outcome identification, natural language processing, causal inference, targeted maximum likelihood estimation, drug utilization, real world data
In post-market drug safety surveillance, machine learning (ML) has been used to support regulatory decision making. In this poster, we highlight some of the completed and ongoing ML applications from the Division of Biometrics VII of Center for Drug Evaluation and Research, in Food and Drug Administration (FDA). We summarize the key findings and provide discussions of issues and concerns when using ML in the regulatory setting.
The ML applications discussed in this poster include: i) two projects using ML and natural language processing (NLP) to improve the accuracy of health outcome identification (HOI) using claims data and electronic health records, ii) a research effort to explore the sample size determination issue with imbalanced data at the planning stage of HOI using ML, iii) a project using NLP to improve the patient age ascertainment in the FDA adverse event reporting system, iv) a research project to apply and compare the ML algorithms to explore risk factors driving opioid prescribing, v) a software development of a ML-based analytic dashboard to explore geo-referenced patterns of prescription drug utilization, and vi) a FDA funded project to evaluate the performance of using targeted maximum likelihood estimation and super learner in causal inference, with potential applications to clinical trials and real-world data. These projects demonstrate the present utility and future potential of ML for regulatory science.