"Challenges in AI-Driven Patient Assessment and Personalized Medicine Predictions"

A study has found that computer algorithms used to predict treatment outcomes for people with schizophrenia perform well on data they were trained on, but their accuracy drops significantly when applied to new, unseen data sets. The research highlights the need for rigorous testing of clinical prediction models on large data sets to ensure their reliability, similar to the process for drug development. Only about 20% of psychiatric prediction models undergo validation on samples other than the ones on which they were developed, indicating a need for more disciplined algorithm development and testing in healthcare.