The term 'natural experiment’ has a murky conceptual history in public health, with definitions ranging from naturally occurring states to unspecified approximations of randomised trials. Craig and colleagues recently defined natural experiments as events creating exposed and unexposed groups outside researcher control, but this is largely indistinguishable from any observational study. One could define natural experiments as cases of randomised treatment assignment by a third party but expanding beyond this creates complications. We argue that grouping study designs with labels like 'natural experiments’ or 'quasi-experiments’ provides potentially misleading guidance for causal inference given the difficulty of justifying the core assumption of ‘as-if randomisation’ across diverse designs. Rather than relying on vague labels, researchers should explicitly state their design and defend the credibility of assumptions needed for causal inference. Expansive definitions of ‘natural experiments’ can potentially bias evidence synthesis by obscuring the rigorous justifications needed for causal inference, ultimately creating more confusion than clarity.
BACKGROUND:Religious and faith-based institutions play a critical role in the social infrastructure that supports population health. While prior research has focused largely on individual religious participation or the presence of congregations within neighbourhoods, far less is known about the population health implications when places of worship close within communities. In this study, we examine church closures as a novel neighbourhood-level indicator and its association with self-rated health (SRH). METHODS:We linked 2015 and 2018 Southeastern Pennsylvania Household Health Survey individual SRH data (n=16 294 adults) across 219 ZIP Code Tabulation Areas to churches that closed between 2009 and 2014. Church closure data were from Data Axle. Church closure density (closures/km²) was modelled using multilevel linear models and spline models, adjusting for individual sociodemographic factors and neighbourhood characteristics. RESULTS:Church closures were concentrated in socioeconomically disadvantaged neighbourhoods with higher proportions of racial/ethnic minority residents (58.6% below median income, Black: 36.0%). Spline regression models revealed significant non-linear associations. Across the lower range of closure density (below the -0.5 SD threshold, approximately the 53rd percentile), each SD increase in closure density was associated with a 0.385-point lower SRH score (95% CI -0.644 to -0.127) in the fully adjusted baseline-adjusted model. Above this threshold, the association was significantly attenuated (β=0.407, 95% CI 0.152 to 0.662), consistent with a near-null net effect in higher-closure-density areas. CONCLUSIONS:Neighbourhood church closures were associated with poorer SRH in specific contexts and can be considered an indicator of institutional loss.
BACKGROUND:Few studies have assessed the impact of large-scale early childhood education and care (ECEC) programmes. In this study, we evaluated the impact of national-level policy changes providing access to free ECEC during children's preschool years on symptoms of mental distress during early adolescence. METHODS:We used data from nine European countries that participated in five waves of the Health Behaviour in School-aged Children (HBSC) Study conducted between 2002 and 2018, including 165 656 participants who were born between 1988 and 2003, and surveyed when they were 11-15 years old. Primary outcomes were feeling low, feeling irritable, feeling nervous and having difficulty sleeping more than once per week in the past 6 months. Our difference-in-differences design compared outcomes across birth cohorts in five treated countries that expanded access to free ECEC to changes in control countries that did not. We examined heterogeneity by gender and family affluence. RESULTS:One year of access to free ECEC reduced the prevalence of feeling low by 1.3 percentage-points (95% CI 2.5 to -0.1), feeling irritable by 2.1 percentage-points (95% CI -4.0 to -0.3) and feeling nervous by 1.8 percentage-points (95% CI -2.7 to -0.9) and did not appreciably affect the prevalence of having difficulty sleeping. We did not observe evidence of a marked beneficial effect across all treated countries, and stratified analyses did not show heterogeneity in these associations by gender or family affluence. Our main results were robust to alternative control groups. CONCLUSION:Our findings suggest that the effects of free ECEC policies on adolescent mental health are context-specific.
BACKGROUND:Poor school attendance adversely affects youth behaviour and health trajectories. We reviewed and synthesised the evidence from randomised controlled trials (RCTs) on interventions to improve school attendance. METHODS:This systematic review and meta-analysis updates the Education Endowment Foundation (EEF) 2022 review on attendance interventions. We synthesised RCT data to inform recommendations. We searched ERIC (via EBSCOhost), PsycINFO, Web of Science and Google Scholar for publications 1 January 2020-16 October 2024 to identify attendance interventions delivered to students, parents/guardians or school staff. We also extracted and analysed RCTs identified in the EEF review. Two reviewers independently extracted data from published articles and assessed risk of bias and certainty of evidence (Grading of Recommendations Assessment, Development and Evaluation, GRADE assessment). We synthesised and meta-analysed data per our protocol (PROSPERO: CRD42024610037). RESULTS:We screened 9366 titles and abstracts and included 61 articles (2000-2024): 43 articles published 2020-2024 plus 18 articles from the previous review (2000-2020), reporting 57 trials of 54 interventions. Pooled mean difference and 95% CIs for days of school attended was 0.63 (-0.34 to 1.61), I2=64.0%, low certainty, for mentoring interventions and 0.43 (0.10 to 0.76), I2=91.9%, moderate certainty, for parental engagement interventions. Other intervention areas including targeted approaches, behavioural programmes and social-emotional interventions had smaller, heterogeneous evidence bases. Risk of bias was moderate-to-low in most trials. CONCLUSION:A large range of interventions exist to address the diverse causes of school absence. Parental engagement demonstrated a small positive effect on attendance, particularly among younger age groups. Most trials were from the USA; implementation and evaluation in other countries will inform effectiveness.
