Lymphatic filariasis risk prediction models: A systematic review of modelling approaches, predictors, validation practices, and methodological quality

Background Risk prediction models are increasingly used to support Lymphatic filariasis (LF) surveillance, identification of residual transmission foci, and elimination planning. However, their methodological quality, validation practices, and translational capacity have not been systematically evaluated. Methodology/principal findings To address this gap, we conducted a systematic review following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines to characterise the methodological approaches, predictor domains, validation practices, model performance assessment, and risk of bias of published LF risk prediction models. The review was not prospectively registered. Studies developing statistical, Bayesian spatial, ecological niche, machine learning (ML), or hybrid models to estimate LF prevalence, infection status, or transmission risk were included. Data were extracted using the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS), and risk of bias was assessed using the Prediction model Risk Of Bias ASsessment Tool (PROBAST). Twenty-six studies published between 2000 and 2025 were included. Most studies were conducted in Africa and India and relied on spatially aggregated epidemiological and environmental data. Climatic (73.1%) and environmental (76.9%) predictors predominated and frequently co-occurred, whereas socio-demographic (57.7%), programmatic (7.7%), and vector-related (7.7%) variables were less consistently incorporated. Bayesian and spatial models were mostly aligned with expert-driven predictor selection and internal validation, whereas ML and hybrid modelling approaches aligned with algorithm-based predictor selection and cross-validation. Despite increasing methodological diversification, external validation (15.4%) and calibration assessment (7.7%) were rarely reported. All studies were assessed as having high or unclear risk of bias in the analysis domain. Conclusions/significance LF risk prediction modelling is methodologically diverse but remains predominantly reliant on internal validation. Limited external validation and calibration assessment constrain generalisability and reliability of published models. Improving external validation, calibration assessment, methodological transparency, and reporting standards will be essential for developing robust prediction models capable of supporting elimination verification and post-elimination surveillance.