Forced displacement is associated with health service delivery disruption beyond the effects of armed conflict

Wait 5 sec.

IntroductionHumanitarian crises are increasing in frequency, duration, and severity. In 2022, armed conflicts caused more than 237,000 battle-related deaths (BRDs), and in 2024, the number of forcibly displaced people surged past 123 million—both the highest levels recorded this century1,2. Concurrently, natural disasters, intensified by climate change, now affect 175 million people annually3. These systemic shocks can exert extraordinary pressure on national health infrastructure, disrupting the supply chains and surveillance systems required to prevent infectious disease outbreaks4,5,6. Among essential health services, routine immunization is deeply embedded within and reflective of the health systems, and is widely recognized as a tracer indicator of broader health system performance, particularly during crises5,7,8. However, despite the growing frequency of these crises, the relationship between the magnitude of an emergency and the extent of service delivery disruption remains unclear due to methodological limitations, conflicting empirical findings, and—perhaps most importantly—theoretical oversights9.Current scholarship on vaccination coverage during humanitarian crises has tended to consider armed conflict and natural disasters separately. The displacement they generate may differ in form: armed conflict tends to produce protracted, often cross-border population movements10,11, whereas disaster-related displacement is more frequently sudden-onset, internal, and short-distance12. In both literatures, population displacement is often treated as a byproduct of these events rather than a distinct pressure5,13,14. We posit that displacement is itself an independent driver of health system disruption—a stressor that has been largely overlooked by existing research. While armed conflict and disasters can contract the local supply of healthcare by destroying facilities or triggering health worker flight, displacement rapidly reshapes the geographic distribution of demand15,16.When populations remain within a localized crisis zone, the public health emergency is usually bounded, allowing authorities to isolate and triage the shock. Conversely, mass displacement forces national health systems to chase a moving target, creating immense logistical difficulty as central administrators attempt to locate mobile populations, understand their needs, and reallocate finite resources under acute time constraints17,18,19. Further, we suggest that this dynamic is nonlinear20,21: many national health systems may possess enough organizational resilience to accommodate at least marginal demographic shifts22, but rapid, large-scale population movements can more frequently breach aggregate logistical and staffing thresholds. Past these breaking points, national service delivery declines precipitously until it hits an operational floor.To test this framework, we analyze longitudinal data from 193 countries between 2000 and 2023, evaluating sudden, large-scale surges in forced displacement (hereafter displacement shocks) against national childhood vaccination coverage. We deploy quasi-experimental difference-in-differences designs to estimate the aggregate impact of these shocks, and utilize nonlinear dose–response modeling to map the continuous threshold boundaries and saturation plateaus where national health systems stabilize or fail. Rather than relying on composite indices intended to describe state fragility, we use country-year displacement data to isolate the immediate demographic pressures pressing on a state’s infrastructure. To ensure our models capture an independent displacement signal, we simultaneously control for the concurrent pressures of armed conflict battle-related deaths (BRDs) and natural disaster fatalities. Finally, we contextualize these macro-level statistical patterns with targeted historical case vignettes, demonstrating the real-world administrative mechanisms through which acute displacement shocks precipitate national health system erosion, frequently leading to outbreaks of preventable diseases.Our analysis focuses primarily on the national level. In many countries, the national government coordinates health security policy, financing, and reporting23. We do not, however, assume that vaccination services are delivered exclusively by a single, legitimate central state: in conflict-affected settings, delivery can be fragmented across contested authorities, and even in stable states, externally supported supply chains may operate alongside or bypass national institutions24,25,26. The national level, however, is generally the unit at which immunization is coordinated internally and with international partners and, critically, is the level at which coverage itself is measured by WUENIC. When coverage at this aggregate level declines under crisis, the domestic population is left vulnerable regardless of which delivery systems have failed. Understanding the thresholds at which these aggregate systems buckle is therefore essential for national and international agencies to anticipate failures and build long-term institutional resilience. Nevertheless, we