JMIR Publications

ABSTRACT

Background:

Influenza surveillance faces delays that hinder timely intervention. While search engine queries offer real-time monitoring, they are susceptible to noise and may not reflect true incidence. Digital antiviral prescription data capture genuine health-seeking behavior at the point of care, offering a potential alternative. However, the predictive validity of such data for influenza forecasting and their optimal integration within a multi-source modeling framework remains insufficiently evaluated.

Objective:

To evaluate the predictive utility of digital antiviral prescription data for influenza forecasting and propose a methodological framework linking source validation to multi-source predictive modeling.

Methods:

We conducted a longitudinal study using weekly province-level data from Jiangxi, China (2022–2024), integrating six data streams. We examined temporal relationships between digital prescription rates and influenza positivity rates using cross-correlation and Granger causality tests. We constructed 38 predictive features and, using a rolling forecasting method, evaluated LASSO alongside ARIMA, Random Forest, Prophet, and Explainable Boosting Machine for 1- to 8-week ahead predictions, with performance assessed using R² and RMSE.

Results:

Digital prescription rates exhibited a leading effect on influenza positivity rates, with the strongest correlation observed at a 1-week lead time, a finding further corroborated by Granger causality tests (P < .001 for all lags). LASSO demonstrated superior or comparable performance across most horizons, achieving the highest mean R2 (0.89) and lowest mean RMSE (4.52). LASSO consistently outperformed alternatives for 5- to 8-week horizons. Top predictors included 1- and 5-week lagged positivity rates and 1-week lagged prescription rates.

Conclusions:

Digital antiviral prescription data is a valid real-time leading indicator of influenza activity. LASSO provides an optimal balance of accuracy and interpretability within a multi-source data framework, offering a practical pathway for provincial influenza early warning.