This article is more than 5 years old. Information may no longer be current.

Machine learning helps predict treatment outcomes for schizophrenia

Researchers identified first-episode drug-naive schizophrenia patients with an accuracy of 78.6% and predicted their responses to antipsychotic treatment with an 82.5% accuracy at the individual level using machine-learning algorithms and the functional connections of the superior temporal cortex, according to study findings.

“Finding biomarkers of schizophrenia at the first episode without the confounding effects of treatment has been challenging, and identifying reliable non-invasive brain imaging markers in the early phase of illness remains an important goal,” Bo Cao, PhD, from the department of psychiatry at University of Alberta in Canada, and the department of psychiatry and behavioral sciences at The University of Texas Health Science Center, and colleagues wrote in Molecular Psychiatry. “Using these biomarkers to make individual predictions of future treatment responses to antipsychotic medicine in the early phase of schizophrenia would be clinically invaluable.”

Researchers used a machine-learning algorithm to examine functional MRI images of 38 first-episode drug-naive untreated schizophrenia patients and 29 healthy controls. Specifically, the algorithm measured functional connections between cortical regions in the brain.

Patients received treatment with risperidone for 10 weeks; the dose increased to 3 mg to 6 mg a day during the first week of administration then stayed the same until the end of the study. Two clinical psychiatrists, who were blind to the purpose of the study, examined patients’ scores on the positive symptoms and hallucination subscale of the positive and negative syndrome scale (PANSS).

The results showed abnormal functional connectivity involving the superior temporal cortex in patients with first-episode drug-naive schizophrenia compared with healthy controls. Cao and colleagues used machine-learning algorithms and functional connections between superior temporal cortex and other cortical regions as input, which allowed them to accurately identify the schizophrenia patients and predict their individual responses to antipsychotic treatment (balanced accuracy: 78.6%; accuracy: 77.8%; sensitivity: 74.4; specificity: 82.8%).

After participants received antipsychotic treatment for 10 weeks, the investigators found that patients with first-episode drug-naive schizophrenia experienced improvement of symptoms as measured by the PANSS total scores (P < .0001) and subtotal scores of positive (P < .0001), negative (P = .0109) and general symptoms (P < .0001). Based on the predicted percentage drop of the PANSS, the model also predicted with 82.5% accuracy whether a patient would respond positively to antipsychotic treatment with risperidone.

“This is the first step, but ultimately we hope to find reliable biomarkers that can predict schizophrenia before the symptoms show up. We also want to use machine learning to optimize a patient's treatment plan,” Cao said in a press release. “In the future, with the help of machine learning, if the doctor can select the best medicine or procedure for a specific patient at the first visit, it would be a good step forward. It will be a joint effort of the patients, psychiatrists, neuroscientists, computer scientists and researchers in other disciplines to build better tools for precise mental health.” – by Savannah Demko

Disclosures: Cao reports no relevant financial disclosures. Please see the study for other authors’ relevant financial disclosures.