抽象的な

Machine Learning using causal inference as a pipeline to identify drug label expansion candidates from real-world data

Yishai Shimoni

Label expansion for on-the-market drugs has huge financial and clinical potential, since it bypasses much of the time-consuming and expensive parts of the drug development process. Historically, identifying such opportunities often required serendipitous identification of positive effects of one treatment over a non-indicated outcome. The large amounts of real-world clinical data that are recently becoming available provide an opportunity to identify such effect in a systematic way, either paving the way to designing clinical trials for label expansion with a high probability of success; or allowing to test hypotheses on real data with minimal risk. Such a data-driven approach, however, presents challenges in that it contains inherent treatment bias, which may confound the observed differences in treatment effect. We have developed a pipeline that utilized causal inference methodology to address these biases and correct for them. We then identify multiple candidates for drug repurposing or label expansion that have a strong foundation in realworld results. We recently applied this pipeline to identify candidates for repurposing in Parkinson’s disease, identifying an insomnia drug that seems to have a beneficial effect for delaying the onset of dementia in these patients