A systematic review published in BMC Infectious Diseases indicates that artificial intelligence and machine learning models can improve infectious disease surveillance by detecting epidemiological shifts earlier than traditional systems, though prospective deployment remains largely unvalidated. Researchers Chunmin Pu, Yang Nian, Shunjiang Ma, Yuyan Tan, Meng He, Xiuzhu Li, and Yu Huang evaluated modeling methodologies across global investigations, assessing how computational frameworks handle predictive disease tracking. The team paid special attention to enteric infections, such as rotavirus gastroenteritis, where conventional monitoring systems often struggle to provide sufficient predictive lead time for timely public health action.
According to the authors, enteric infections place a heavy burden on children and strain resources in low-resource healthcare systems. Standard surveillance mechanisms continue to rely on laboratory-confirmed case reporting. Because these diagnostic records move slowly through routine hospital information systems, electronic medical records, and laboratory information systems, public health authorities frequently experience delayed notification. Digital health infrastructure combined with artificial intelligence now offers potential routes to reconstruct disease tracking around integrated, near-real-time data sources to overcome these longstanding operational delays.
To assess the current state of computational surveillance, the investigators conducted their systematic review in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 guidelines. They searched PubMed and IEEE Xplore for studies published through July 2026, followed by a supplementary PubMed literature search in September 2026. The search targeted statistical, machine learning, deep learning, and multi-source artificial intelligence approaches applied to infectious disease surveillance or forecasting, prioritizing enteric diseases and methodologically transferable models. Additional relevant investigations were identified by screening reference citations.
Study selection and data extraction were conducted independently by two reviewers, who also evaluated methodological quality and risk of bias across the literature. For twenty-three prediction-model studies, the reviewers applied the formal Prediction model Risk Of Bias ASsessment Tool, known as PROBAST. Five ecological association studies underwent narrative appraisal, whereas five contextual studies were analyzed separately without formal bias appraisal. Across the entire search, thirty-three studies satisfied the inclusion criteria, separating into twenty-eight formal modeling investigations and five distinct contextual analyses.
The twenty-eight formal investigations encompassed five statistical models, eight machine learning approaches, four deep learning architectures, and eleven multi-source digital surveillance frameworks. The remaining five contextual studies explored environmental monitoring, virus-host associations, computational simulations, surveillance signals, and network dynamics. Across these studies, data sources broadened over time. While earlier methods relied primarily on meteorological indicators and clinical records, newer models integrated electronic medical records, telehealth data, wastewater sampling, internet search trends, social media metrics, water-quality measurements, and genomic sequencing streams.
The synthesis demonstrated that machine learning and deep learning methodologies were increasingly implemented alongside conventional statistical techniques to model complex nonlinear and lagged exposure-response dynamics. These computational frameworks demonstrated particular utility in integrating heterogeneous data streams. Furthermore, multi-source digital surveillance systems—specifically those incorporating wastewater-based monitoring and syndromic data designs—consistently reported earlier outbreak signal detection than single-source monitoring designs, allowing faster identification of epidemiological changes.
Despite these predictive capabilities, the reviewers identified significant methodological limitations throughout the current literature. Most available evidence relies on retrospective study designs spanning single-site, national, and multinational datasets. Crucially, prospective real-world deployment and external validation were largely absent across the evaluated literature. The authors concluded that practical operational adoption will depend less on retrospective benchmark accuracy and far more on robust external validation, model interpretability, and seamless integration with existing institutional surveillance infrastructure.
Subject of Research: Machine learning and multi-source data integration in infectious disease surveillance
Article Title: Machine learning–based surveillance and forecasting of infectious diseases: a systematic review of modeling approaches and multi-source data integration
Article References: Pu, C., Nian, Y., Ma, S., Tan, Y., He, M., Li, X., & Huang, Y. (2026). Machine learning–based surveillance and forecasting of infectious diseases: a systematic review of modeling approaches and multi-source data integration. BMC Infectious Diseases. https://doi.org/10.1186/s12879-026-14824-z
Image Credits: AI Generated
DOI: 10.1186/s12879-026-14824-z
Keywords: machine learning, infectious disease surveillance, disease forecasting, deep learning, enteric infections, rotavirus gastroenteritis, multi-source digital surveillance, wastewater-based epidemiology, syndromic surveillance, electronic medical records, telehealth, genomic data, PRISMA 2020, PROBAST, external validation, lagged exposure-response, hospital information systems, systematic review
News Source: Blake Davidson. (November 18, 2026). Machine Learning Aids Disease Surveillance but Lacks Real-World Prospective Validation. Scienmag.



