Work

The Xaant Resting Autonomic Signature (X-RAS): prospective clinical validation of an autonomic biomarker for anxiety and depressive disorders

Year: 2026
Type: preprint
Study design: Observational Study
Abstract: Background Anxiety and depressive disorders affect over 500 million people globally, with a treatment gap exceeding 75% in low- and middle-income countries. Current diagnostic frameworks rely on clinical interview and self-report, with test–retest reliability of 0.20 to 0.67 for these disorders in DSM-5 field trials. Autonomic nervous system dysregulation is a well-established biological correlate of anxiety and depression. We report the first, to our knowledge, prospective clinical validation of an autonomic biomarker-based artificial intelligence system against a blinded psychiatrist reference standard in routine clinical practice. Methods We conducted a prospective, two-site diagnostic accuracy study at tertiary hospitals in Delhi and Kerala, India (CTRI/2023/07/054984). The Xaant Resting Autonomic Signature (X-RAS) platform, a three-layer LSTM with self-attention previously developed on an independent cohort of 2,349 participants and frozen prior to the trial, analysed electrodermal activity and blood volume pulse from 15-minute resting-state recordings and produced a four-class diagnostic classification. Of 180 participants enrolled, 170 completed the protocol. Out of these, 114 clinical participants with DSM-5 depression, anxiety, or comorbid disorders, and 56 reference group participants recruited from a long-term meditation practice tradition were considered. The reference standard was the DSM-5 diagnosis established by the site psychiatrist, who was blinded to the platform output. The primary outcome was Cohen's kappa for AI-psychiatrist diagnostic agreement within the clinical cohort. Results Within the clinical cohort, the platform achieved 93.0% agreement with clinical diagnosis (κ = 0.884, 95% CI 0.789 to 0.956), with 100% sensitivity for generalised anxiety disorder (GAD) detection in the GAD subgroup (n = 15, 95% CI 78.2 to 100). All eight classification disagreements fell along the depression–comorbid boundary, and the AI classification agreed with the patient's own self-categorisation in five of the eight. Performance was consistent across the two sites (Delhi 94.6%, Kerala 91.4%, p = 0.527) and across age, sex, and education subgroups. Classical autonomic markers separated the clinical and reference groups with moderate to large effect sizes (d = 0.53 to 0.84). Two of the physiological constructs of the development-cohort, namely Hypoactivity and Experience, were reproduced at the group level in independently diagnosed patients. A per-minute slope analysis of the tonic EDA signal showed a transdiagnostic reversal, where the baseline settled downward during rest in the reference group and drifted upward across all three clinical subgroups (Kruskal–Wallis p = 3.0 × 10⁻²²). Conclusions Resting-state autonomic AI analysis provides objective diagnostic support for anxiety and depressive disorders, with Cohen’s κ = 0.884 (95% CI 0.789 to 0.956) against psychiatrist-assigned DSM-5 diagnoses in routine outpatient practice. Direct comparison with published inter-rater or test-retest reliability estimates is not appropriate given differences in methodology and reference standards. The platform's low cost, portability, and suitability for non-specialist operators support scalability in resource-limited settings. External validation in primary care, community, telehealth, and non-Indian populations is the next step for the programme. Trial registration: Clinical Trials Registry of India, CTRI/2023/07/054984, registered July 2023.
Language: English
Field: Medicine
Open Access status: green
OpenAlex ID: W7235521219
Cites: 36
Cited by:
Related to: 10