Cognitive performance and differentiation of B-SNIP psychosis Biotypes: Algorithmic Diagnostics for Efficient Prescription of Treatments (ADEPT) - 2

Abstract

Objective

The B-SNIP consortium validated neurobiologically defined psychosis Biotypes (BT1, BT2, BT3) using cognitive and psychophysiological measures. B-SNIP’s biomarker panel is not practical for most settings. Previously, B-SNIP developed an efficient classifier of Biotypes using only clinical assessments (called ADEPT-CLIN) with acceptable accuracy (∼.81). Adding cognitive performance may improve ADEPT’s performance.

Method

Clinical assessments from ADEPT-CLIN plus 18 cognitive measures from 1907 individuals with a B-SNIP psychosis Biotype were used to create an additional diagnostic algorithm called ADEPT-COG. Extremely randomized trees were used to create this low burden classifier.

Results

Total Biotype classification accuracy peaked at 94.6 % with 65 items. A reduced set of 18 items showed 90.5 % accuracy. Only 9–10 items achieved a one-vs-all (e.g., BT1 or not) accuracy of ∼.95, considerably better than using clinical assessments alone. The top discriminators of psychosis Biotypes were antisaccade proportion correct, BACS total, symbol coding, antisaccade correct response latency, verbal memory, digit sequencing, stop signal reaction times, stop signal proportion correct, Tower of London, and WRAT Reading. Except for antisaccade proportion correct and Tower of London, there was no overlap of the top discriminating items for B-SNIP Biotypes and DSM psychosis categories.

Conclusions

This low-burden algorithm using clinical and cognitive measures achieved high classification accuracy and can support Biotype-specific etiological and treatment investigations in clinical and research environments. It may be especially useful for clinical trials.