Clinical speech AI for schizophrenia: Diagnostic accuracy of speech- and language-based digital biomarkers in a PRISMA-DTA systematic review and meta-analysis.
Xiaoqi Tang, Junmei Chen, Yuehe Huang et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis evaluated the diagnostic accuracy of speech- and language-based digital biomarkers for schizophrenia, analyzing 46 reports. The primary patient-level healthy-control set showed a sensitivity of 0.804 and specificity of 0.829, with an SROC AUC of 0.865. However, the study concludes that while these models demonstrate internal discrimination, their real-world diagnostic utility remains unestablished.
Key findings
- Sensitivity of 0.804 (95% CI 0.743-0.853) and specificity of 0.829 (0.800-0.855) were found in the primary healthy-control set.
- The exploratory psychiatric/mixed-comparator set showed a sensitivity of 0.637 (0.505-0.751) and specificity of 0.882 (0.736-0.953).
- Substantial heterogeneity was observed for sensitivity (τ² = 0.328) but limited for false-positive rate (τ² = 0.021).
- No primary dataset provided verified independent external validation.
Why it matters
The findings highlight the potential of speech and language models as diagnostic tools for schizophrenia, but underscore the need for further validation before clinical implementation. Clinicians should be aware of the current limitations in real-world applicability.