Ethnic inequalities in respiratory virus epidemics in England: a mathematical modelling study

This mathematical modelling study demonstrates that ethnic inequalities in respiratory virus transmission in England are driven by distinct demographic characteristics and social mixing patterns, which result in varying attack rates across ethnic groups that are further modulated by local population structures and pathogen infectiousness.

The Gist

Imagine the spread of a respiratory virus (like the flu or a cold) as a giant game of "musical chairs" played across England. But instead of chairs, the players are passing around "infection tokens." The goal of this study was to figure out why some groups of people end up with way more tokens than others, and whether the rules of the game change depending on which city you're playing in.

Here is the story of the study, broken down into simple concepts:

1. The Players and the "Contact Count"

The researchers looked at a massive survey called Reconnect, which asked over 12,000 people in England: "How many people did you talk to or see yesterday?"

They found that the number of people you see (your "contact count") isn't just random. It's like a social battery:

• Black and Mixed ethnic groups tended to have a "fuller battery" on average, meaning they interacted with more people daily.
• Asian ethnic groups tended to have a "smaller battery," interacting with fewer people.
• White ethnic groups fell somewhere in the middle, but generally had fewer contacts than Black and Mixed groups.

The Twist: Even when the researchers adjusted for things like age, job, income, and how many people live in your house, these differences remained. It wasn't just that Black people were younger or lived in bigger houses; the culture of socializing itself seemed to differ by group.

2. The "Virus Storm" Simulation

Since you can't ethically infect thousands of people to see what happens, the researchers built a digital twin of England—a computer simulation. They created a virtual population with the exact same mix of ages, ethnicities, and social habits as the real world.

Then, they let a "digital virus" loose. They asked: If a virus starts spreading, who gets hit the hardest?

The Results:

• The White group (who make up the majority of the population) had the lowest rate of infection.
• The Black and Mixed groups had the highest rates. In some scenarios, they were getting infected at nearly double the rate of the White group.
• The Asian group was in the middle, slightly higher than the White group but lower than the others.

Why? It's like a forest fire. If you have a group of trees (people) that are packed closer together and have more branches touching (more social contacts), the fire spreads faster through that specific patch of forest, even if the wind (the virus) is the same.

3. The "City Map" Effect

The researchers didn't just look at England as one big blob. They zoomed in on specific cities like Birmingham, Liverpool, London, and York.

Think of each city as a different ecosystem:

• Birmingham is like a dense, bustling marketplace. The mix of ages and ethnicities there meant that the "infection tokens" flew around very quickly. The gap between the White group and the Black/Mixed groups was huge here.
• Liverpool and York are more like quiet villages with a different mix of people. Here, the gap between groups was much smaller.

The study showed that a "one-size-fits-all" public health plan (like a national lockdown or a mass vaccination campaign) might work well in Liverpool but fail to protect the most vulnerable people in Birmingham. Local context is king.

4. The "Recipe" for Inequality

The study identified three main ingredients that mix together to create these inequalities:

1. Demographics: Who lives where? (e.g., younger populations tend to have more friends and parties).
2. Mixing Patterns: Who talks to whom? (People tend to talk to others like themselves more often).
3. The "Extra" Social Factor: Even after accounting for age and jobs, some ethnic groups simply have more social interactions per person than others.

The Big Takeaway

Imagine you are trying to stop a leak in a boat. If you only look at the boat from a distance (national data), you might think the hole is in the middle. But if you look closer (local data), you see the hole is actually in the corner where the water is rushing in fastest.

The lesson: To stop a virus from spreading unfairly, we can't just use a generic map. We need to understand the specific "social neighborhoods" of different cities. Public health strategies need to be tailored to the local mix of people and their social habits to ensure everyone gets the protection they need, not just the average person.

In short: Where you live, who you are, and who you hang out with all combine to determine how likely you are to catch a virus. Ignoring these details leaves some communities behind.

Technical Summary

1. Problem Statement

Respiratory infections in England exhibit significant disparities across ethnic groups, with individuals from Black, Mixed, and South Asian backgrounds historically facing higher risks of infection, hospitalization, and severe outcomes compared to White individuals. While demographic factors (age, household size, income) and structural inequalities are known contributors, the specific role of social mixing patterns and contact heterogeneity in driving these transmission dynamics remains underexplored.

The study aims to:

• Quantify how differences in social contact patterns and demographic characteristics between ethnic groups contribute to transmission inequalities.
• Determine whether these inequalities persist after adjusting for demographic variables.
• Model how local demographic structures (e.g., specific cities) alter the expected burden of disease and the relative risk between ethnic groups.

2. Methodology

Data Sources

• Reconnect Survey (2024–2025): A nationally representative social contact survey in the UK with 13,238 participants. The study utilized data from 12,484 participants after exclusions. It collected self-reported ethnicity (2021 Census categories: White, Black, Asian, Mixed, Other), demographics (age, gender, household size, income, employment), and detailed contact diaries (number of contacts, age, sex, ethnicity of contacts).
• Census 2021 & Family Resources Survey: Used to obtain population distributions for age, household size, employment status, and income at national and local levels (England, London, and six major cities: Birmingham, Leicester, Liverpool, Manchester, York).

Statistical Analysis

• Bayesian Negative Binomial Regression: The authors modeled the number of contacts as a function of ethnicity and demographic covariates (age, employment, income, household size, gender, urban/rural status, weekday/weekend).
  • This approach allowed for the estimation of both the mean number of contacts and the dispersion (variance) of the contact distribution.
  • Rate Ratios (RR) were calculated to compare contact rates of non-White groups against the White reference group, adjusted for demographics.

