Structural causal influence (SCI) captures the forces of social inequality in models of disease dynamics

Sudam Surasinghe, Swathi Nachiar Manivannan, Samuel V. Scarpino, Lorin Crawford, C. Brandon OgbunugaforarXiv 2024 · DOI · arXiv

TL;DR

Researchers have introduced Structural Causal Influence (SCI), a new metric that quantifies how social inequalities—like unequal access to healthcare or varying exposure risks—drive the spread of infectious diseases. By applying causal inference to epidemiological models, the study demonstrates that even minimal interaction between communities can cause a "less vulnerable" group to face a sustained epidemic it would have otherwise avoided.

What Problem Does This Paper Address?

Traditional epidemiological models often treat populations as homogeneous "soups" or use simple stratifications that fail to capture the formal impact of social forces. While we know qualitatively that poverty, systemic racism, and lack of healthcare access drive disease, we lack a rigorous way to measure how much these structural inequalities influence the overall trajectory of an outbreak.

The authors argue that existing metrics, like the basic reproduction number ($R_0$), provide a population-wide average that can mask the reality of what’s happening in specific sub-communities. This "modeling gap" makes it difficult for public health officials to design interventions that specifically target the root causes of transmission variability.

Methods

The paper introduces the Structural Causal Influence (SCI) metric, grounded in Judea Pearl’s framework of causal inference (specifically "do-calculus").

The core idea is the "perfect intervention." To calculate the SCI for a specific sub-community ($i$), the researchers compare two scenarios:

  1. The Full Model: The disease dynamics of the entire population, including all social inequalities ($R_0^{SD}$).
  2. The Intervened Model: A hypothetical scenario where community $i$ is "removed" or perfectly protected from infection ($R_0^{do(\Gamma_i)}$).

The SCI ($C_i$) is the normalized difference between these two values: $$C_i = \frac{R_0^{SD} - R_0^{do(\Gamma_i)}}{R_0^{SD}}$$

The authors tested this metric across three different modeling environments:

  • A standard SIR (Susceptible-Infectious-Recovered) model modified for social determinants.
  • A complex Ordinary Differential Equation (ODE) model of Hepatitis C (HCV) among people who inject drugs (PWID).
  • A spatial Agent-Based Model (ABM) where individuals move and interact in a simulated 2D environment.

Datasets

The paper primarily uses simulated data and parameters derived from existing literature rather than a single new empirical dataset.

  • HCV Parameters: Based on historical data of the PWID population in New York City (approx. 170,000 individuals).
  • ABM Simulations: The researchers generated synthetic outbreak data using 1,000 simulation runs to establish mean values for growth rates and reproduction numbers.

Key Insights and Ideas

  • The "Vulnerability Trap": One of the most striking findings is that a "less vulnerable" community might have an isolated $R_0$ below 1 (meaning the disease would naturally die out). However, even 1% mixing with a more vulnerable community can push that $R_0$ above 1, leading to a sustained epidemic in both groups.
  • Negative SCI: The metric can be negative. This happens when a specific sub-community has such low transmission that it effectively "dilutes" the overall population's infection rate. A negative SCI for Community A indicates that the epidemic in the total population is actually less severe than it is in the more vulnerable Community B.
  • Inequality as a Systemic Risk: The SCI formally shows that inequality isn't just a problem for the marginalized; it changes the fundamental math of the disease for everyone. High SCI values in vulnerable groups indicate they are "driving" the epidemic for the entire population.

Key Results

  • HCV Case Study: In the Hepatitis C model, Community 2 (the more vulnerable group with lower needle clearance and treatment rates) consistently showed a much higher SCI ($C_2$ ranging from 0.53 to 0.91) compared to Community 1 ($C_1$ ranging from 0 to 0.36).
  • Mixing Sensitivity: In the ABM, the researchers found that as mixing ($m$) increased, the SCI of the less vulnerable community ($C_1$) moved from negative to positive. This transition marks the point where the "safe" community stops suppressing the total population's growth rate and starts contributing to it.
  • Intervention Strategy: The authors show that if a community has a high SCI, targeting interventions there is mathematically the most efficient way to reduce the $R_0$ of the entire population.

Limitations and Future Work

  • Calculation Complexity: Calculating SCI requires the ability to estimate $R_0$ for both full and intervened models. In highly complex, real-world scenarios with many overlapping sub-communities, this could become computationally expensive.
  • Perfect Interventions: The "do-operator" assumes a perfect intervention (zero transmission), which is rarely achievable in public health. Future work could explore "soft interventions" that reflect more realistic policy outcomes.
  • Scaling: The current study focuses on two-community models. The authors suggest that future research should apply SCI to models with multiple, intersecting social identities and more granular demographic data.

Summary updated

Abstract

Mathematical modeling has played a central role in understanding how infectious disease transmission manifests in populations. These models have demonstrated the importance of key community-level factors in structuring epidemic risk, and are now routinely used in public health for decision support. One barrier to their broader utility is that the existing canon does not often accommodate social inequalities as distinct formal drivers of variability in transmission dynamics. Given decades of evidence supporting the organizational effects of inequalities in structuring society more generally, and infectious disease risk more specifically, addressing this modeling gap is of critical importance. In this study, we build on previous efforts to integrate social forces into computational epidemiology by introducing a metric, the structural causal influence (SCI). The SCI uses causal analysis to provide a measure of the relative vulnerability of sub-communities within a susceptible population, shaped by differences in characteristics such as access to therapy, exposure to disease, and other determinants driven by social forces. We develop our metric in a simple case and apply it to a context of public health importance: Hepatitis C virus in a population of persons who inject drugs. In addition, we demonstrate the flexibility of the SCI using an agent-based model of an infectious disease. Our use of the SCI reveals that, under specific parameters in a multi-community model, the "less vulnerable" community may achieve a basic reproduction number below one, ensuring disease extinction. However, even minimal transmission between communities can increase this number, leading to sustained epidemics within both communities.
BibTeX
@misc{surasinghe2024structural,
  author = {Sudam Surasinghe and Swathi Nachiar Manivannan and Samuel V. Scarpino and Lorin Crawford and C. Brandon Ogbunugafor},
  title = {{Structural causal influence (SCI) captures the forces of social inequality in models of disease dynamics}},
  year = {2024},
  eprint = {2409.09096},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2409.09096},
}
Figure