also examine subnational dynamics, looking at Nigeria as a descriptive case study (Supplementary Case Study 2) to understand the micro-level dynamics that may drive these broader country-level erosions.By disentangling population displacement from conflicts and disasters, our analysis demonstrates that population movement likely provides a powerful, independent signal of health system degradation. Our approach identifies operational thresholds, defined in both absolute numbers and population proportions, where system resilience collapses and coverage gaps emerge. Even in the absence of precise real-time data on refugee flows, order-of-magnitude indicators of population flight can alert decision-makers that a system is approaching its structural breaking point. In other words, these findings equip global health leaders with actionable risk metrics, enabling them to undertake actions like proactively deploying targeted resources and stabilizing supply chains before systemic service collapse occurs.ResultsGlobal associations between displacement and vaccination coverageAnalysis of 44,472 country-year-vaccine observations (Fig. 1) revealed a robust inverse relationship between population displacement and national vaccination coverage. Higher displacement was associated with lower vaccination rates, with coverage falling from an average of 90.6% in the lowest displacement quintile to 80.4% in the highest. Figure 2 shows meaningful descriptive associations between large-scale population movements and the erosion of childhood immunization.Fig. 1: Study design, data sources and analytical structure.Full size imagea The five source datasets merged into the analytic panel and their contributed variables: national childhood immunization coverage from the World Health Organization (WHO) and United Nations Children’s Fund (UNICEF) Estimates of National Immunization Coverage (WUENIC); forcibly displaced persons, comprising refugees and internally displaced persons (IDPs), from the United Nations High Commissioner for Refugees (UNHCR) Refugee Data Finder; battle-related deaths from the Uppsala Conflict Data Program (UCDP) Country-Year dataset on organized violence (version 25.1); natural-disaster mortality from the Emergency Events Database (EM-DAT), maintained by the Centre for Research on the Epidemiology of Disasters (CRED), Université catholique de Louvain (UCLouvain); and gross domestic product (GDP) per capita from the World Bank World Development Indicators (WDI). These were merged into a country-year-vaccine panel spanning 193 countries, 2000–2023 and 10 antigens (n = 44,472 observations). Italicized entries are datasets used only in robustness analyses: the Social and Economic Rights Fulfillment Index (SERF) Right to Health score, Varieties of Democracy (V-Dem) health equality, and Internal Displacement Monitoring Centre (IDMC) displacement data. b Countries in the panel by United Nations geographic region (light bars), with the subset experiencing at least one displacement shock shown as a nested darker segment, not an additional count. A displacement shock is defined as ≥50,000 displaced persons with ≥50% year-over-year growth; 51 of 193 countries qualify, and 19 qualify at the ≥500,000-person, 100%-growth threshold used in the primary specification. c Analytical structure. Core global analyses (solid outlines) generate Figs. 2–5: ordinary least squares (OLS) regression, difference-in-differences average treatment effects on the treated (ATTs), event-study models with HonestDiD pre-trend sensitivity analysis, and dose–response curves from generalized additive models (GAMs) fitted by restricted maximum likelihood (REML) at basis dimension k = 8. The national and subnational case studies (dashed outlines; Supplementary Case Studies 1 and 2) are descriptive and are not used to estimate treatment effects; they draw on the International Organization for Migration Displacement Tracking Matrix (IOM DTM) and the Demographic and Health Surveys (DHS). Robustness analyses are listed beneath. No statistics were derived in this figure.Fig. 2: Descriptive statistics and regression results.Full size imageSpearman ρ correlations between vaccinations (a) and between primary independent variables (b), computed with pairwise-complete observations; no P values are displayed (a, n = 1295–4661 country-years; b, n = 37,500–44,472 country-year-vaccine observations). Boxplots of vaccination coverage stratified by displacement quintile, battle-related death (BRD) tier, and disaster death tier (c). Boxplots in (c) (n = 38,405 country-year-vaccine observations with non-missing coverage) show the median (center line) and the 25th and 75th percentiles; whiskers extend to the most extreme observation within 1.5 times the interquartile range of the box. Outlying observations are not plotted and the vertical axis is truncated at the 5th and 99th percentiles of the pooled coverage distribution for legibility; no observations are excluded from the underlying distributions. Battle-related-death tiers in (c) are shown at four levels (0, 1–24, 25–999 and ≥1000 deaths), whereas the regression models in (e) and all difference-in-differences analyses use the conventional three-level classification (no conflict