Estimated prebiotic intake and depression in US adults: repeated cross-sectional analyses of NHANES 2015-2018 and 2021-2023.

The Syrian conflict, now in its fourteenth year, has produced one of the most severe mental health emergencies in contemporary humanitarian history. Epidemiological surveys consistently report posttraumatic stress disorder (PTSD) prevalence of 23–43%, major depression of 30–51%, and generalised anxiety of 18–49% amongst conflict-exposed populations (–). This burden is compounded by near-total collapse of psychiatric infrastructure: Syria has 0.37 psychiatrists per 100,000 people nationally, whilst Northwest Syria—home to nearly four million people—relies on just two practicing psychiatrists (, ). Mental health services in Northwest Syria are sustained by NGOs operating under severe constraints, with the WHO mhGAP Intervention Guide () serving as the standard for task-shifting to non-specialist community health workers (CHWs). Bou-Orm et al. () documented the operational challenges facing MHPSS services in this fragmented geography. AI-powered mental health interventions have attracted growing attention as instruments of scale (, ). Multi-agent architectures, in which specialised LLM-based agents collaborate on complex clinical tasks, have shown improved assessment accuracy over single-model approaches on English-language benchmark datasets (–); we are explicit that this evidence concerns benchmark performance and not clinical deployment outcomes, and certainly not deployment in humanitarian settings. Direct transfer of these tools to Arabic-speaking, post-conflict settings is structurally constrained rather than already demonstrated to fail: Arabic NLP lags behind English in clinical domains (, ), Syrian dialect is nearly absent from NLP benchmarks (), and the ethical terrain of algorithmic deployment in fragile contexts is undertheorised (). To our knowledge, no peer-reviewed deployment of multi-agent mental health AI in Syrian Arabic-speaking populations has been published; this absence—rather than a demonstrated failure—is the starting point of the present analysis. Throughout, we use mental health triage in a deliberately narrow sense: an initial-contact decision-support task that detects symptom-relevant signals from a user-initiated or community-health-worker–mediated conversation, assigns a four-tier severity classification (mild, moderate, severe, crisis), recommends a referral destination aligned with mhGAP tiered care, and supports longitudinal re-contact for deterioration monitoring. Triage in this sense is consultative rather than diagnostic: it produces a recommendation that a human supervisor accepts, modifies, or overrides, and the boundary of clinical responsibility rests with that supervisor rather than with the system, which makes no clinical diagnosis.