Speech latency may predict schizophrenia trial response, boosting drug-placebo separation

Researchers have identified a promising new speech biomarker that could significantly enrich clinical trials by reducing sample size requirements and enhancing statistical outcomes. By using speech latency, participants who are likely to show a high placebo response can be identified and excluded.

The study in Biological Psychiatry, showed that when these participants were removed from the main analysis, the treatment-placebo effect increased dramatically, by as much as two to three times the original results.

Speech latency is an objective measure of verbal response time derived from standard clinical assessments. It is sensitive to cognitive, social, and motivational factors and can be measured using recordings of psychiatric interviews.

The single, interpretable speech biomarker identified in this study was extracted from screening interviews with 406 participants with schizophrenia from three countries representing eight languages in a Phase III study of the antipsychotic drug brilaroxazine (RP5063).

“Our study has important implications for reducing clinical trial costs and burden,” says lead investigator Alex S. Cohen, Ph.D., Department of Psychology and Center for Computation and Technology, Louisiana State University, Baton Rouge, LA, and Quantic Innovations, Inc., Benton, AR.

“Using patient characteristics like speech biomarkers can help identify patients who are most likely to benefit from the medication being tested. Because these measures are objective and easy to collect, they could make trials more reliable and help reduce the risk of failure.”

Clinical trials for neuropsychiatric drugs are essential for evaluating the efficacy and safety of novel treatments yet can fail when the participants chosen are too different from each other, or when it is hard to determine who will respond to the treatment.

Enrichment strategies to decrease heterogeneity, reduce confounding psychiatric conditions, and identify participants likely to demonstrate a true pharmacological response rather than a placebo effect can successfully improve clinical trial outcomes, reduce associated costs, and help bring effective treatments to patients faster.

Among the many potential speech biomarkers related to schizophrenia, speech latency was chosen because it is tied to social communication, motivation, and cognition. It is conceptually and empirically tied to psychomotor slowing, whereby a longer pause suggests a disruption in the neural circuits responsible for translating thought into speech.

Co-investigator Laxminarayan Bhat, Ph.D., President and CEO, Reviva Pharmaceuticals, Cupertino, CA, explains, “These vocal biomarker findings reinforce brilaroxazine’s potential to address major unmet needs as a next-generation treatment for schizophrenia by confirming its broad-spectrum efficacy across core symptom domains, including negative symptoms, and corroborating our Phase III clinician-assessed outcomes.”

By enriching the sample using the speech-latency ratio, statistical significance was achieved with approximately half the sample size, with much larger effects on key symptom and functional domains. These effects were particularly notable in total, positive, and negative symptoms, of which 80%, 73%, and 57% of patients, respectively, showed a statistically significant improvement.

Dr. Cohen comments, “We were surprised by how pervasive the improvements were for the vocal biomarker-enriched sample. Improved treatment-placebo separation was observed for negative symptoms, the primary endpoint and nearly every secondary endpoint.”

Beyond its clinical accuracy, the study highlights a shift toward interpretable AI. While modern digital phenotyping often relies on “black box” algorithms, the researchers found that the most striking aspect of speech latency is the transparency and simplicity of the measure itself.

Co-investigator Mark Opler, Ph.D., MPH, Chief Research Officer, Clario, Philadelphia, PA, notes, “While the process of discovering it, validating it, and learning to analyze it correctly was very complicated and involved a collaboration between multiple organizations over many years, the core construct is not complex. While there is clearly a role for multimodal, AI-driven methods in digital phenotyping, the field should also strive for clear, interpretable measures.”

This study demonstrates that speech latency analysis is useful for evaluating clinical trial results. Given that speech latencies can be automatically and quickly computed, they could help inform participant screening. However, the authors caution that speech latencies should not be used as an endpoint. Understanding when and why speech latencies improve and optimizing measures of their sensitivity are important.

John Krystal, MD, Editor of Biological Psychiatry, concludes, “While the neurobiology of speech latency in schizophrenia is not yet well understood, this research marks significant progress in shedding light on this issue. By identifying these specific biomarkers, we can better predict treatment response and, ultimately, improve patient outcomes.”