Abstract
Background
Improved outcome prediction in individuals at high risk for psychosis may facilitate targeted early intervention. Studies suggest that improved outcome prediction may be achieved through the use of neurocognitive or neuroimaging data, on their own or in addition to clinical data. This study examines whether adding resting-state functional connectivity data to validated clinical predictors of psychosis improve outcome prediction in the prodromal stage.
Methods
This study involves 137 adolescents and young adults at Clinical High Risk (CHR) for psychosis from the Shanghai At Risk for Psychosis (SHARP) program. Based on outcome after one-year follow-up, participants were separated into three outcome categories: good outcome (symptom remission, N = 71), intermediate outcome (ongoing CHR symptoms, N = 30), and poor outcome (conversion to psychosis or treatment-refractory, N = 36). Resting-state fMRI data were acquired for each participant and processed using the Conn toolbox, including rigorous motion correction. Multinomial logistic regression analysis and leave-one-out cross-validation were used to assess the performance of three prediction models: 1) a clinical-only model using validated clinical predictors from the NAPLS-2 psychosis-risk calculator, 2) an fMRI-only model using measures of functional connectome organization and within/between-network connectivity among established resting-state networks, and 3) a combined clinical and fMRI prediction model. Model performance was assessed using the harmonic mean of the positive predictive value and sensitivity for each outcome category. This F1 measure was compared to expected chance-levels using a permutation test with 1,000 sampled permutations in order to evaluate the statistical significance of the model’s prediction.
Results
The clinical-only prediction model failed to achieve a significant level of outcome prediction (F1 = 0.32, F1-chance = 0.26 □ 0.06, p = .154). The fMRI-only model did predict clinical outcome to a significant degree (F1 = 0.41, F1-chance = 0.29 □ 0.06, p = .016), but the combined clinical and fMRI prediction model showed the best performance (F1 = 0.46, F1-chance = 0.29 □ 0.06, p < .001). On average, positive predictive values (reflecting the probability that an outcome label predicted by the model was correct) were 39% better than chance-level and 32% better than the clinical-only model. Analyzing the contribution of individual predictor variables showed that GAF functional decline, a family history of psychosis, and performance on the Hopkins Verbal Learning Test were the most influential clinical predictors, whereas modular connectome organization, default-mode and fronto-parietal within-network connectivity, and between-network connectivity among language, salience, dorsal attention, cerebellum, and sensorimotor networks were the leading fMRI predictors.
Discussion
This study’s findings suggest that functional brain abnormalities reflected by alterations in resting-state functional connectivity precede and may drive subsequent changes in clinical functioning. Moreover, the findings show that markers of functional brain connectivity may be useful for improving early identification and clinical decision-making in prodromal psychosis.
Heritability of Resting State EEG Functional Connectivity Patterns
