Computational speech markers of negative symptoms show evidence of being robust to antipsychotic dose and extrapyramidal symptoms in schizophrenia

There is increasing interest in using computational speech and language analysis for the objective assessment of symptom severity in schizophrenia spectrum disorders (SSD), including negative symptom severity. However, the extent to which antipsychotic side effects influence speech markers of negative symptoms is unclear. This study investigated associations between computational speech and negative symptom severity, and examined whether speech features are associated with antipsychotic dose and extrapyramidal symptoms (EPS). Sixty-seven participants with SSD completed speech-focused tasks and clinical assessment. Seventeen speech features capturing relevant acoustic, timing, and linguistic characteristics were extracted from participant responses and examined for associations with negative symptoms, EPS, and antipsychotic dose. Eight timing features and one linguistic feature were significantly correlated with negative symptom severity (p < .05, false-discovery-rate-corrected; FDR), and these relationships were specific to negative symptoms rather than overall psychiatric symptom severity. No features were significantly correlated with antipsychotic dose or EPS after FDR correction, and Bayesian analyses provided moderate evidence for the null hypothesis (i.e., absence of an association) for most features. Two features showed Bayesian evidence for the alternative hypothesis, however; the association between speech and negative symptoms remained significant after removing the clinical covariate effects. The results provide support for computational speech markers of negative symptom severity, and both frequentist and Bayesian evidence that these speech markers are generally not confounded by antipsychotic side effects. The results help advance the clinical validation of computational speech assessment and analysis for measuring negative symptom severity in SSD.