Machine Learning for Predicting Rare Clinical Outcomes—Finding Needles in a Haystack

Abstract

Assessing the risk of individual patients for experiencing specific clinical outcomes is essential for clinicians to make decisions. In practice, many of these clinical outcomes are rare, which makes the risk assessment process like finding needles in a haystack. Machine learning (ML) 1 aims to develop computational algorithms from massive data to extract informative signals and holds great potentials in this scenario. Recently, ML algorithms have demonstrated the capability to achieve superior performance compared with clinically used risk calculators in outcome prediction with real world patient data, such as claims or electronic health records (EHRs).

Topics & Concepts