Key Takeaways
Extended Takeaways
- In this community sample, full-interview speech models performed substantially better than patient-only verified segments for major depressive disorder, with best F1-scores of 77.48% (±5.59) versus 45.05% (±24.84), suggesting that longer conversational samples may be important if speech screening is implemented in prenatal care.
- For generalized anxiety disorder, full-interview models also outperformed verified-segment and wav2vec2 approaches, achieving a best F1-score of 79.57% (±14.81) compared with 64.89% (±16.68) and 51.62% (±14.09), respectively; simpler acoustic feature engineering outperformed the pretrained deep model in this dataset.
- Models based only on pregnancy and sociodemographic variables had limited discrimination for both major depressive disorder and generalized anxiety disorder, with best F1-scores of 39.33% (±5.36) and 33.04% (±15.50), indicating that standard risk factors alone may miss clinically diagnosed cases captured by speech patterns.
- Adding pregnancy and sociodemographic characteristics to the best speech pipelines did not improve classification: generalized anxiety disorder remained at 79.57%±14.81, major depressive disorder declined to 75.00%±17.09, and unpaired t tests showed no statistical difference versus speech-only models for major depressive disorder (P value=∼.82) or generalized anxiety disorder (P value=1.0).
- The strongest candidate markers differed somewhat by diagnosis: pause duration, F0, and shimmer were most influential for major depressive disorder, whereas harmonics-to-noise ratio, jitter, and F0 were most influential for generalized anxiety disorder, which may help clinicians interpret why monotony, pausing, and voice-quality changes could carry diagnostic signal.
- These findings come from a small but diagnostically rigorous sample of 146 participants, including 19 with major depressive disorder and 28 with generalized anxiety disorder, with outcomes established by SCID-5 rather than symptom scales; however, 72% of participants identified their ethnic background as White, which limits generalizability across more diverse pregnant populations.
