A novel fever prevalence indicator for early warning of acute respiratory infections using wastewater-based epidemiology and machine learning.

The rapid spread of acute respiratory infections (ARIs) poses a major challenge to global public health, yet current surveillance systems relying on sentinel hospitals and laboratory testing often lack timeliness and precision. In this study, we present an integrated framework combining wastewater-based epidemiology (WBE) with machine learning to establish a fever prevalence-based early warning model for ARIs. Using wastewater data on paracetamol and ibuprofen consumption (n = 603) from Dalian, Beijing, Guangzhou, and other cities, we evaluated statistical, unsupervised, and supervised learning approaches (11 models) for anomaly detection. Based on model validation and evaluation, the interquartile range, isolation forest, neural networks, and logistic regression models demonstrated favorable performance. However, only the warning thresholds derived from the interquartile range method (3.3%) and isolation forest (3.5%) indicated that November 2021 was a non-epidemic period and November 2024 was an epidemic period, which is consistent with the influenza-like illness surveillance results from sentinel hospitals. Our findings demonstrate that WBE, when coupled with machine learning, can provide timely and precise community-level monitoring of fever prevalence, while also offering early-warning capability for fever-dominant emerging respiratory infections and previously unidentified etiological agents such as "Disease X". This approach effectively addresses the limitations of conventional surveillance, such as incomplete coverage and delayed response, and provides a scalable, evidence-based tool to strengthen public health preparedness and response.