Machine learning for risk prediction of infectious diseases: a scoping review of application, predictive performance, and reporting gaps.
TL;DR
ML demonstrates substantial potential for infectious disease risk prediction, however, methodological improvements-especially calibration assessment, external validation, and standardized reporting-are urgently needed.
Abstract
Background
Infectious diseases remain a major cause of global morbidity and mortality, with machine learning (ML) emerging as a promising approach for risk prediction. However, the heterogeneity of ML applications across diseases, algorithms, and validation practices impedes evidence synthesis and identification of best practices.
Objective
This scoping review aims to systematically map the application of ML algorithms for infectious disease risk prediction, describe disease targets, study populations, data sources, performance metrics, and identify methodological strengths, limitations, and research gaps.
Method
We conducted a scoping review following JBI methodology and PRISMAScR guidelines. Four databases (PubMed, Scopus, Science Direct, IEEE Xplore) were searched from inception to 31 May 2026. Two reviewers independently screened records, extracted data, and assessed quality using PROBAST + AI.
Results
From 168 included studies, ML prediction spanned individual (13.7%), population (56.0%), and spatial (30.4%) levels. Respiratory infections (44.6%) and COVID-19 (23.8%) dominated, while neglected tropical diseases comprised 19.1%. XGBoost, Random Forest and LSTM variants showed high performance, with spatial models yielding the highest median AUC (∼0.90), followed by population (0.84) and individual (0.82) models. Critical gaps included rare calibration reporting (3.0%), limited external validation (10.7%), and severe geographic skew-43.5% from China/USA, only 8 from low-income countries.
Conclusion
ML demonstrates substantial potential for infectious disease risk prediction. However, methodological improvements-especially calibration assessment, external validation, and standardized reporting-are urgently needed. Future research should prioritize multi-level modeling approaches and contextually adapted models for high-burden regions.