Importance of social inequalities to contact patterns, vaccine uptake, and epidemic dynamics

The study investigates how socio-demographic and economic inequalities shape epidemic dynamics through their effects on exposure (via social contacts, mobility, and work conditions) and outcomes (via health status, access to care, and protection attitudes). Traditional models largely stratify by age while neglecting other key heterogeneities. The research asks: which individual characteristics and subgroups most differentiate epidemic-relevant behaviours (contact patterns and vaccination uptake), and how do these differences translate into epidemiological outcomes? Using Hungary’s MASZK survey during COVID-19, the authors analyse variations in contacts and vaccination by education, employment, income, settlement, and other factors, and develop an extended age-structured SEIR framework that incorporates these social dimensions to quantify unequal epidemic burden.

Existing infectious disease models typically rely on age-stratified contact matrices and age-varying epidemiological parameters, often ignoring socio-economic and demographic heterogeneities beyond age. Prior work has documented that pandemics disproportionately affect certain groups, with inequalities arising from differences in exposure, compliance with NPIs, health status, and access to care. Calls to integrate social and behavioural factors into epidemic models have intensified, highlighting limited understanding of how socio-economic divides translate into differential transmission. Empirical studies during COVID-19 showed heterogeneous impacts of NPIs, mobility reductions, and contact patterns across socio-economic strata in multiple countries. This study addresses these gaps by jointly analysing contact patterns and vaccination uptake across social strata and embedding these findings into an extended compartmental model.

Data: Monthly cross-sectional, nationally representative CATI surveys (MASZK) in Hungary from 04/2020 to 07/2022 (~1000 respondents/month), recording proxy face-to-face contacts (≥15 minutes within 2 m with at least one unmasked), by peer age groups; and rich socio-demographics (education, employment, income, settlement, gender, health status, smoking), COVID-19-related behaviours and attitudes, and vaccination status. Sampling weights via raking; response rate ~49%. Contacts involving children reported by guardians. Periodization into waves and interim periods; outliers removed at 98th percentile; uncertainty estimated via bootstrap (n=1000). Statistical analysis: For determinants of contacts, negative binomial regression with interactions between age group and each socio-demographic variable X: μ_i = α + β1 age_group_i + β2 X_i + β3 age_group_i*X_i + ε_i; y_i ~ Neg-Bin(exp(μ_i), φ). Importance quantified via average marginal effects (AMEs) by age group and the maximum confidence level at which AME excludes zero across categories, ranking variables per period. For vaccination uptake (binary), logistic regression with age-by-X interactions; robustness checks include controlling for Oxford Stringency Index and individual vaccination status in contact models and assessing NPIs’ effects on vaccination. Decoupled contact matrices: Conventional age contact matrix C_ij is decoupled by respondent subgroup d (e.g., employed, education levels, income levels, settlement), yielding matrices C_{d,ij} computed as total contacts from age i in subgroup d to age j, divided by subgroup population N_{d,i}; symmetrized. Only respondent characteristics are known; alter characteristics other than age are unavailable. Epidemiological models: Extended age-structured SEIR/SEIRD models where force of infection uses subgroup-specific age contact matrices C_{d,ij}. Compartments: S, E, I, R; for real-case modelling, D is added (SEIRD). Differential equations govern transitions with parameters: transmission probability per contact β, incubation rate ε, recovery rate μ (and IFR for mortality in SEIRD). Vaccination uptake is modelled as age- and subgroup-dependent probability that moves individuals to immune/recovered states, allowing comparison of averted infections versus a no-vaccination scenario while holding contact structure fixed. Simulations use static bootstrapped contact matrices; example parameters shown in figures include μ=0.4, ε=0.25, R0=2.5; seeded with 5 infectious individuals; medians and IQRs over 1000 runs. Hungarian 4th wave application: Population stratified simultaneously by age, employment, and income to model 09/2021–01/2022; SEIRD calibrated to daily deaths via Approximate Bayesian Computation (ABC). Attack and mortality rates computed by subgroup and age to disentangle age effects from socio-economic factors.

