A small but intriguing study has revealed a machine-learning algorithm can effectively predict the social outcomes of patients with depression or psychosis better than human experts. The research found the AI tool outperformed clinician predictions when estimating a patient's social functioning up to a year in the future.

The study followed over 400 subjects split in three groups – a healthy control, those at clinically high-risk of developing psychosis, and those suffering from recent onset depression. The machine learning model was trained on a variety of patient data including neuroimaging and combined baseline data, then tasked with predicting each patient's social outcome one year into the future. These social outcomes were defined as an ability to maintain relationships with others and an ability to undertake social interactions.

In the patients at a high risk of psychosis the machine remarkably correctly predicted future social outcomes 83 percent of the time. Seventy percent of the depression patients were correctly predicted. In both cases the machine outperformed expert prognostication.

"Predicting social outcomes is important as among young people and emerging adults in OECD countries the top causes of 'disability' – and poor social functioning is included in that – are mostly disorders of mental health, including those that typically present with a first episode of psychosis," explains Stephen Wood, lead researcher on the study.

"By being able to better predict what will happen to people at high risk of psychosis or with recent onset depression over time, we are able to provide individualized treatments to clients when they first present to mental health services and potentially improve their social functioning."

This latest work in AI-driven diagnostic modeling suggests certain levels of unconscious bias present in human clinical interactions could be transcended using these computer models. One commenter on this new study hypothesized human clinicians may underpredict future social outcomes for some patients due to what is dubbed as "optimism bias".

Despite the reasonably small data set, this study points toward a fascinating future where clinicians can more effectively design treatments with the aid of diagnostic computer models. Other recent studies have reported success with AI systems designed to do everything from read mammograms and predict breast cancer risk to more generally guess a person's overall lifespan using a series of medical images.

The new research was published in the journal JAMA Psychiatry.