Application of joint modelling to the analysis of transition to psychosis

Abstract

Background and Aims: Psychosis is a mental condition that is associated with serious adverse effects on people's personal lives and a huge economic burden on society. Criteria for identifying people at high risk of developing psychosis, referred to as clinical high risk (CHR) criteria, were established during the 1990s and 2000s. These criteria have given hope that treatments can be provided at an early stage to delay or prevent the onset of psychosis. However, while individuals satisfying these criteria, called CHR individuals, certainly have a much higher risk of developing psychosis than the general population, a considerable proportion of them do not transition from an at-risk state to psychosis. Therefore, much research has been conducted in the past twenty years or so seeking predictors of transition to psychosis among CHR individuals. Such research work has almost entirely been based on fixed predictors, which are predictors whose values do not change or pertain to only the baseline time point, i.e. at study entry. Even though longitudinal data, i.e. repeated measurements over time, can be obtained and are common in transition to psychosis studies, they are rarely utilised in the analysis. Predictors with longitudinal measurements are called time-dependent predictors (TDPs). TDPs carry information about the progression of the illness and they can provide continuous updates of the transition risk prediction over time according to the changes in their values. Such updating of prediction is called dynamic prediction and could contribute to the provision of timely and personalized treatment to the patients concerned. This doctoral work has two aims: (1) to examine the inclusion of TDPs in the analysis of transition to psychosis among CHR individuals and (2) to examine the implementation of dynamic prediction in predicting transition to psychosis among CHR individuals.

Method: The relatively new statistical methodology, joint modelling, was used to incorporate TDPs into the analysis of transition to psychosis data. Both simulated data and real data were used to assess the performance of joint modelling and the potential benefit of the inclusion of longitudinal data in the analysis and prediction of the risk of transition.

Results: Compared to the conventional approach of using only baseline values, the inclusion of longitudinal data via joint modelling was shown to provide better statistical inference in the estimation of the effect of the predictors on the risk of transition. Moreover, dynamic prediction through the use of TDPs was shown to have the potential to provide better prediction of the risk of transition than the use of fixed predictors only.

Conclusion: The inclusion of longitudinal data in the form of TDPs via joint modelling in the analysis of transition to psychosis is certainly a worthwhile endeavour. Dynamic prediction of transition through the incorporation of TDPs in prediction models has great potential in improving the prediction of transition and enabling the provision of personalized treatment.