Machine-Learning Holds Promise for Identifying First-Episode Psychosis

May 21, 2020

Artificial intelligence combined with electronic health records may detect FEP a year before onset in asymptomatic adults.


A novel approach is believed to be remarkably accurate in predicting first-episode psychosis (FEP). According to the article, “Dynamic ElecTronic hEalth reCord deTection (DETECT) of Individuals at Risk of a First Episode of Psychosis: A Case Control Development and Validations Study,” published in Lancet Digital Health, FEP can be detected nearly a year before onset in asymptomatic individuals.

The study was based on electronic health records (EHR) data from the IBM Explorys Platform that included people who received a diagnosis of FEP and subsequent schizophrenia from 1990 to 2018. A dataset of more than 60 million unidentifiable individuals was used to devise a model known as Dynamic ElecTronic hEalth reCord deTection (DETECT).

With this model, the researchers sought to determine whether EHR together with artificial intelligence could predict risk of developing FEP with some measure of precision: “DETECT was developed to reliably identify individuals potentially at risk of FEP 1 year in advance to ensure that risk prediction was done before the manifestation of more obvious behavioural [sic] pathology and functional morbidity,” according to the researchers.

The authors of the study who were both from  Kings College in London and Lundbeck, which funded the study, noted several limitations. The model design and validation methods may have compromised results because they were conducted in a case-control setting. Further study is needed to establish true analysis of an entire population of individuals with schizophrenia.

There is promise in the results, however, which point to the need for more research: “DETECT could potentially be useful to screen EHRs on a large scale. The primary purpose of such large-scale screening would be to detect individuals at risk of FEP who might be missed by the current health-care pathways.”

Further reading about the ethics of large-scale population screening can be found in an article by Ioana Alina Cristea and Florian Naudet in their accompanying piece in Lancet Digital Health, Is large-scale population screening coming to psychiatry?



1. Raket RR, Jaskolowski J, Kinon BJ. Dynamic ElecTronic hEalth reCord deTection (DETECT) of individuals at risk of a first episode of psychosis: a case-control development and validation study. Lancet Digit Health. 2020;2(5):E229-E239.