From Correlation to Mechanism: What Pandemic- Inequality Research Could Gain from Mathematical Modeling
What the NEJM Article Is and What It Found
In July 2026, the New England Journal of Medicine published a Perspective titled "The Inequality–Pandemic Cycle — Rethinking Preparedness," authored by Matthew Kavanagh, Joseph Stiglitz, Monica Geingos, Winnie Byanyima, and Michael Marmot on behalf of the Global Council on Inequality, AIDS, and Pandemics. The article's central claim is provocative: economic inequality, not technical preparedness, is the strongest predictor of how badly a country suffers during a pandemic.
The evidence behind this claim is substantial. The authors note that the United States ranked highest on the Global Health Security Index and received top marks from the World Health Organization's Joint External Evaluations, yet its Covid-19 death rate ran more than twice the global average. Meanwhile, countries with far lower preparedness rankings — Vietnam, Mauritius, Uruguay, Ethiopia, even Norway — fared considerably better. Across subnational, national, and cross-national comparisons, the authors found that inequality correlated consistently with worse Covid mortality, HIV incidence, and AIDS-related deaths, while neither global health security scores nor overall poverty levels showed the same association.
From this evidence, the authors propose a self-reinforcing cycle: inequality makes outbreaks more likely to escalate into pandemics; inequality then drives how severe and prolonged those pandemics become; and pandemics, in turn, deepen inequality, making the next outbreak harder to control. They identify four specific mechanisms driving this cycle — social determinants of exposure (crowded housing, precarious work, inability to isolate), breakdowns in visibility, governance, and trust (undercounted or stigmatized populations falling outside public health systems), unequal international financing (rich countries could spend trillions on response while poor countries could not), and unequal access to science (life-saving vaccines and treatments developed quickly but distributed slowly and unevenly). The article closes with a policy agenda: social protection during outbreaks, community-embedded governance, reformed international financial mechanisms, and open licensing of pandemic technologies.
This is careful, well-evidenced work. It establishes that these four mechanisms matter and correlates them with outcomes, but it does not offer a dynamical account of how a given configuration of inequality converts into a specific epidemic trajectory — a single wave that resolves, a series of relapses, or a stagnant plateau that persists indefinitely. That is one area where mathematical modeling has something to contribute, and — as the following sections discuss — it is an area epidemiology has already been working in for some time.
The Gap Between Correlational Findings and Dynamical, Predictive Models
Correlational research of the kind the NEJM authors present answers a specific and valuable question: which factors are statistically associated with worse outcomes? It does not, by itself, answer a related question: given a specific configuration of those factors, what trajectory is the epidemic likely to follow, and roughly where might the outcome shift from one qualitative type to another?
This is not a minor technical distinction. It is the difference between knowing that a factor matters and having some account of how it operates within a system involving feedback, delay, and threshold effects. A country's health system, immune response, and social conditions do not act as isolated, additive contributors to an outcome; they interact over time, with the effect of any one factor depending on the state of the others. Correlational studies, however rigorous, are structurally better suited to describing associations across cases than to tracing the evolving trajectory within a single case.
Dynamical models attempt to close this gap by encoding cause-and-effect relationships as a system of equations that evolves over time, producing a trajectory: a modeled curve of infections, recovery, relapse, or stabilization, and — in favorable cases — parameter ranges that separate one qualitative outcome from another. This is one gap that our own model, which treats immunity and infection as a predator-prey system governed by a cyclic three-dimensional Lotka-Volterra model, was built to explore. It is best understood as a single, illustrative example of a broader class of tools — not a uniquely capable one — and epidemiology already has a substantial, independent line of work doing comparable things, discussed below.
What a Dynamical Model Can Add to Correlational Findings
A few things become more available once a qualitative, mechanism-rich dataset like the NEJM Council's is translated into some dynamical model — ours or another.
The first is candidate regimes. Our model, for instance, suggests that depending on two composite parameters — the degree of localization of the immune or public-health response, and its efficacy — an infectious process tends toward one of a small number of qualitatively distinct outcomes: full recovery, a single wave followed by resolution, a relapsing course that eventually stabilizes into a chronic plateau, or sustained oscillations that do not fully resolve. These emerge from the governing equations rather than being assigned after the fact. Applied loosely to the NEJM's country comparisons, this offers one way to reframe qualitative outcome differences (the US doing poorly despite high preparedness, Vietnam doing well despite low preparedness) as points on a parameter diagram — though we'd emphasize this is a suggestive reframing, not a fitted or validated one at this stage.
