A four-domain framework for AI-enabled mobile health: from sensing to equitable governance

Artificial intelligence (AI) is widely expected to strengthen mobile health (mHealth) responses to the dual burden of non-communicable and infectious disease, yet translation into demonstrable population health benefit remains limited. This narrative review conceptualizes AI-mHealth as the mobile and wearable layer of an Internet of Things (IoT)-enabled connected health ecosystem. Because the literature has largely examined sensing, modeling, and implementation as discrete problems, the dependencies that govern system-level utility remain poorly characterized. We therefore appraise the evidence across four linked domains: Sensing, Inference, Intervention, and Governance. We searched PubMed/MEDLINE and the Web of Science Core Collection from October 2015 to September 2026. Sensing performance depended on device and measurement context. Wrist-worn photoplethysmography showed no consistent skin-tone-related error at rest, whereas error in darker skin tones exceeded that in lighter skin tones more than fourfold at higher exercise intensities. Inference models improved risk stratification but lacked the interpretability required to support clinical accountability, and model compression for on-device deployment degraded subgroup performance unevenly. Just-in-time adaptive interventions improved proximal response, although effect durability varied across microrandomized trials, and remote heart failure management has not yet demonstrated a consistent effect on readmission. Algorithmic bias, the digital divide, deficiencies in data stewardship and cybersecurity, limited human oversight, and regulatory latency form a mutually reinforcing system of risk across the AI-mHealth lifecycle. Our principal finding is an asymmetry: upstream limitations propagate downstream automatically, whereas their correction requires deliberate institutional design. We propose an operational roadmap of four implementation phases, from technical readiness to accountability at scale. Each phase is mapped to the framework domains it engages and to the specific Sustainable Development Goal 3 targets it serves, spanning communicable disease screening, chronic disease management, workforce capacity, and universal coverage. Aggregate accuracy concealed reliability loss in the subgroups most in need of intervention, and the roadmap therefore makes subgroup-level equity auditing a condition of deployment.