Publication:

Artificial intelligence for drowning prevention: a scoping review for public health practice

Drowning remains a major and preventable cause of injury mortality worldwide, with a disproportionate burden in low- and middle-income countries. Timely recognition of distress and rapid rescue can be missed within existing supervision systems. Artificial intelligence (AI) and machine learning (ML) technologies are increasingly proposed to strengthen drowning prevention through hazard prediction, real-time surveillance, wearable alerts, and drone-assisted response. We conducted a scoping review (2015–2023) of studies describing AI/ML applications for drowning prevention, detection, or rescue across electronic databases and selected grey literature. Evidence was synthesized by functional domain (hazard prediction, incident detection, rescue support) and conceptually mapped to the Haddon Matrix to examine potential contributions across pre-event, event, and post-event phases. Given heterogeneity in study designs and outcomes, findings were appraised descriptively with attention to validation context and implementation considerations. Thirty-eight studies met inclusion criteria. Most evaluated computer-vision systems in controlled pool or coastal environments, with fewer examining wearable sensors, drone-assisted localization, or environmental risk prediction. Reported performance metrics were frequently derived from simulation or prototype datasets, with limited independent field validation. Few studies assessed integration into lifeguard workflows, cost-effectiveness, or applicability in resource-limited settings. AI/ML technologies may complement established drowning prevention strategies by enhancing hazard surveillance and supporting rapid response. However, the current evidence base remains largely pre-implementation. Future research should prioritize real-world validation, transparent reporting, and evaluation of integration within injury prevention systems to determine whether these tools reduce drowning in practice.