Socioeconomic-Aware Pandemic Simulation via a Hybrid SVIR–Machine Learning Framework

Objectives: Pandemic policy choices can generate large and uneven economic consequences through mortality and morbidity, disruptions to labour supply and productivity, and pressure on health-system capacity. Sound decision-making, therefore, depends on simulation tools that can represent how epidemic dynamics evolve over time under alternative policies. During the last pandemic, simulation models were widely used to support health policy decisions, often relying on aggregate indicators and limited information on socioeconomic conditions and policy measures, especially under data uncertainty (Moran et al., 2025). Compartmental epidemic models such as SVIR (susceptible–vaccinated–infected–recovered) provide an interpretable mechanistic description of disease transmission and vaccination dynamics. Recent AI– mechanistic hybrids can learn time-varying transmission and improve forecasting, but socioeconomic access and vaccine-acceptance mechanisms remain only partially integrated in policy-relevant ways (Bousquet et al., 2022; Grimm et al., 2022). We propose a socioeconomic-aware pandemic simulation approach: a hybrid SVIR–machine learning (ML) framework that learns time-varying, group-specific epidemic drivers from socioeconomic conditions and policy measures, while preserving an interpretable transmission structure. The proposed hybrid SVIR–ML framework explicitly links epidemic dynamics to socioeconomic measures and standard health and economic indicators. By doing so, it illustrates a possible shift in how pandemic simulation outputs are measured and interpreted, moving beyond average case counts while remaining compatible with common indicators used in health policy analysis. Mechanistic models remain central for scenario analysis, but can be hard to calibrate when behaviour and policy change quickly; AI methods can extract signals from rich data, but may lack structure for counterfactual policy evaluation (Ye et al., 2025). Methods: We couple a stratified SVIR model (geography ×socioeconomic status ×age) with ML functions that map measurable determinants to a small set of epidemiologically meaningful parameters. Specifically, time-varying transmission and contact scaling, βg,r (t), and vaccine acceptance/uptake, ag (t) ∈(0,1), are learned from surveillance time series together with contextual indicators (e.g., mobility). Non-pharmacological interventions, seasonality) and socioeconomic/access measures (e.g., deprivation indices, income/education proxies, travel time or provider density). Where appropriate, selected clinical parameters may be fixed from external evidence to focus estimation on behavioural and contextual drivers. Model outputs are designed to support health-economic analysis by translating stratified infections and outcomes into standard indicators (e.g., healthcare utilisation and costs, QALYs/DALYs, and distributional summaries using distributional cost-effectiveness concepts) (Asaria et al., 2016). Empirically, we draw on evidence that uptake and infection risk can vary systematically with deprivation (Lopez-Sanchez et al., 2024). Results: The framework is intended to improve calibration under rapid behavioural and policy shifts and may help quantify how socioeconomic differences in access and acceptance contribute to observed epidemic dynamics. It has the potential to support counterfactual evaluation of targeted policies while making explicit the efficiency–equity trade-offs relevant for health-demand management and resource planning (e.g., demand surges for outpatient care, hospital beds, and vaccination services). Discussion: Embedding socioeconomic determinants within a mechanistic backbone and reporting results in standard health and economic indicators may strengthen the policy relevance of pandemic simulations. The proposed framework aims to enable equity-informed preparedness and to support health-economic appraisal of interventions under realistic constraints and uncertainty.