Causal Policy Analytics from Infectious Disease Databases

Abstract

This paper develops a causal policy analytics framework applying structural causal modelling, propensity score matching (PSM), and sensitivity analysis to infectious disease surveillance databases. Using linked administrative records from 847 sub-national jurisdictions across 42 countries covering three specified respiratory pathogen episodes—seasonal influenza A/H3N2 (2017–2018), SARS-CoV-2 (2020–2022), and respiratory syncytial virus (RSV, 2022–2024)—the study estimates average treatment effects on the treated (ATT) for five non-pharmaceutical interventions: school closures, mask mandates, travel restrictions, vaccination campaigns, and contact tracing. A logistic regression propensity score model incorporating 14 measured confounders and concurrent-NPI indicators, with 1:1 nearest-neighbour matching without replacement, controls for observed confounding. Under the stated identification assumptions, vaccination campaigns show the largest estimated reduction in standardised case incidence (−24.1%, 95% CI [−30.8, −17.4]), followed by school closures (−18.4%) and contact tracing (−15.6%). Rosenbaum sensitivity analysis indicates that vaccination and school closure estimates are robust to moderate unmeasured confounding (Γ > 2.4), while travel restriction estimates are sensitive to relatively small hidden biases (Γ = 1.6). These findings demonstrate that causal inference methods applied to routine surveillance data can yield policy-relevant evidence, though interpretation must account for interference, policy co-occurrence, and measurement limitations inherent in observational epidemic data.

Keywords: Causal inference; infectious disease; propensity score matching; non-pharmaceutical interventions; sensitivity analysis; policy evaluation; target trial emulation