Dense Cities May Shield Their Poorest Residents From Diabetes, Israeli Study Finds

For decades, public health policy has rested on a deceptively simple assumption: the poorer a community, the sicker it is, and the relationship runs in a straight line. Health systems around the world, including Israel’s, allocate preventive resources on exactly that logic, directing screening programs, health promotion campaigns, and chronic disease interventions toward the municipalities that sit lowest on official socioeconomic rankings. But a new analysis of 127 Israeli municipalities suggests that this monotonic view of health inequality misses something fundamental. According to a study published in Frontiers in Epidemiology, the socioeconomic gradient in diabetes prevalence is not a fixed feature of disadvantage—it bends, and in some cases flattens, depending on how densely populated a municipality happens to be.

The research team, led by Yuval Arbel of Western Galilee College together with Israel Rachevski, Amichai Kerner of Bar Ilan University, and Miryam Kerner of the Technion and Kaplan Medical Center, assembled municipality-level administrative data covering roughly 96 percent of Israel’s 132 municipalities. Their outcome measure was age-standardized diabetes prevalence—the number of diagnosed diabetes cases per 1,000 inhabitants, adjusted to remove distortions from differing age structures—based on 2017 to 2019 data. Socioeconomic status came from the official municipal ranking published by the Israeli Central Bureau of Statistics, a composite index reflecting income, education, employment, and living conditions on a scale from 1, the most disadvantaged, to 10. Population density, measured in inhabitants per square kilometer, served as the proxy for urban context.

The descriptive picture alone is striking. Average diabetes prevalence across the sample was 68.71 cases per 1,000 inhabitants, but the range ran from 38.27 in the healthiest municipality to 142.84 in the most affected—more than a threefold difference. The bivariate statistics told a familiar story: diabetes prevalence correlated strongly and negatively with socioeconomic rank, with a Pearson coefficient of −0.633, and negatively with logged population density at −0.273. But the researchers were not content to stop at correlation. Their central question was whether the shape of the socioeconomic gradient itself changes as density changes, and answering that required a more flexible statistical architecture.

Their approach was an ordinary least squares regression with a quadratic specification: diabetes prevalence was modeled as a function of socioeconomic rank, the square of that rank, the natural logarithm of population density centered at the sample median of about 2,713 inhabitants per square kilometer, and—crucially—interaction terms between density and both the linear and squared SES components. Those interactions allow the slope and the curvature of the SES–diabetes relationship to vary with density, rather than forcing a single gradient onto all 127 municipalities. Heteroskedasticity-robust HC3 standard errors guarded against undue influence from outlying municipalities in the relatively small cross-section. The model explained about 44 percent of the between-municipality variation in diabetes prevalence.

The headline result concerns what density does at the bottom of the socioeconomic distribution. At SES rank 1, the marginal effect of logged density was −14.73, statistically significant at the one percent level, meaning that an exact 10 percent increase in population density was associated with roughly 1.40 fewer diabetes cases per 1,000 inhabitants. At rank 2 the effect was −9.66, equivalent to about 0.92 fewer cases per 1,000 for the same density increase, and at rank 3 it remained significant at −5.64, or roughly 0.54 fewer cases. From rank 4 onward, the density association dissolved into statistical noise, hovering near zero across the middle and upper reaches of the SES scale. In other words, the apparent protective association of urban density is concentrated precisely where diabetes burden is highest.

The complementary analysis examined how the SES gradient itself shifts across density contexts, evaluating conditional coefficients at the 25th percentile (1,927 inhabitants per square kilometer), the median (2,713), and the 75th percentile (4,531) of the density distribution. At the two lower density levels, neither the linear nor the quadratic SES component reached statistical significance. But at the 75th-percentile density, the conditional quadratic coefficient turned substantially more negative, at −0.628, and became significant at the five percent level. The researchers calculated that this detectable curvature emerges at densities above approximately 4,227 inhabitants per square kilometer—though they are careful to frame this as an approximate density region, not a discrete threshold. Predicted curves plotted at the three density levels fan out dramatically at the low end of the SES distribution and converge in the middle and upper portions, visual confirmation that density matters most for the most disadvantaged places.