The recently updated UK Medical Research Council's guidance on the use of natural experiments to evaluate population health interventions, published in 2025, adopted a broad definition of natural experiments as any event or process that divides a population into groups that differ in their exposure to the intervention of interest and that is not under the control of a researcher. Many authors adopt narrower definitions that only include cases where exposure is determined in an 'as-if random' or 'plausibly exogenous' way. I explain why we used a broad definition in our guidance, why narrower definitions are hard to apply consistently and what might be gained or lost by adopting a broad or a narrow one.
BACKGROUND:Mortality from alcohol-related liver disease (ARLD) in England rose after the COVID-19 pandemic. Existing reports lack data on prior healthcare use, which is key to understanding care pathways before death. This study examined long-term trends in ARLD mortality before, during and after the pandemic and assessed changes in relation to prior inpatient care. METHODS:In this population-based observational study using English national mortality records linked to NHS hospital data, all ARLD deaths in adults aged 25-84 were analysed from January 2004 to December 2024. Monthly mortality trends were examined by age, sex, deprivation, region and prior inpatient care. Interrupted time series estimated changes in mortality after the pandemic onset. RESULTS:Over the study period, 97 058 ARLD deaths occurred, with about one-third out-of-hospital. Following the pandemic, ARLD mortality increased by 27.7% (95% CI 23.3% to 32.3%), representing 5571 excess deaths between March 2020 and December 2024. The step-increase was greater for out-of-hospital deaths (34.9%, 95% CI 28.4% to 41.7%) than for in-hospital deaths (22.9%, 95% CI 18.2% to 27.8%), and the increase in out-of-hospital deaths was driven by deaths in people who had no recent discharge in the previous year (75.7%, 95% CI 61.7% to 90.9%). CONCLUSIONS:The increase in deaths from ARLD has been driven by a disproportionate increase in those without recent inpatient care. Efforts to understand ARLD mortality should increasingly be focused on the out-of-hospital setting.
BACKGROUND:Children in households affected by adult substance misuse face intersecting educational risks. While associations between household adversity and school disengagement are documented, research often relies on siloed or self-report data. This study used national linked data to estimate associations with children's school attendance and exclusion, using adult health and justice contacts as administrative proxies for substance misuse exposure during household cohabitation. METHODS:This retrospective cohort study used the Secure Anonymised Information Linkage (SAIL) Databank in Wales. Children aged 11-15 were followed across academic years between 2017 and 2019. Exposure was defined as residing with an adult having either a substance misuse health contact, a justice contact or both health and justice contacts. Outcomes were attendance rate and ever being excluded from school. Multilevel models adjusted for child mental health, special educational needs, adverse childhood experiences and deprivation. RESULTS:Among 180 388 children, 14.7% were exposed to household substance misuse. Risks peaked in the combined health and justice group, showing significantly lower attendance (2.81% reduction, CI 2.97 to 2.64) and the highest exclusion odds (OR 6.67, CI 4.01 to 11.08) versus unexposed peers. Deprivation independently predicted poorer outcomes and amplified the negative effects of household adversity. INTERPRETATION:Household substance misuse, particularly involving the justice system, is associated with extreme education disruption vulnerability. These risks remain largely invisible in educational records. Policy must prioritise integrated data sharing to improve early identification and cross-sectoral support. Current siloed approaches are insufficient to break the link between household adversity and educational disengagement.
