The analysis of large-scale multidomain databases has become a key avenue in the quest for identifying predictors of psychosis. However, despite large scientific efforts, until today no clinically viable predictive models have been derived from these databases. This might to some extent result from methodological shortcomings of classical statistical approaches that are typically applied to these data. New methods such as multivariate pattern analysis (MVPA) hold the promise to overcome these drawbacks and have therefore been successfully applied in multiple areas of predictive medicine. Most importantly, MVPA facilitates predictions at the single-subject level which is a prerequisite of early recognition findings becoming part of routine diagnostic algorithms. One potential application area of MVPA is the automated classification of clinically relevant disease outcomes. In this regard, the ‘first generation' of MVPA studies provided evidence that these methods can differentiate healthy individuals from different psychiatric populations such as patients with depressive, bipolar or schizophrenic disorders. Recently, MVPA methods have been successfully employed in the more challenging classification of different patient populations. However, in terms of early recognition, the most interesting area of application of these methods is the individualized stratification of patients into high-risk and established disease stages as well as the prediction of treatment response. Along this path, predictive modelling could potentially evolve into a clinical tool enabling clinicians to personalize preventive therapy, thus avoiding unnecessary interventions and making the most efficient use of limited clinical resources. The present chapter gives an overview of the methodology of MVPA and current results obtained using this methodology in the emerging field of predictive psychiatry. Clinical implications and further potential applications are discussed.
Paper
Full text
Pattern Recognition Methods in the Prediction of Psychosis
Semantic Scholar · Psychology · 2016
Abstract
The analysis of large-scale multidomain databases has become a key avenue in the quest for identifying predictors of psychosis. However, despite large scientific efforts, until today no clinically viable predictive models have been derived from these databases. This might to some extent result from methodological shortcomings of classical statistical approaches that are typically applied to these data. New methods such as multivariate pattern analysis (MVPA) hold the promise to overcome these drawbacks and have therefore been successfully applied in multiple areas of predictive medicine. Most importantly, MVPA facilitates predictions at the single-subject level which is a prerequisite of early recognition findings becoming part of routine diagnostic algorithms. One potential application area of MVPA is the automated classification of clinically relevant disease outcomes. In this regard, the ‘first generation' of MVPA studies provided evidence that these methods can differentiate healthy individuals from different psychiatric populations such as patients with depressive, bipolar or schizophrenic disorders. Recently, MVPA methods have been successfully employed in the more challenging classification of different patient populations. However, in terms of early recognition, the most interesting area of application of these methods is the individualized stratification of patients into high-risk and established disease stages as well as the prediction of treatment response. Along this path, predictive modelling could potentially evolve into a clinical tool enabling clinicians to personalize preventive therapy, thus avoiding unnecessary interventions and making the most efficient use of limited clinical resources. The present chapter gives an overview of the methodology of MVPA and current results obtained using this methodology in the emerging field of predictive psychiatry. Clinical implications and further potential applications are discussed.