Predictive modelling approaches for childhood diarrhoeal disease: a systematic review (2000–2025)

Background Childhood diarrhoeal disease remains a leading cause of child mortality worldwide. Although predictive models have been increasingly proposed for early warning, their methodological quality and validation have not been systematically appraised using formal risk of bias frameworks. This systematic review aimed to identify, describe, and critically appraise individual-level prediction models and population-level forecasting studies for childhood diarrhoeal outcomes, focusing on model development, validation, reporting quality, and operational readiness. Methods Following PRISMA 2020 guidelines (PROSPERO: CRD420251232269), we searched seven databases and grey literature sources from January 2000 to September 18, 2025. We included studies that developed or validated multivariable prediction models for diarrhoeal outcomes in children, including statistical, machine learning, and hybrid approaches. Individual-level models were assessed using the Prediction Model Risk of Bias Assessment Tool (PROBAST), and population-level studies were appraised qualitatively for methodological credibility, with emphasis on validation. Results From the 20,662 records screened, 38 studies met the inclusion criteria. Publications accelerated after 2020, with 78.9% of included studies published in this period. The evidence base comprised 25 single-country studies and 13 multi-country or global analyses. Statistical models were most commonly used, as reported in 18 studies, while ensemble or hybrid approaches such as hybrid statistical–machine learning models were applied in 11 studies, and machine learning methods in eight studies. Predictor selection varied widely, with socio-demographic and household factors included in 23 studies and environmental or climatic variables in 18. Among the 15 individual-level prediction models assessed using PROBAST, five showed high overall risk of bias. Of the 19 population-level forecasting studies appraised qualitatively, only three undertook rigorous out-of-sample temporal validation. The remaining four studies were classified as other modelling categories and were summarised descriptively. Conclusions Research on childhood diarrhoeal prediction has expanded rapidly across individual-level, population-level, and related classification models. Although some studies reported stronger validation practices, these remained uncommon, leaving evidence for transportability and implementation readiness limited for most models. Future progress will depend on closer alignment between modelling objective, validation strategy, appraisal framework, and transparent reporting.