The second is the possibility of explicit thresholds. Correlational analysis can show that more inequality associates with worse outcomes on average, but it is not designed to identify a specific point at which a system tips from one qualitative regime to another. A calibrated dynamical model, in principle, can: our framework identifies a boundary, expressed through an efficiency parameter, that separates conditions favoring full recovery from conditions permitting a slide into chronic persistence. Whether a given country's actual conditions sit near that boundary is an empirical question our current work has not yet tested against data, but it illustrates the kind of diagnostic a fitted model could eventually offer alongside, not instead of, inequality rankings.
The third is a quantitative prediction worth testing. Because our epidemic-level model reduces to a well-defined oscillatory system, it yields a specific candidate prediction for the period of epidemic waves, expressed as the geometric mean of the infectious duration and the immunity-decay time. This is the kind of falsifiable, quantitative claim that a correlational study does not produce on its own; once calibrated, it could be checked against the timing of successive Covid-19 waves in high- versus low-inequality settings, alongside comparable predictions from other dynamical frameworks used in this space.
A Concrete Template: Fitting the NEJM's Own Examples
The NEJM article itself supplies several specific, well-documented episodes that could serve as candidate data points for calibrating a dynamical model rather than remaining as illustrative anecdotes.
Singapore's outbreak among migrant workers in crowded dormitories, missed for a period by an otherwise robust national testing system, is a case where a subpopulation's infection state became functionally invisible to the wider public-health response. In dynamical terms, this describes a breakdown in the feedback coupling between a local infected group and the population-level response — a parameter that, in a model built around population feedback (rather than fixed, uniform response rates), would show up as an effectively decoupled compartment. Fitting the actual case-count trajectory from that outbreak against such a model, with and without that feedback term active, could help test whether removing the coupling reproduces the delayed detection and subsequent surge that was actually observed.
The months-long delay before mpox cases diagnosed locally in the Democratic Republic of Congo triggered an international response is a comparable governance-scale example. This is not simply a story about institutional failure; it is a specific, dateable interval during which a feedback loop that a model would normally assume to be active was absent. That interval — the length of the delay, and the trajectory of cases before and after international response finally engaged — is a measurable quantity a fitted model could be checked against, testing whether the observed acceleration in cases during the delay period is broadly consistent with what the model predicts when the relevant coupling parameter is set near zero.
The gap in vaccine rollout timing — high-income countries reaching priority populations within six months, while many low-income countries had not yet begun vaccinating by that point — offers a fairly clean quantitative input. That six-month interval is a plausible proxy for the localization and efficacy parameters that, in a predator-prey-style model, influence whether a population moves toward rapid suppression of an outbreak or toward a prolonged relapsing or plateau course. Comparing actual case trajectories in early- versus late-vaccinated countries against the regimes such a model predicts for correspondingly high versus low parameter values would be one way to convert the NEJM's observed six-month gap into a testable dynamical hypothesis, alongside other modeling approaches suited to the same question.
Epidemiology Already Has a Family of Comparable Models
It's worth being clear that the predator-prey framework above is only one example from a much larger, already-active toolkit, and epidemiology has not been waiting for dynamical approaches to inequality — it has been building them for over a decade, with a marked acceleration since Covid-19.
Three established lines of work are directly relevant. First, SEIR-type compartmental models have been extended to stratify contact rates, susceptibility, and other transmission determinants by socioeconomic status, the same way models are routinely stratified by age; this lets researchers mechanistically embed exposure inequality (crowding, precarious work, inability to isolate) directly into the model structure rather than treating it as an external correlate. Second, agent-based models built on synthetic populations — households, workplaces, schools — have been used to reproduce observed inequities in infection, hospitalization, and death by social stratum, and to run counterfactual scenarios; a widely cited example modeled Santiago, Chile, and found that more equitable social-distancing policies would have prevented more than 80% of the deaths actually recorded in that period. Third, mobility-data-augmented compartmental models incorporate real movement patterns that differ sharply by income group, capturing how unequal ability to reduce mobility during lockdowns prolongs transmission in lower-income areas. Separately, more theoretical work — for instance Esseau-Thomas and colleagues' model of the feedback between lockdown policy and income inequality — has formalized the inequality-worsens-pandemics-worsens-inequality cycle mathematically in its own right, arriving at a threshold result for lockdown policy that is conceptually close to the kind of tipping-point finding a Lotka-Volterra approach also aims at.
The practical implication is that a research program built on the NEJM Council's four mechanisms does not need to adopt any single modeling framework, including ours, to gain dynamical traction. SES-stratified SEIR models, agent-based synthetic-population models, mobility-augmented compartmental models, and predator-prey-style immune-infection models are best read as a family of complementary tools, each suited to slightly different questions and data availability — and a research group could reasonably choose among them, or combine several, depending on which of the Council's four mechanisms it is trying to formalize.