Mathematical Modeling

• Stratified Stochastic SEIR Model: A compartmental model (Susceptible-Exposed-Infected-Recovered) was implemented in R (odin2, dust2).
  • Stratification: The population was stratified by Age, Ethnicity, and Contact Level (Low, Medium, High).
  • Contact Group Classification: Using the regression parameters, synthetic populations were generated to fit a three-component Poisson mixture model, classifying individuals into low, medium, and high-contact groups based on their age and ethnicity strata.
  • Mixing Matrix: A mixing matrix was constructed combining age-specific and ethnicity-specific mixing patterns from the Reconnect survey, assuming independence between age and ethnicity mixing structures.
• Simulation Scenarios:
  • Baseline: Full heterogeneity (demographics, contact distribution, and mixing patterns).
  • Scenario 1 (Demographics Only): Ethnicity-specific contact parameters set to 1 (removing contact distribution differences).
  • Scenario 2 (National Demographics): Used national-level demographic distributions for all ethnic groups (removing local demographic structure differences).
  • Scenario 3 (Homogeneous Mixing): Replaced the ethnicity-specific mixing matrix with a constant matrix (removing assortative mixing).
• Local Simulations: The model was run using local demographic distributions for six major English cities to assess spatial variation in $R_0$ and attack rates.
• Parameters: Simulations covered a range of basic reproduction numbers ($R_0$ from 1.4 to 8.0) to represent various respiratory pathogens.

3. Key Results

Contact Patterns and Demographics

• Demographic Differences: White individuals in England are, on average, older, live in smaller households, and have higher retirement rates compared to Black, Mixed, and Asian groups.
• Contact Rates (Adjusted): Even after adjusting for demographics, significant disparities in contact rates remained:
  • Black Ethnicity: Higher contact rates (Rate Ratio [RR]: 1.18; 95% CI: 1.11–1.26).
  • Mixed Ethnicity: Significantly higher contact rates (RR: 1.31; 95% CI: 1.14–1.52) and a more dispersed distribution (higher proportion of extreme contact numbers).
  • Asian Ethnicity: Lower contact rates (RR: 0.85; 95% CI: 0.79–0.91).
  • Dispersion: Asian and Black groups showed less dispersed contact distributions than White groups, while Mixed groups showed higher dispersion.

Epidemic Dynamics and Attack Rates

• National Level: In simulated outbreaks, the White ethnic group consistently had the lowest attack rates.
  • Mixed and Black groups had the highest relative attack rates (approx. 1.4–1.5 times higher than White groups at $R_0=2$).
  • Asian groups had slightly elevated attack rates (approx. 1.05–1.08 times higher) compared to White groups.
• Drivers of Inequality:
  • When demographic structures were harmonized (Scenario 2), heterogeneity in attack rates decreased significantly, particularly for Mixed and Black groups.
  • When mixing patterns were homogenized (Scenario 3), heterogeneity dropped further, indicating that assortative mixing (preference for mixing within one's own ethnic group) is a critical driver of inequality.
  • Residual differences in contact distribution (regression parameters) still contributed to inequality, but to a lesser extent than demographics and mixing.

Local (City-Level) Variation

• Birmingham: Showed the highest inequality. The relative attack rate for Black and Mixed groups was approximately double that of White groups (RR $\approx$ 2.0) at lower $R_0$ values. This is attributed to a large age gap between ethnic groups and larger household sizes in non-White populations in this city.
• Liverpool and York: Cities with higher proportions of White populations showed lower relative inequalities (RR < 1.6).
• $R_0$ Variation: Local demographic structures impacted the effective transmissibility. For a fixed transmission rate ($\beta$), $R_0$ was up to 8.7% higher in Manchester compared to York, leading to absolute incidence differences of 2–3%.

4. Key Contributions

1. Quantification of Contact Heterogeneity: Provided the first large-scale, adjusted quantification of how ethnic groups differ in social contact patterns in England, revealing that Black and Mixed groups have systematically higher contact rates independent of age or income.
2. Decomposition of Drivers: Demonstrated that ethnic inequalities in infection risk are driven by a triad of factors: demographic structure (age/household size), contact distribution (mean and dispersion), and mixing patterns (assortativity).
3. Context-Dependence of Inequality: Showed that the magnitude of ethnic health inequalities is not static; it varies significantly by location (city) and pathogen transmissibility ($R_0$).
4. Methodological Framework: Integrated a large-scale social contact survey with a stratified stochastic SEIR model to project how micro-level contact behaviors scale up to macro-level epidemic disparities.

5. Significance and Implications

• Public Health Policy: National-level estimates of infection risk may mask significant regional variations. Interventions based solely on national averages may be ineffective or inequitable.
• Tailored Interventions: Mitigation strategies (e.g., vaccination campaigns, non-pharmaceutical interventions) must be tailored to local demographic contexts. For example, cities like Birmingham require different strategies than York due to distinct ethnic contact structures.
• Equity: The findings suggest that "one-size-fits-all" approaches could inadvertently widen health inequalities. Understanding the specific drivers (e.g., household size vs. mixing patterns) allows for more targeted interventions.
• Future Research: The study highlights the need to integrate data on vaccine coverage, socioeconomic deprivation, and severity outcomes to fully understand the progression from infection risk to clinical burden.

In conclusion, the study establishes that ethnic inequalities in respiratory epidemics are structural and driven by complex interactions between demographics, social behavior, and local population composition. Addressing these inequalities requires granular, location-specific data and modeling.