Heterogeneous susceptibility inference in real time

Compartmental models remain the default tool for real-time outbreak analysis because they can be fitted rapidly as surveillance data accrue. The standard SIR model assumes homogeneous susceptibility to infection, leading to systematic overestimation of herd immunity thresholds and final epidemic sizes. We evaluate a one-parameter SIR extension capturing population-level heterogeneity in susceptibility at negligible computational cost. Using simulated outbreaks across realistic epidemiological settings, this heterogeneity can be estimated with increasing precision as outbreaks unfold. Ignoring heterogeneity results in biased and misleadingly precise estimates of the remaining epidemic size. Heterogeneity also remains estimable after a moderately effective intervention begins in fast-moving outbreaks, informing when the intervention can end. When a single intervention period is considered, with no reinstatement or vaccination, moderately effective interventions minimise the final epidemic size, a benefit amplified by heterogeneity. Real-time estimates of population-level heterogeneity, with quantified uncertainty, should be used to inform public health decision-making.Preprint. Not yet peer-reviewed. Code: https://github.com/t-refae/HSIR_fitting and https://github.com/t-refae/HSIR.