Dynamic prediction of transition to psychosis using joint modelling

Area of Science:

• Psychiatry and Mental Health
• Statistical Modeling
• Clinical Prediction

Background:

• Research on predicting transition to psychosis in ultra-high risk (UHR) individuals typically uses only baseline data.
• Longitudinal data, such as psychopathology, are often collected but underutilized in existing prediction models.
• Dynamic prediction, updating risk assessments with accumulating data, has not been applied in the UHR context.

Purpose of the Study:

• To explore the utility of dynamic prediction for enhancing psychosis onset prediction in UHR individuals.
• To determine if dynamic prediction models outperform traditional baseline-only models.
• To investigate the application of joint modeling for dynamic prediction in this population.

Main Methods:

• Utilized data from the NEURAPRO study, including 304 UHR individuals.
• Employed joint modeling, an emerging statistical methodology, to implement dynamic prediction.
• Compared prediction accuracy (sensitivity, specificity, likelihood ratios) of dynamic versus baseline-only models.

Main Results:

• Dynamic prediction using joint modeling demonstrated significantly improved sensitivity, specificity, and likelihood ratios compared to baseline predictors.
• The updated predictions reflect the evolving clinical status of individuals over time.
• This method provides more accurate and timely risk assessments.

Conclusions:

• Dynamic prediction, particularly using joint modeling, offers a significant advancement over conventional methods for predicting psychosis onset in UHR individuals.
• This approach can serve as a valuable tool for clinicians to refine prognostic judgments.
• It supports timely, personalized treatment decisions based on unfolding patient symptomatology.