In the Journals

Machine learning model predicts functional outcomes in early-stage mental illness

Findings from a multimodal, multisite machine learning analysis revealed that prediction models determined 1-year social-functioning outcomes in up to 83% of patients at clinical high-risk for psychosis and 70% of patients with recent-onset depression.

Prior research has indicated that psychosis can be predicted in individual patients in clinical high-risk states using clinical, neurocognitive, neurophysiological and MRI data, Nikolaos Koutsouleris, MD, from the department of psychiatry and psychotherapy at Ludwig-Maximilian-University, Germany, and colleagues explained in JAMA Psychiatry. Using machine learning has strengthened these findings, demonstrating that clinical baseline data may be linked to predictors of functional and treatment outcomes in first-episode psychosis and depression.

“Although it has been conceptually suggested, the question of whether behavioral and MRI-based data could be efficiently combined within sequential prognostic algorithms to optimize predictive power has yet to be empirically tested,” Koutsouleris and colleagues wrote.

This multisite naturalistic study examined whether using clinical, imaging-based and combined machine learning can identify predictors of social and role functioning in patients in clinical high-risk states for psychosis or with recent-onset depression.

The investigators also compared the geographic, transdiagnostic and prognostic generalizability of machine learning with human prognostication, and evaluated the sequential prognosis encompassing clinical and combined machine learning. Participants in high-risk states, participants with recent-onset depression and those with recent-onset psychosis, as well as healthy control participants, were followed for 18 months.

The study included 116 people at high-risk for psychosis and 120 with recent-onset depression. Koutsouleris and colleagues identified generalizable clinical, imaging-based and combined prediction models of social functioning impairments in patients at increased risk for psychosis and recurrent depression.

Using clinical baseline data, machine-learning prediction models correctly identified the 1-year social-functioning outcomes with a balanced accuracy of 76.9% of patients in high-risk states and 66.2% of those with recent-onset depression. Using structural neuroimaging, models predicted outcomes in 76.2% of patients at high risk for psychosis and 65% of patients with early-stage depression. Combined models correctly predicted 1-year outcomes in 82.7% of patients in high-risk states and 70.3% of those with recent-onset depression.

Researchers found that lower functioning prior to study entry was a transdiagnostic predictor. They also reported that machine learning outperformed expert prognostication. The results showed that neuroimaging machine learning plus clinical machine learning provided a 1.9-fold increase of prognostic certainty in uncertain cases of patients at clinical high-risk and a 10.5-fold increase for patients with recent-onset depression.

“To further elucidate the clinical, environmental, and neurobiological factors that facilitate or limit the transferability of the risk calculators presented here, external and prospective validation is needed in ethnically diverse patient populations recruited at sites beyond the European catchment areas of this study,” the authors wrote. “This is the next important step toward quantifying the feasibility and utility of precision psychiatry approaches for the secondary and tertiary prevention of severe mental illnesses.”

Questions remain about how the prognostic test described in this study can be implemented in the clinic, Aristotle N. Voineskos, MD, PhD, of the Centre for Addiction and Mental Health, Toronto, and the department of psychiatry, University of Toronto, said in a related comment.

“If neuroscience is to be embedded or included as a part of clinical care, it will require all of the skills and knowledge of the psychiatric services community, implementation experts, quality improvement experts, and people with lived experience, along with clinical neuroscientists,” he wrote. “The continuous cycle of precision medicine, implementation, and the learning health care system could serve as a model for psychiatry, given our field’s lack of progress to date.” – by Savannah Demko

Disclosures: Koutsouleris reports honoraria for talks presented at education meetings organized by Otsuka/Lundbeck. Please see the study for all other authors’ relevant financial disclosures. Voineskos reports funding from the Canada Foundation for Innovation, the Canadian Institutes for Health Research, the Centre for Addiction and Mental Health Foundation, the NIMH, the Ontario Ministry of Research and Innovation and the University of Toronto.

Findings from a multimodal, multisite machine learning analysis revealed that prediction models determined 1-year social-functioning outcomes in up to 83% of patients at clinical high-risk for psychosis and 70% of patients with recent-onset depression.

Prior research has indicated that psychosis can be predicted in individual patients in clinical high-risk states using clinical, neurocognitive, neurophysiological and MRI data, Nikolaos Koutsouleris, MD, from the department of psychiatry and psychotherapy at Ludwig-Maximilian-University, Germany, and colleagues explained in JAMA Psychiatry. Using machine learning has strengthened these findings, demonstrating that clinical baseline data may be linked to predictors of functional and treatment outcomes in first-episode psychosis and depression.

“Although it has been conceptually suggested, the question of whether behavioral and MRI-based data could be efficiently combined within sequential prognostic algorithms to optimize predictive power has yet to be empirically tested,” Koutsouleris and colleagues wrote.

This multisite naturalistic study examined whether using clinical, imaging-based and combined machine learning can identify predictors of social and role functioning in patients in clinical high-risk states for psychosis or with recent-onset depression.

The investigators also compared the geographic, transdiagnostic and prognostic generalizability of machine learning with human prognostication, and evaluated the sequential prognosis encompassing clinical and combined machine learning. Participants in high-risk states, participants with recent-onset depression and those with recent-onset psychosis, as well as healthy control participants, were followed for 18 months.

The study included 116 people at high-risk for psychosis and 120 with recent-onset depression. Koutsouleris and colleagues identified generalizable clinical, imaging-based and combined prediction models of social functioning impairments in patients at increased risk for psychosis and recurrent depression.

Using clinical baseline data, machine-learning prediction models correctly identified the 1-year social-functioning outcomes with a balanced accuracy of 76.9% of patients in high-risk states and 66.2% of those with recent-onset depression. Using structural neuroimaging, models predicted outcomes in 76.2% of patients at high risk for psychosis and 65% of patients with early-stage depression. Combined models correctly predicted 1-year outcomes in 82.7% of patients in high-risk states and 70.3% of those with recent-onset depression.

Researchers found that lower functioning prior to study entry was a transdiagnostic predictor. They also reported that machine learning outperformed expert prognostication. The results showed that neuroimaging machine learning plus clinical machine learning provided a 1.9-fold increase of prognostic certainty in uncertain cases of patients at clinical high-risk and a 10.5-fold increase for patients with recent-onset depression.

“To further elucidate the clinical, environmental, and neurobiological factors that facilitate or limit the transferability of the risk calculators presented here, external and prospective validation is needed in ethnically diverse patient populations recruited at sites beyond the European catchment areas of this study,” the authors wrote. “This is the next important step toward quantifying the feasibility and utility of precision psychiatry approaches for the secondary and tertiary prevention of severe mental illnesses.”

Questions remain about how the prognostic test described in this study can be implemented in the clinic, Aristotle N. Voineskos, MD, PhD, of the Centre for Addiction and Mental Health, Toronto, and the department of psychiatry, University of Toronto, said in a related comment.

“If neuroscience is to be embedded or included as a part of clinical care, it will require all of the skills and knowledge of the psychiatric services community, implementation experts, quality improvement experts, and people with lived experience, along with clinical neuroscientists,” he wrote. “The continuous cycle of precision medicine, implementation, and the learning health care system could serve as a model for psychiatry, given our field’s lack of progress to date.” – by Savannah Demko

Disclosures: Koutsouleris reports honoraria for talks presented at education meetings organized by Otsuka/Lundbeck. Please see the study for all other authors’ relevant financial disclosures. Voineskos reports funding from the Canada Foundation for Innovation, the Canadian Institutes for Health Research, the Centre for Addiction and Mental Health Foundation, the NIMH, the Ontario Ministry of Research and Innovation and the University of Toronto.