Individual prediction of long-term outcome in adolescents at ultra-high risk for psychosis: Applying machine learning techniques to brain imaging data

Abstract

An important focus of studies of individuals at ultra-high risk (UHR) for psychosis has been to identify biomarkers to predict which individuals will transition to psychosis. However, the majority of individuals will prove to be resilient and go on to experience remission of their symptoms and function well. The aim of this study was to investigate the possibility of using structural MRI measures collected in UHR adolescents at baseline to quantitatively predict their long-term clinical outcome and level of functioning. We included 64 UHR individuals and 62 typically developing adolescents (12-18 years old at recruitment). At six-year follow-up, we determined resilience for 43 UHR individuals. Support Vector Regression analyses were performed to predict long-term functional and clinical outcome from baseline MRI measures on a continuous scale, instead of the more typical binary classification. This led to predictive correlations of baseline MR measures with level of functioning, and negative and disorganization symptoms. The highest correlation (r = 0.42) was found between baseline subcortical volumes and long-term level of functioning. In conclusion, our results show that structural MRI data can be used to quantitatively predict long-term functional and clinical outcome in UHR individuals with medium effect size, suggesting that there may be scope for predicting outcome at the individual level. Moreover, we recommend classifying individual outcome on a continuous scale, enabling the assessment of different functional and clinical scales separately without the need to set a threshold. Hum Brain Mapp 38:704-714, 2017. © 2016 Wiley Periodicals, Inc.

Individualized Diagnostic and Prognostic Models for Patients With Psychosis Risk Syndromes: A Meta-analytic View on the State of the Art

“…A total of 19 ML studies (73%) employed a support vector machine algorithm (10,30,(37)(38)(39)(40)(41)(42)(43)(44)(45)(46)(47)(48)(49)(50)(51)(52)(53), while the rest used Gaussian process (11) or convex hull classification (54), randomized trees (55), greedy algorithm (20), random forest (5), or LASSO regression (56,57). All ML models were computed with CV, whereas studies using Cox regression applied bootstrapping (28,(58)(59)(60)(61)(62), reported apparent results (i.e., the model is tested in the same sample from which it was derived) (63-68), or lacked a validation procedure.…”

“…One model was built on P300 amplitude from event-related potentials and sociopersonal adjustment (62). Functional outcomes were predicted with neuroanatomical (63,9,19) and blood-based biomarkers (11), and 2 studies combined clinical and MRI measures (10,30). There were no effects of data modality on SE (p = .172) or false positive rate (p = .606) ( Table S2).…”

“…The two prevailing statistical approaches to address the challenge of single-subject prediction are machine learning (ML) methods (e.g., support vector machine, LASSO [least absolute shrinkage and selection operator] regression, random forest), which can handle large databases and different data domains (24,25), and Cox proportional hazard regression, a form of multivariate survival analysis (26) able to investigate time-toconversion trajectories. Recent research applying these methods has produced prognostic models able to stratify CHR patients into different risk classes according to their pretest risk enrichment (27) or a set of combined predictors (28,29), or to predict patients' functional outcomes based on different data modalities with performance accuracies of up to 83% (10,30). Despite the great potential of these models, their applicability is still hindered by the methodological heterogeneity in the field.…”

Brain morphologic changes in early stages of psychosis: Implications for clinical application and early intervention

“…In addition, given that ARMS subjects are liable to develop poor functioning [178][179][180][181] and impaired quality of life 182,183 regardless of later psychosis onset, long-term functional disability, but not only psychosis onset itself, is an important target for prevention. 184 Interestingly, recent neuroimaging studies suggest that volume reduction predominantly in the fronto-limbic and subcortical regions, [184][185][186][187] cortical surface patterns, 188 progressive reduction in the white matter integrity of the corpus callosum, 189 and prefrontostriatal hyperactivation 190 may be possible neurobiological predictors of long-term functional outcome. However, because of the relatively small sample size and/or lack of treatment information during the follow-up period, these findings need replication and validation in a larger, wellcontrolled high-risk cohort.…”

Realising stratified psychiatry using multidimensional signatures and trajectories

“…For example, while multivariate signatures based on symptomatology were used in studies predicting outcome for first-episode psychosis [64] and symptom severity and persistence in post-traumatic stress disorder (PTSD) [65], only univariate and dichotomised outcomes were addressed and categorical diagnosis was presumed. When neuroimaging data were used as the signature, machine learning classifiers were used to predict categorical diagnosis [66], transition from an at-risk state to a dichotomised ‘psychosis versus health’ outcome [67] and dichotomised clozapine response [68, 69], or univariate aggregate predicted univariate global assessment of function (GAF) [70]. Only one study [71] used machine learning to predict multiple outcomes in major depressive disorder although again, these were dichotomised and did not model trajectories as multidimensional constructs.…”