Real-time digital prescriptions unlock influenza dynamics: evidence from 21 million transactions.
retrospective_cohort · Level III
Where this comes from
- Record sourced from PubMed, PMID 41775925.
- Also identified by DOI 10.1038/s41746-026-02513-9 and PMC identifier 13076620.
- Licence recorded as CC BY-NC-ND.
- Because redistribution is not established, this page shows the abstract only. Follow the links below for the full text.
Abstract
Traditional influenza surveillance suffers from 1-2 week reporting delays that compromise outbreak response. We analyzed 21.08 million digital prescription transactions from China's largest on-demand medication delivery platform across 31 provinces (2022-2024) and demonstrated that prescription data provide causally validated epidemic proxies. Digital prescriptions exhibited 2-week predictive lead time (convergent cross-mapping skills Δρ<sub>CCM</sub> = 0.339; p < 0.001; 28/31 provinces), substantially exceeding environmental predictors (Δρ<sub>CCM</sub> = 0.032-0.196) while matching online search indices (Δρ<sub>CCM</sub> = 0.349). Critically, prescriptions demonstrated bidirectional causal coupling with laboratory-confirmed influenza positivity (forward: 28/31; reverse: 22/31 provinces). This dynamical signature was absent in online search (reverse: 2/31) and environmental variables (reverse: 0-5/31), distinguishing validated disease signals from confounded correlates. Prescriptions also exhibited greater environmental sensitivity compared with laboratory surveillance (air pollutants: 14-26/31 versus 3-7/31 provinces). Leveraging this validated proxy, a spatiotemporal deep learning framework integrating GNN, Mamba, and LSTM achieved 96-day forecasting of daily prescription rates (mean absolute error (MAE) = 1.166; MAE < 3.0 in 29/31 provinces). Digital prescriptions thus enable both immediate epidemic detection (24 h data availability) and actionable long-range forecasting, providing an additional validated data stream for multi-source epidemic surveillance.