Dynamic prediction of transition to psychosis using joint modelling

Considerable research has been conducted seeking risk factors and constructing prediction models for transition to psychosis in individuals at ultra-high risk (UHR). Nearly all such research has only employed baseline predictors, i.e. data collected at the baseline time point, even though longitudinal data on relevant measures such as psychopathology have often been collected at various time points. Dynamic prediction, which is the updating of prediction at a post-baseline assessment using baseline and longitudinal data accumulated up to that assessment, has not been utilized in the UHR context. This study explored the use of dynamic prediction and determined if it could enhance the prediction of frank psychosis onset in UHR individuals. An emerging statistical methodology called joint modelling was used to implement the dynamic prediction. Data from the NEURAPRO study (n = 304 UHR individuals), an intervention study with transition to psychosis study as the primary outcome, were used to investigate dynamic predictors. Compared with the conventional approach of using only baseline predictors, dynamic prediction using joint modelling showed significantly better sensitivity, specificity and likelihood ratios. As dynamic prediction can provide an up-to-date prediction for each individual at each new assessment post entry, it can be a useful tool to help clinicians adjust their prognostic judgements based on the unfolding clinical symptomatology of the patients. This study has shown that a dynamic approach to psychosis prediction using joint modelling has the potential to aid clinicians in making decisions about the provision of timely and personalized treatment to patients concerned. (C) 2018 Elsevier B.V. All rights reserved.