Understanding and managing the risk of water-related diseases under hydrometeorological extremes

Leptospirosis is a neglected tropical disease caused by pathogenic Leptospira sp. We present lessons learned from a 3 year project studying the links between the water environment and leptospirosis incidence in Malaysia, where the disease is considered endemic, and where higher transmissions have been linked to monsoonal floods. We use a risk framework to identify the hydrometeorological and environmental drivers; occupational, recreational and domestic exposure routes; and sociodemographic factors that contribute to infection. Our analysis reveals that the link between hydrometeorology and cases is present but highly nonlinear; through different statistical and machine learning algorithms, we find that hydrometeorological signals only result in moderately performing regression models but accurate classification models. A higher predictive accuracy can be achieved by a black-box model (deep learning) than by more explanatory, grey-box models (quasi-Poisson and random forest). We further identify spatial heterogeneity in the controlling environmental factors, but these findings are challenging to merge with the temporal models due to the static nature of available environmental data. Lastly, a case-control questionnaire study reveals that "time spent outdoors" most increases the likelihood of individual infection; however, integrating this into predictive modelling is not straightforward. We lastly share our experiences with low-cost hydrological sensing in remote forested recreational areas as an effort to improve their spatial representation. We identify key challenges/research directions going forward in understanding the risk of diseases under a changing water environment: (1) developing a holistic model structure that can capture the role of all underlying drivers (2) consideration of the scale of modelling - individual vs. spatiotemporal, and the data requirements at each scale (3) the need for spatiotemporal data on rodents to explicitly represent the vector infection pathway (4) continuing discourse to bridge the knowledge disciplines and methodologies, including improving stakeholder communication of quantitative outputs and limitations of models.