Real-Time Influenza Dynamics: Nature Paper Analyzes 21 Million Chinese Digital Prescriptions

Revolutionizing Flu Surveillance with Digital Prescriptions

In a groundbreaking study published in npj Digital Medicine, a Nature journal, researchers from China's leading medical institutions have harnessed 21.08 million digital prescription records to unlock real-time insights into influenza dynamics across the country. This innovative approach addresses longstanding delays in traditional influenza surveillance systems, which often lag by 1-2 weeks, hindering timely public health responses. By analyzing transactions from Meituan Health, China's dominant on-demand medication delivery platform capturing about 70% of the online pharmacy market, the team demonstrated how prescription data for antiviral drugs like oseltamivir and baloxavir marboxil serve as a robust proxy for epidemic activity.

The study, spanning January 2022 to December 2024 across all 31 provinces, reveals prescription rates varying dramatically from 0.22 to 8.22 per 100,000 population—a 37-fold difference. These patterns synchronized strongly with laboratory-confirmed influenza positivity rates, with Spearman's correlation coefficients ranging from 0.29 to 0.80 (p < 0.05 in all provinces). This marks a pivotal advancement in digital epidemiology, particularly timely as China grapples with resurgent flu seasons post-COVID-19 restrictions.

Influenza Burden in China: A Persistent Public Health Challenge

Seasonal influenza imposes a heavy toll in China, with estimates of 50 to 170 million infections annually between 2010 and 2020, contributing to thousands of deaths. Recent data indicate peaks in southern provinces reaching 59.62% influenza-like illness (ILI) rates during 2022/2023—the highest in a decade—while northern regions hit 57.60%. In late 2025, at least 17 provinces reported high epidemic levels, overwhelming hospitals and prompting school closures in some areas.

China's vast population (over 1.4 billion) and diverse climates exacerbate transmission: southern regions experience semi-annual peaks (summer and winter), while northern areas see single winter surges. Traditional challenges include low healthcare-seeking rates (24-31% for severe cases), underreporting of mild illnesses, and reliance on sentinel hospitals for weekly lab data. Vaccination coverage remains low, around 3-5% nationally, far below global targets, amplifying the need for enhanced surveillance.

Digital platforms like Meituan Health bridge these gaps by capturing real-world therapeutic demand from urban and rural users alike, reflecting both mild and moderate cases that evade lab detection.

Methodology: From Big Data to Causal Insights

Led by Rui Shen and corresponding authors Luzhao Feng from Peking Union Medical College (PUMC) and Hongbo Liu from China Medical University (CMU), the research employed convergent cross-mapping (CCM)—a dynamical systems tool—to quantify predictive lead times and causality. CCM tests whether one time series can predict another, revealing nonlinear couplings absent in linear correlations.

• Data Aggregation: Anonymized provincial-level prescriptions for flu antivirals, cross-validated against Chinese CDC lab positivity, Baidu search indices, and environmental factors (temperature, humidity, pollutants).
• Causal Analysis: Bidirectional CCM confirmed prescriptions' forward (28/31 provinces) and reverse (22/31) causality with lab flu, unlike unidirectional search data (2/31 reverse) or environmentals (0-5/31).
• Forecasting Model: Hybrid spatiotemporal deep learning with Graph Neural Networks (GNN) for spatial dependencies, Mamba for efficient long-sequence modeling, and LSTM for temporal patterns. Trained on 2022-2023 data, tested on 2024.

This rigorous validation distinguishes prescriptions as true epidemic signals, filtered through clinical (physician approval) and economic (payment) barriers.

Key Findings: 2-Week Lead Time and Superior Forecasting

Digital prescriptions outperformed environmental predictors by providing a 2-week lead (Δρ_CCM = 0.339, p<0.001 in 28/31 provinces), matching search indices but with causal proof. They also showed heightened sensitivity to air pollutants (14-26/31 provinces vs. 3-7/31 for labs), capturing behavioral responses like reduced outdoor activity during poor air quality.

The forecasting model excelled: mean absolute error (MAE) of 1.166 for 96-day horizons, under 3.0 in 29 provinces, surpassing baselines like LSTM (MAE 1.380). Pearson r exceeded 0.8 in 28 provinces, with best performance in less urbanized areas (e.g., Xizang MAE 0.193).

These results underscore prescriptions' role in multi-source surveillance.

Leading the Charge: Peking Union Medical College and China Medical University

PUMC's School of Population Medicine and Public Health, under Prof. Luzhao Feng—Deputy Dean and prolific epidemiologist—anchors this research. Feng's team, part of the Key Laboratory of Pathogen Infection Prevention (Ministry of Education), has pioneered influenza modeling and vaccine effectiveness studies. PUMC, a top-tier institution collaborating with Chinese CDC, exemplifies how higher education drives national health security.

CMU's School of Public Health, contributing via Xueying Xu and Hongbo Liu, brings northeastern perspectives, enhancing spatiotemporal granularity. These universities integrate big data with public health expertise, training future leaders via labs like State Key Laboratory of Respiratory Health.Discover university opportunities in China.

Their code is open-source on GitHub, fostering global academic collaboration.

Overcoming Traditional Surveillance Hurdles

China's sentinel system—relying on ILI sentinel hospitals and lab tests—faces delays, geographic biases toward urban severe cases, and weekly reporting. Digital prescriptions counter this with daily, nationwide granularity, capturing 70-80% of symptomatic cases via platforms ubiquitous in daily life.

• Real-time: 24-hour availability vs. weeks.
• Comprehensive: Mild cases included.
• Scalable: No new infrastructure needed.

Unlike Google Flu Trends' pitfalls (media-confounded searches), prescriptions pass clinical-economic filters, ensuring reliability.

Advanced Forecasting: 96 Days Ahead

The GNN-Mamba-LSTM model leverages graph structures for inter-province flows, Mamba's state-space efficiency for long horizons, and LSTM's memory. Tested prospectively on 2024 data, it achieved high fidelity even in high-burden metros like Beijing (39.67% error reduction vs. baselines).

Provincial variations highlight needs: Western provinces forecast easiest due to stable patterns; eastern metros require refined mobility inputs.

Public Health Implications and Policy Pathways

This proxy enables early warnings, optimizing resource allocation amid China's flu surges. Integration with CDC systems could boost vaccination campaigns (currently low at ~3%) and antiviral stockpiling.Read the full Nature study.

Environmentally, heightened sensitivity flags pollution-flu interactions, informing air quality alerts. Globally, as digital pharmacies boom (projected $364B by 2030), causal-validated proxies promise scalable surveillance.

Stakeholder Perspectives: From Researchers to Platforms

Prof. Feng emphasizes: "Digital prescriptions bridge the gap between symptoms and labs, empowering proactive responses." Meituan Health's data-sharing exemplifies public-private synergy.

Challenges persist: User demographics skew younger/urban; OTC/self-care missed. Future: Multimodal fusion with wastewater, wearables.

Photo by xiaoyu xie on Unsplash

Future Outlook: Digital Epidemiology's Horizon

Chinese universities like PUMC and CMU are at the forefront, training AI-savvy epidemiologists. As China's digital health ecosystem matures, expect expanded applications to RSV, COVID hybrids.

Prospects include personalized alerts via apps, boosting equity in rural areas. For academics, this opens research jobs in digital public health.

In conclusion, this Nature paper positions university-led innovation as vital for China's health resilience. Explore Rate My Professor for insights on public health faculty, higher ed jobs in epidemiology, or career advice for aspiring researchers. University jobs await—post a job today.