One subtlety deserves emphasis, because it guards against an overly enthusiastic reading. At high density, the estimated turning point of the quadratic relationship sits at approximately SES rank 1.05—almost exactly the bottom boundary of the observed scale. That means the data do not support characterizing the high-density relationship as a broad inverted-U in which intermediate socioeconomic levels carry the greatest risk. Instead, predicted diabetes prevalence peaks near the very bottom of the SES range and declines across virtually the entire remaining distribution. The quadratic evidence, the authors note, should be interpreted cautiously; the parsimonious quadratic form imposed on an ordinal nine-level ranking cannot uniquely identify the true functional form of the SES–diabetes relationship.

What might explain the pattern? The authors are explicit that their observational design cannot establish causation, and that population density should not be read as the causal mechanism itself. Rather, density likely serves as a proxy for a bundle of municipal characteristics: proximity to services, transportation accessibility, healthcare infrastructure, demographic composition, and the walkability and service provision that urban economists associate with agglomeration. For socioeconomically disadvantaged residents, who may lack private vehicles and face constraints on healthcare access, the resources concentrated in denser municipalities could matter disproportionately. The interpretation aligns with a substantial place-effects literature in health geography, as well as emerging evidence that the built environment and socioeconomic context interact in non-linear ways to shape obesity and diabetes risk. At the same time, density brings its own exposures, and the same urban form that puts clinics within walking distance can also concentrate pollution and alter consumption environments.

The policy implications cut against entrenched practice. If municipalities with identical socioeconomic rankings can carry markedly different disease burdens depending on their density context, then allocating preventive resources by rank alone risks misallocation. Two disadvantaged municipalities—one sprawling and sparse, one compact and dense—may not need the same intervention intensity, even though their composite SES scores are indistinguishable. The authors recommend complementing socioeconomic indices with direct, outcome-based indicators such as observed disease prevalence, identifying municipalities whose burden is unexpectedly high relative to their rank and local characteristics, and incorporating density and other contextual variables into the design of screening and lifestyle interventions. For chronic disease prevention, where risk factors are tightly entangled with environmental and behavioral conditions, such flexibility may be especially consequential.

The study carries limitations that shape how far its conclusions travel. Five municipalities with missing data were excluded, leaving open the possibility of some selection; the analysis is cross-sectional and cannot disentangle behavioral, environmental, and institutional confounders; and because observations are municipalities rather than individuals, the findings describe between-place differences, not individual-level associations between a person’s socioeconomic status and their diabetes risk. Residual confounding by unmeasured municipal characteristics cannot be excluded even with age standardization. Still, the core message is difficult to dismiss: socioeconomic gradients in health may not operate uniformly across space, and the interactive structure the Israeli data reveal offers a template for testing whether the same density-dependent heterogeneity appears in other countries and for other outcomes. If it does, the humble practice of ranking places by wealth and intervening accordingly may need a considerably more textured replacement.

Subject of Research: The relationship between municipal socioeconomic status, population density, and diabetes prevalence in Israel

Article Title: Socioeconomic gradients in diabetes prevalence across Israeli municipalities: the role of population density

Article References: Arbel, Y., Rachevski, I., Kerner, A., & Kerner, M. (2026). Socioeconomic gradients in diabetes prevalence across Israeli municipalities: the role of population density. Frontiers in Epidemiology, 6, Article 1848489. https://doi.org/10.3389/fepid.2026.1848489

Image Credits: AI Generated

DOI: 10.3389/fepid.2026.1848489

Keywords: diabetes prevalence, socioeconomic status, population density, health inequalities, Israel, municipalities, epidemiology, public health policy, urban health, chronic disease prevention, place effects, quadratic regression