Modelling approaches for predicting the distribution of skin NTDs: A systematic review

Skin neglected tropical diseases (NTDs) such as cutaneous leishmaniasis, lymphatic filariasis, mycetoma, and podoconiosis affect millions in endemic regions, but are under-recorded despite causing significant burdens. Predictive modelling has been used to estimate the distribution and prevalence of some of these diseases, and predictions may be useful for identifying at-risk populations and guiding interventions. This review synthesises the literature on modelling approaches to predict skin NTD distributions, aiming to identify prevalent methodologies, evaluate their strengths and limitations, highlight research gaps, and provide recommendations for enhancing their utility.

We conducted a systematic literature review from three databases and included studies published from 2000-2024. Studies were included if they employed statistical models or machine learning algorithms to predict the distribution of skin NTDs. Two independent reviewers screened titles, abstracts, and full texts. Data extracted included disease, study region, source of epidemiological data, model types and predictors.

From 2,870 retrieved records, 68 met the inclusion criteria. The most modelled skin NTDs were cutaneous leishmaniasis (n = 26) and lymphatic filariasis (n = 18). Geostatistical modelling was the most common approach, followed by ecological niche modelling, with MaxEnt and generalised linear models constituting the predominant model types. Common environmental covariates included climate, land cover and land use, elevation, and soil data. The types of epidemiological data varied, with many studies relying on passive surveillance and pseudoabsence data. The risk of bias was high among ecological niche models.

Environmental and geostatistical models can inform targeted interventions for skin NTDs, aiding efficient resource allocation and public health planning. However, data limitations, especially the absence of true absence data, underreporting and variations in surveillance sensitivity, can reduce model accuracy and undermine decision-makers' confidence. Future studies should focus on incorporating information about case identification into modelling frameworks, including a broader spectrum of environmental and socio-economic determinants, and ensuring validation across diverse geographic regions.