BACKGROUND:Stressful life events are recognised determinants of mental health, yet their effects in later life remain unclear. Reflecting increasing longevity, our study examined links between stressful life events, both personal and work-related, and mental health outcomes among middle-aged and older Australians aged 50 and over. METHODS:We analysed 95 603 person-year observations from 11 187 individuals across 16 waves (2006-2021) of the nationally representative Household, Income and Labour Dynamics in Australia Survey. Key personal (eg, separation, physical violence, serious personal injury or illness, detention in jail) and work-related (eg, retirement, fired or made redundant, major worsening in finances) stressful life events were assessed. Mental health outcomes were assessed using the five-item Mental Health Inventory (MHI-5; range: 0-100) and the Mental Component Summary (MCS; range: -1.21-76.19), with lower scores indicating poorer mental health. Fixed-effects generalised least-squares models estimated within-individual changes, with stratification by age and gender. RESULTS:Personal and work-related stressful life events were consistently associated with poorer mental health outcomes. Separation (MHI-5: β = -3.790, 95% CI -5.235 to -2.345; MCS: β = -2.261, 95% CI -3.153 to -1.368), physical violence (MHI-5: β = -4.603, 95% CI -6.251 to -2.954; MCS: β = -2.562, 95% CI -3.601 to -1.523) and serious personal injury or illness (MHI-5: β = -3.620, 95% CI -4.916 to -2.323; MCS: β = -2.463, 95% CI -3.258 to -1.667) were associated with substantial reductions in mental health. Major worsening in finances produced the largest declines overall (MHI-5: β = -4.191, 95% CI -6.287 to -3.552; MCS: β = -2.714, 95% CI -3.561 to -1.867). Effect sizes varied across gender and age groups. CONCLUSION:Personal and work-related stressful life events adversely affect mental health among Australian middle-aged and older adults, highlighting the need for targeted policies, interventions and community support.
BACKGROUND:Rates of intimate partner homicides of women in the USA are high relative to other high-income countries, especially those involving firearms. We evaluated one potential policy solution by examining state Extreme Risk Protection Order (ERPO) firearm transfer laws and intimate partner homicide of women. METHODS:The primary outcome was state rates of intimate partner homicide deaths among females aged 10-64 years per 100 000 using data from the Supplementary Homicide Reports (2014-2022). We also examined firearm homicide rates using vital statistics data (2014-2023), overall and stratified by racialised groups. This study used a two-stage least-squares instrumental variable approach where our exposure was the annual count of ERPO petitions filed in each state per 100 000 population and our instrument was the presence of an ERPO policy. RESULTS:Overall, we did not observe statistically significant reductions in state rates of intimate partner homicides (beta=-0.05 per 100 000; 95% CI -0.11 to 0.01) or firearm homicides among women (beta=-0.04 per 100 000; 95% CI -0.13 to 0.05) associated with ERPO implementation. However, among white women, each annual increase of 1 per 100 000 in the rate of ERPO petition filings was associated with a statistically significant decrease in the rate of firearm homicides of -0.04 per 100 000 (95% CI -0.08 to -0.01, p=0.022). CONCLUSIONS:Our findings suggest the potential for ERPO use to prevent fatal violence against white women and underscore the need for equitable policy solutions for black and Indigenous women who face the highest rates of intimate partner violence.
BACKGROUND:Physical activity (PA) is a widely recognised modifiable risk factor for cognitive health. The role of diet and their combined associations on cognition in midlife remains unclear. We assessed the additive and interactive associations of moderate-to-vigorous PA (MVPA) and diet quality on cognition in midlife. METHODS:At age 46, participants from the 1970 British Cohort Study wore activPAL accelerometers for 1 week and completed the Oxford WebQ dietary questionnaire, from which the Pyramid-Based Mediterranean Diet Score (PyrMDS) was derived. Global cognitive z-scores were derived from standardised tests of processing speed, verbal memory and fluency at age 46 and 51. Multivariate linear regression examined independent and mutually adjusted associations between MVPA and PyrMDS at age 46 with cognitive z-scores at both timepoints. Combined MVPA-PyrMDS tertiles and interaction terms were also tested. RESULTS:In a sample of n=3028 (55% female), mean MVPA was 52±25 min/day and mean PyrMDS was 6.31±1.65. In mutually adjusted models at age 51, only PyrMDS was associated with global cognitive z-scores after covariate adjustment (per hour MVPA: β=0.228, 95% CI -0.026 to 0.481; per point PyrMDS: β=0.045, 95% CI 0.025 to 0.065). In combined analyses, PyrMDS was associated with better age 51 global cognitive z-scores irrespective of MVPA (high-PyrMDS:low-MVPA vs low-PyrMDS:low-MVPA: β=0.185, 95% CI 0.046 to 0.325). No interactions were observed. CONCLUSION:Diet quality demonstrated prospective associations with better cognition independently of MVPA, with associations remaining more consistent than for PA after adjustment of confounders. Diet quality may represent an under-recognised lifestyle target for cognitive health.
