Adaptive Negative-Control Calibration for Multi-Platform Digital Nowcasting

Madamalla (2025) shows multi-platform models outperform single-platform approaches for outbreak detection, but the review also flags demographic, linguistic, and misinformation biases that vary over time and by platform. Borrowing the negative-control strategy from Bayesian vaccine safety surveillance (Bu et al., 2023), this project proposes a hierarchical model that learns an empirical bias distribution per platform-region-time using a large set of negative control queries/topics (terms/events structurally similar to the disease signal but with no causal link, e.g., sports team searches, non-health trending news). We layer in online “media shock” detection—change-points triggered by exogenous events or algorithmic manipulation (Gombar, 2025)—to adaptively reweight platforms during periods when bias spikes. This departs from current digital epi practice by explicitly modeling bias as a time-varying latent process informed by controls, rather than treating platform signals as noisy proxies to be fused. The payoff is faster, more reliable signal detection with principled uncertainty calibration during volatility (e.g., misinformation waves), and a general recipe for bias-aware nowcasting that can plug into hybrid systems Madamalla advocates.

References:

1. Digital Epidemiology in Action: A Cross-Platform Review of Social Media and Internet-Based Surveillance for Infectious Disease Outbreaks. Esha Madamalla (2025). International Journal of Innovative Science and Research Technology.
2. Algorithmic Manipulation and Information Science: Media Theories and Cognitive Warfare in Strategic Communication. Marija Gombar (2025). European Journal of Communication and Media Studies.
3. Bayesian safety surveillance with adaptive bias correction. Fan Bu, M. Schuemie, A. Nishimura, Louisa H. Smith, K. Kostka, Thomas Falconer, Jody-Ann McLeggon, P. Ryan, G. Hripcsak, M. Suchard (2023). Statistics in Medicine.