Prediction Tool for Individual Outcome Trajectories

The following is a summary of “Prediction Tool for Individual Outcome Trajectories Across the Next Year in First-Episode Psychosis in Coordinated Specialty Care,” published in the November 2022 issue of Psychiatry by Basaraba, et al.

Creating trustworthy, validated individual-level prediction tools for important outcomes in coordinated specialty care (CSC) settings for patients with a first episode of psychosis may be instructive for joint clinical and client decision-making. For a study, researchers sought to create a tool that predicted outcomes for mental hospitalization, education/work status, or both during the course of a client’s subsequent year of quarterly follow-up examinations. Additionally, it provides physicians and clients with an educational visualization of the predictions.

All patients engaged in the OnTrackNY program had individual-level data gathered for them at registration and during quarterly follow-ups using standardized forms. Individuals aged 16 to 30 with recently developed (<2 years) nonaffective psychosis receive person-centered, recovery-oriented, and evidence-based psychosocial and pharmacological therapies through the OnTrackNY program, a network of CSC sites across New York State. Although data collection was ongoing, the study's data were gathered between October 2013 and December 2018, and the analysis took place between July 2020 and May 2021. To construct an internally validated model, the data were split into a training/cross-validation set and a holdout test set (~20% of the sample), which was used for external validation. To forecast the trajectory of events at the individual level, random probability forest models were created. Thirteen site-level demographic and economic census variables and 43 individual-level clinical and demographic data were obtained during OnTrackNY enrollment, 25 time-varying and updated at quarterly follow-up evaluations.

1,298 people between the ages of 16 and 30 made up the study’s entire sample, of which 341 (26.3%), 949 (73.1%), and 8 (<1%) were female. 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), prediction models performed well for 1-year trajectories of education/work across all. validation sets. AUCs above 0.70 were routinely achieved for forecasts of future mental hospitalizations occurring three months in advance, whereas AUCs below 0.60 were consistently attained for predictions occurring six months in the future or later. A prototype interactive visualization tool depicting individual-level school/work trajectories and related attributes was built in light of the strong externally validated performance for forecasting education and employment.

According to the study, precise prediction tools for first-episode psychosis patients’ outcomes may be created. The tools guide joint clinical and client decision-making. In the context of a learning healthcare system, future research should examine the effectiveness of its use, including optimal communication to support joint clinician/client decision-making. Before any strategy was suggested for the outcome, further study was necessary to create better-performing prediction models for upcoming mental hospitalizations.


Leave a Comment

Your email address will not be published. Required fields are marked *