• Determinants of contact patterns: Employment status, education level, and income are the strongest predictors (conditional on age) of contact numbers across pandemic periods. Employed and higher-educated groups consistently report more contacts; higher-SES groups adapt contact numbers more to changing epidemiological conditions and NPIs, reducing during waves and increasing during interim periods. Mid-low educated individuals had higher workplace contacts during some periods due to vocational, on-site work.
• Contact matrices by subgroup: Decoupled age-contact matrices (by education and employment) show pronounced structural differences beyond age, indicating that age-only matrices miss important heterogeneities.
• Extended SEIR outcomes: Simulations using subgroup-stratified contact matrices yield unequal attack rates. Employed individuals have the highest attack rates across age groups; mid/high-educated individuals show higher infection among ages 45–60. By settlement and income, capital residents and high-income individuals (especially 60+) exhibit higher attack rates. Conventional age-only models can overestimate overall and age-specific attack rates relative to the extended model that incorporates social dimensions.
• Vaccination uptake disparities: Higher-SES groups (especially by income) have higher vaccination uptake across all ages and periods. Modelling vaccination shows larger averted attack rates in subgroups with higher uptake, but the benefit is modulated by exposure: for 60+, not-employed individuals (with higher vaccination rates) avert fewer infections due to already low contact rates.
• Hungarian 4th wave (modelled): Employed, higher-income, and younger people experienced higher infection rates; conversely, not-employed, low-income, and older individuals had the highest mortality. After stratifying by age, not-employed consistently have lower attack rates within age groups. Among employed, infection by income is age-dependent: for 15–30, higher income shows higher infection; for 30+, lower income more often shows higher infection (except 60–70, where high income remains most infected). Mortality increases with age; lower-income groups generally suffer higher mortality, with an exception due to data sparsity for employed, high-income.
• Policy implication: Ignoring socio-demographic heterogeneity beyond age can bias predicted epidemic size and obscure unequal burdens across groups.

The findings address the central questions by demonstrating that employment, education, and income significantly shape contact patterns and vaccination uptake, which in turn drive unequal infection and mortality burdens across social strata. Incorporating these dimensions into transmission models alters predicted attack rates and reveals disparities masked by age-only stratification. Higher-SES and employed groups have more contacts and higher infections but lower mortality owing to younger age structures and higher vaccination; disadvantaged groups, despite fewer contacts and lower infection rates, suffer higher mortality, highlighting the compounded role of age, health, and access factors. These insights underscore the relevance of integrating social determinants into epidemiological models for accurate forecasting and for designing equitable, targeted NPIs and vaccination strategies.

This work extends conventional age-structured epidemic modelling by integrating socio-demographic heterogeneities in contact patterns and vaccination uptake, supported by representative survey data. It shows that employment, education, and income are key dimensions driving contact behaviour and vaccine decisions, and that extended models change epidemic forecasts relative to age-only approaches. Applying the framework to Hungary’s 4th COVID-19 wave reveals higher infection rates among employed and higher-SES groups but higher mortality among disadvantaged, older groups. The approach provides a basis for assessing socially stratified epidemic burdens and for tailoring interventions. Future research should collect and integrate contact data stratified by both ego and alter socio-demographics, improve initialization by subgroup-specific epidemic states, jointly model dynamic feedbacks between vaccination, risk perception, and contacts, and further evaluate targeted policies under realistic behavioural heterogeneity.

• Data limitations: Contact diaries capture only alter age, not other socio-demographic attributes, so decoupled matrices are along ego dimensions only. Potential recall and reporting biases. Some subgroup cells (e.g., employed high-income older) are sparse.
• Modelling assumptions: Static contact matrices within periods; homogeneous initialization across socio-demographic groups in calibrated SEIRD due to data gaps; limited exploration of dynamic interplay between vaccination behaviour and contacts; results sensitive to assumed parameters though robustness checks were conducted.
• Causal inference: Observational design prevents causal attribution; NPIs and other contextual factors may confound behaviour-outcome relationships despite controls.
• Generalizability: Findings are specific to Hungary’s context and time periods, though methods are general.