Predicting drowning from sea and weather forecasts: development and validation of a model on surf beaches of southwestern France

Objective To predict the risk of drowning along the surf beaches of Gironde, southwestern France. Methods Data on rescues and drownings were collected from the Medical Emergency Center of Gironde (SAMU 33). Seasonality, holidays, weekends, weather, and sea conditions were considered potentially predictive. Logistic regression models were fitted with data from 2011–2013 and used to predict 2015–2017 events employing weather and ocean forecasts. Results Air temperature, wave parameters, seasonality, and holidays were associated with drownings. Prospective validation was performed on 617 days, covering 232 events (rescues and drownings) reported on 104 different days. The area under the curve (AUC) of the daily risk prediction model (combined with 3-day forecasts) was 0.82 [95% confidence interval (95% CI) 0.79−0.86]. The AUC of the 3-hour step model was 0.85 (95% CI 0.81−0.88). Conclusions Drowning events along the Gironde surf coast can be anticipated up to 3 days in advance.Preventative messages and rescue preparations could be increased as the forecast risk increased, especially during the off-peak season, when the number of available rescuers is low.