Multilevel random-effects (RE) models are standard in social and geographical epidemiology. They rest on the assumption of exchangeability, which means that, before seeing the data, we have no reason to distinguish one group from another: group means become draws from one common distribution. RE models were developed for repeated measurements within individuals where a shared physiology underwrites the assumption, and partial pooling improves prediction. A silent transfer later applied the same machinery to schools, hospitals, neighbourhoods and, recently, to intersectional strata. There the transfer breaks down. Social positions are produced by directional forces such as racism, not drawn from a common distribution. Within a group, the variation between individuals is not measurement error but meaningful individual heterogeneity shaped by unmeasured life courses. We review three readings of exchangeability, from the strictest to the most permissive, and they agree: social and intersectional categories are not exchangeable. Assuming it would require the absence of the very inequalities under study. In many intersectional analyses, shrinkage erases the small and deviating strata, where inequality lives. Modelling injustice as dispersion around a mean domesticates rather than denounces it. For describing realised health inequalities we recommend Simple-Means MAIHDA (S-MAIHDA). Each stratum is estimated from its own individuals, and uncertainty is reported rather than smoothed. RE models keep their legitimate uses, such as prediction, cluster-randomised trials and the propensity models of causal S-MAIHDA. Exchangeability is not a property of the software. It is a claim about the world, to be justified in every application.
In their recent article, Menon et al. study household out-of-pocket health expenditure (OOPE) in Kerala from 2017 to 2020. Using data from the Consumer Pyramids Household Survey, they show 32% reduction in OOPE during this period, largely explained by decline in medication spending. They attribute this reduction to Karunya Arogya Suraksha Padhathi (KASP), a publicly financed health insurance scheme in Kerala. In this commentary, we critically question this attribution. We show that the mechanisms of OOPE reduction postulated by Menon et al. are not supported by the findings of the article. In fact, previous rounds of the National Sample Survey show that any hospitalisation-focused insurance scheme such as KASP cannot be expected to produce large declines in OOPE. In our view, strengthening of government primary healthcare in Kerala since 2017 is a much more plausible explanation for the findings. For financial protection of patients in India, government primary healthcare services need to be improved at scale, with a focus on availability of free medications.
Suicide remains a leading cause of premature mortality in high-income countries, yet prevention efforts often lag emerging risks. We conducted a scoping review of regional modelling studies from high-income countries published over 2015 to 2025 that used longitudinal administrative datasets to estimate regional suicide and self-harm risk, excluding individual-level models. Seventy-one studies met the inclusion criteria, concentrated in the United States (25%), Australia (14%), South Korea (13%) and the United Kingdom (11%). Most studies relied on death records to construct outcomes and applied count-based or spatial models. Among the most frequently examined region-level determinants, higher urbanicity/population density was predominantly associated with lower risk, higher unemployment and material deprivation with higher risk, and higher per-capita income with lower risk. Social determinants, including social fragmentation and divorce rates, also showed relatively consistent associations with higher risk. Descriptive country-stratified findings showed that higher material deprivation was most commonly associated with higher risk in the four countries contributing the most studies. Urbanicity likely functions as a broad contextual indicator capturing multiple characteristics of place rather than a single protective mechanism. Relatively few studies combined panel models, lagged predictors, and machine-learning methods to make fuller use of the longitudinal and high-dimensional information available in administrative data. Common study limitations included short or irregular time series, single-year measures, outcome misclassification and aggregation issues. Expanded access to linked administrative data and analytical methods suited to longitudinal high-dimensional data may strengthen regional risk modelling and, when integrated with local decision-making and service responses, support more timely, place-based prevention.
Background The food environment may influence neurodegeneration and mortality. This study examined associations of food deserts (FD) with brain health and survival using UK Biobank data. Methods We analyzed 262,847 participants aged ≥50 years at baseline (2006–2010) with linked geospatial (2018–2020), neuroimaging (2014–2023; 2019–2023), dietary, biological, and health data. FD exposure was aggregated at the local authority (LA) level. Outcomes included incident dementia, Parkinson's disease (PD), and all-cause mortality (ACM) from January 1, 2020, to January 1, 2024. Neuroimaging sub-analyses (n = 2429) assessed brain volumes and white matter microstructure (WMM) using diffusion MRI metrics (FAmean, MDmean, ICVFmean, ISOVFmean, ODmean). Cox, multistate survival, and mixed-effects models were used. Spatial clustering examined co-localization between FD and outcomes. Area-level FD was examined as the primary exposure, with FI evaluated mainly as a secondary effect modifier. Results Women had lower FD and FI scores and better cardiovascular and social health. FD (per SD increase) was associated with higher ACM risk (HR = 1.06; 95% CI: 1.02–1.09) after adjustment for demographics and deprivation, attenuating after full adjustment. FD also predicted faster WMM decline among women (ODmean β = +0.1124; p = 0.028), though neuroimaging associations did not survive FDR correction. Spatial analyses revealed threefold higher odds of concordance between FD and ACM, and between FD and FAmean decline. FI did not modify FD–ACM associations. Conclusions Living in food deserts was linked to higher mortality and faster white matter decline, particularly among women. Findings highlight the role of area-level food environments in brain health and aging.
Background. The aftermath of firearm violence is shaped by the same structural conditions that determine exposure, yet their role in producing differential post-exposure experiences remains underspecified. This study identifies the mechanisms through which these conditions generate inequitable aftermath experiences across four exposure typologies. Methods. Secondary qualitative analysis of 33 in-depth interviews with adults who had experienced firearm violence (physical injury, witnessing, bereavement, chronic community violence) in the United States. Guided by intersectionality as a sensitizing framework, analysis used iterative coding, cross-case matrices, and temporal analysis. Results. Seven mechanisms were identified: financial foreclosure, racialized institutional stigma, institutional recognition hierarchy, navigation burden, temporal abandonment, agency suppression, and compounding exposure. Mechanisms operated across rather than within exposure typologies, compounded one another systematically, and shaped identity from within while constraining material access. Outlier analysis specified the structural conditions under which each mechanism was suppressed. Intersectional analysis found that race and class operated both distinctly and in combination, producing qualitatively distinct burdens not reducible to either axis alone. Conclusions. The inequitable aftermath of firearm violence is structurally produced, not inevitable. The mechanisms identified provide a framework for intervention that targets institutional design failures, racialized processes, and temporally bounded support systems rather than individual-level coping. Intersectional analysis further demonstrates that race and class compound one another in ways that single-axis frameworks cannot capture, with direct implications for how interventions are designed and for whom.
