Practical challenges for precision medicine

Abstract

Precision medicine promises treatments tailored to individual patient profiles. Machine learning models have been heralded as the tools to accelerate precision medicine by sifting through large amounts of complex data to pinpoint the genetic, sociodemographic, or biological markers that predict the right treatment for the right person at the right time. However, the initial enthusiasm for these advanced predictive tools is now facing a sobering reality check. On page 164 of this issue, Chekroud et al. (1) show that machine learning models that predict treatment response to antipsychotic medication among individuals with schizophrenia in one clinical trial failed to generalize to data from new, unseen clinical trials. The findings not only highlight the necessity for more stringent methodological standards for machine learning approaches but also require reexamination of the practical challenges that precision medicine is facing.