Prediction tools can be developed for those with first-episode psychosis

Accurate prediction tools can be developed for people with first episodes of psychosis, which could potentially help clinician decision-making, researchers reported in JAMA Psychiatry.

Cale N. Basaraba, MPH, of the New York State Psychiatric Institute, and colleagues sought to develop an individual-level prediction tool that used machine-learning methods to predict a trajectory of education/work status or psychiatric hospitalization outcomes over the next year of a patient's life following an episode of psychosis.

“Identifying adolescents or young adults with a psychotic disorder as soon as possible after their first episode of psychosis and connecting them with care has been shown to lead to better outcomes,” Basaraba and colleagues wrote. “Specifically, the use of what in the United States is referred to as coordinated specialty care, a multi-element team-based approach, has shown significant advantages over traditional care, including increased treatment engagement, improved quality of life and greater symptom reduction.”

The researchers obtained individual-level data for patients enrolled in the OnTrackNY program, a community service clinic, at enrollment and quarterly follow-ups. The OnTrackNY program provides person-centered, recovery-oriented, evidence-based psychological and pharmaceutical interventions to those aged 16 to 30 years with psychosis.

The authors separated the data obtained into a training/cross-validation set to perform internally validated model development and a separate holdout test for external validation. In addition, random probability forest models were developed to predict individual-level trajectories of outcomes.

Basaraba and colleagues collected 43 individual-level demographic and clinical features at enrollment, 25 of which were time-varying and updated at follow-up assessments, as well as 13 site-level demographic and economic census variables.

The total study sample included 1,298 participants aged 16 to 30 years (73.1% men). The prediction model performed well for 1-year trajectories of education/work across all validation sets, with areas under the receiver operating characteristic curve (AUCs) ranging from 0.68 (95% CI, 0.63-0.74) to 0.88 (95% CI, 0.81-0.96), the authors reported.

Accuracy for psychiatric hospitalizations 3 months ahead reached an AUC above 0.70, while predictions for hospitalizations at 6 months were consistently poor, with AUCs below 0.60.

“Future work should study the effectiveness of its deployment, including proper communication to inform shared clinician/client decision-making in the context of a learning health care system,” Basaraba and colleagues wrote. “At present, more work is needed to develop better performing prediction models for future psychiatric hospitalizations before any tool is recommended for this outcome.”