Identifying overlapping and distinctive traits of autism and schizophrenia using machine learning classification.

Abstract

Introduction: Autism spectrum disorder (ASD) and schizophrenia spectrum disorder (SSD) share some symptoms. We conducted machine learning classification to determine if common screeners used for research in non-clinical and subclinical populations, the Autism-Spectrum Quotient (AQ) and Schizotypal Personality Questionnaire - Brief Revised (SPQ-BR), could identify non-overlapping symptoms.

Methods: 1,397 undergraduates completed the SPQ-BR and AQ. Random forest classification modelled whether SPQ-BR item scores predicted AQ scores and factors, and vice versa. The models first used all item scores and then the least/most important features.

Results: Robust trait overlap allows for the prediction of AQ from SPQ-BR and vice versa. Results showed that AQ item scores predicted 2 of 3 SPQ-BR factors (disorganised, interpersonal), and SPQ-BR item scores successfully predicted 2 of 5 AQ factors (communication, social skills). Importantly, classification model failures showed that AQ item scores could not predict the SPQ-BR cognitive-perceptual factor, and the SPQ-BR item scores could not predict 3 AQ factors (imagination, attention to detail, attention switching).

Conclusions: Overall, the SPQ-BR and AQ measure overlapping symptoms that can be isolated to some factors. Importantly, where we observe model failures, we capture distinctive factors. We provide guidance for leveraging existing screeners to avert misdiagnosis and advancing specific/selective biomarker identification.