The Broader, Model-Agnostic Argument
None of this depends on any one model, including ours, being the correct or preferred tool for the task. The more general point is structural: wherever researchers have moved beyond correlation and identified specific, well-defined causal mechanisms — as the NEJM Council has done with its four channels of social determinants, governance and trust, financing, and access to science — the underlying dataset is, in principle, compatible with dynamical modeling, and epidemiology already has multiple mature frameworks capable of doing that work.
This creates an opportunity that extends beyond pandemic preparedness specifically. Any body of research that has isolated distinct causal channels, rather than simply reporting an association between a broad factor and an outcome, has effectively identified candidate parameters for some dynamical system — and in epidemiology, several such systems already exist and are actively used. One useful next step, among several viable approaches, is building or adapting a model in which those parameters are made explicit and testing whether its predicted regimes and thresholds are consistent with observed cases; this is a step that current inequality-pandemic research has generally not yet taken, but it is one for which the field already has tools on hand, not one that requires inventing a new modeling paradigm. The NEJM Council's inequality-pandemic work is a clear and current example of research that has done the qualitative groundwork and could, using any of several existing dynamical frameworks, be extended toward more predictive, falsifiable claims.
References
Kavanagh, M. M., Stiglitz, J. E., Geingos, M., Byanyima, W., & Marmot, M. G., for the Global Council on Inequality, AIDS, and Pandemics. (2026). The Inequality–Pandemic Cycle — Rethinking Preparedness. New England Journal of Medicine. https://doi.org/10.1056/NEJMp2607839
Global Council on Inequality, AIDS and Pandemics. (2025). Breaking the inequality-pandemic cycle: Building true health security in a global age. United Nations/UNAIDS. https://www.unaids.org/sites/default/files/2025-11/2025_global-council-inequality-report_en.pdf
Haider, N., Yavlinsky, A., Chang, Y.-M., et al. (2020). The Global Health Security Index and Joint External Evaluation score for health preparedness are not correlated with countries' COVID-19 detection response time and mortality outcome. Epidemiology and Infection, 148, e210.
Dimaschko, Ju. E., Shlyakhover, V. E., & Iabluchanskyi, M. I. (2023). Immunity and Infection as Predator and Prey: Common Concept of Infectious Disease and Epidemic. ResearchGate. https://doi.org/10.13140/RG.2.2.17366.63040
Esseau-Thomas, C., Galarraga, O., & Khalifa, S. (2022). Epidemics, pandemics and income inequality. Health Economics Review, 12, 7. https://doi.org/10.1186/s13561-022-00355-1
Richard, D. M., & Lipsitch, M. (2024). What's next: Using infectious disease mathematical modelling to address health disparities. International Journal of Epidemiology, 53(1), dyad180. https://doi.org/10.1093/ije/dyad180
Tizzoni, M., Nsoesie, E. O., Gauvin, L., Karsai, M., Perra, N., & Bansal, S. (2022). Addressing the socioeconomic divide in computational modeling for infectious diseases. Nature Communications, 13, 2897. https://doi.org/10.1038/s41467-022-30688-8
Gozzi, N., Perrotta, D., Paolotti, D., & Perra, N. (2021). Estimating the effect of social inequalities on the mitigation of COVID-19 across communities in Santiago de Chile. Nature Communications, 12, 2429. https://doi.org/10.1038/s41467-021-22601-6
Nande, A., Sheen, J., Walters, E. L., et al. (2021). The effect of eviction moratoria on the transmission of SARS-CoV-2. Nature Communications, 12, 2274. https://doi.org/10.1038/s41467-021-22521-5
Chang, S., Pierson, E., Koh, P. W., et al. (2021). Mobility network models of COVID-19 explain inequities and inform reopening. Nature, 589, 82–87. https://doi.org/10.1038/s41586-020-2923-3
Mena, G. E., Martinez, P. P., Mahmud, A. S., Marquet, P. A., Buckee, C. O., & Santillana, M. (2021). Socioeconomic status determines COVID-19 incidence and related mortality in Santiago, Chile. Science, 372, eabg5298.
Zelner, J., Trangucci, R., Naraharisetti, R., et al. (2022). There are no equal opportunity infectors: Epidemiological modelers must rethink our approach to inequality in infection risk. PLOS Computational Biology, 18, e1009795.
Furceri, D., Loungani, P., Ostry, J. D., & Pizzuto, P. (2021). Will COVID-19 have long-lasting effects on inequality? Evidence from past pandemics. IMF Working Paper WP/21/127.
Comments
Post a Comment