BACKGROUND:Socioeconomic inequalities in children's out-of-home care (OHC) are a major public health problem. Though widely documented, no review has systematically synthesised evidence across high-income countries. We aimed to compare the risk of OHC for children experiencing disadvantaged versus advantaged socioeconomic conditions across Organisation for Economic Co-operation and Development countries. METHODS:We conducted a systematic review and meta-analysis, searching eight databases for studies published between January 1980 and April 2026. We included observational studies reporting quantitative associations between socioeconomic conditions and OHC in representative child populations. Two reviewers independently screened records, extracted data and assessed quality. We used random-effects meta-analysis stratified by exposure type and effect measure, along with harvest, forest, line and scatter plots, to synthesise findings. RESULTS:Of 12 328 unique records screened, 150 studies from 13 countries met inclusion criteria, yielding 447 estimates. Of these, 390 showed positive associations between disadvantaged socioeconomic conditions and OHC across area deprivation, parental employment and education, income and eligibility for social assistance. The meta-analysis included 79 estimates across 19 strata. Very high heterogeneity across pooled estimates limited interpretation. However, associations were consistently positive and most consistent for area deprivation and eligibility for social assistance. Sensitivity analyses using higher-quality studies strengthened our main findings. CONCLUSIONS:Across high-income settings, disadvantaged socioeconomic conditions were consistently associated with increased risk of OHC. Policies addressing the socioeconomic drivers of childhood adversity may help reduce care entry. Limitations of studies included socioeconomic conditions rarely being a primary study focus, inconsistent confounder adjustment and high heterogeneity. Future research should evaluate the impact of social and economic policy reforms on OHC risk. PROSPERO REGISTRATION NUMBER:CRD42021266991.
BACKGROUND:The aim of this study is to investigate educational differences in labour market participation expressed by working life expectancy and working years lost due to various labour market states among men and women. METHODS:This study used a random sample of 100 000 individuals from the Swedish register-based Swedish Work Illness and labour market Participation cohort, including individuals born 1945-1975 (aged 30-60 in 2005). The multistate expected labour market affiliation method was applied to estimate working life expectancy and expected working years lost due to unemployment, sickness absence, other non-work, disability pension, early old-age pension and death from 2006 to 2020 or until age 65. Education was categorised as ≤9 years, 10-11 years, 12 years, 13-14 years and ≥15 years ranging from compulsory education only to university education. RESULTS:Those with lower educational attainment could expect substantially shorter remaining working lives than those with higher educational attainment. At age 30, the difference between the highest and lowest educational groups was 5.6 years for men and 8.9 years for women. A clear educational gradient was observed for expected time in unemployment and disability pension, with lower education associated with more time in these states. This pattern was less clear for sickness absence, other non-work, early old-age pension and death. CONCLUSIONS:Shorter working lives after age 30 among those with lower educational attainment highlight marked inequalities in labour market participation. This has both immediate and long-term consequences for individuals which can in turn lead to reduced income and possibly poorer health.
BACKGROUND:Prior research indicates that food insecurity is negatively associated with cardiometabolic health and plays a role in cardiovascular disease (CVD) risk, which also increases with age. Food insecurity disproportionately affects ageing women in the USA, who face higher poverty and economic insecurity rates than their male counterparts, increasing their risk for hunger, poor nutrition and chronic conditions. We examined the association of food security and assistance with CVD morbidity and mortality risks in postmenopausal women through a target trial emulation. METHODS:A retrospective cohort study was conducted among 41 227 postmenopausal women from the Women's Health Initiative Observational Study free of CVD at baseline. Approximately 96% reported food security and 3% reported reliance on food assistance at 6 years of follow-up (analytical baseline). Furthermore, 16% of 1861 women reporting food insecurity also reported food assistance. Inverse probability of treatment weighting (IPTW) within Cox regression models was performed using demographic, socioeconomic, lifestyle and health characteristics collected prior to the analytical baseline. RESULTS:Overall, 6037 eligible women experienced CVD outcomes over ~24 years of follow-up. In IPTW Cox regression models, food security (HR=0.65, 95% CI 0.47 to 0.89) and food assistance (HR=0.47, 95% CI 0.46 to 0.48) were inversely related to CVD risk. Similar results were obtained for all-cause and chronic disease-specific mortality risks. CONCLUSION:Food security and assistance are generally associated with lower CVD risks after menopause, with implications for food assistance programmes, targeted nutritional counselling, community interventions and holistic care among ageing women in the USA.
BACKGROUND:Long-term historical trajectory of violence remains insufficiently understood. METHODS:This study examines long-term changes in natural/non-natural mortality by comparing cause-of-death data from late medieval and contemporary Milan. It aims to reconstruct medieval patterns and assess changes in age-specific and sex-specific distributions of violent and accidental deaths. Causes of death recorded in the Liber Mortuorum (LM), a civil death register in use in Milan since at least 1451, were compared with 21st-century mortality data from the Milan province. Causes of death in the LM were translated from Latin and classified as natural or non-natural (accidents, homicides or suicides). Comparisons were made using the age-adjusted proportional mortality ratio (PMR) and the female-to-male ratio. RESULTS:The LM recorded 16 288 deaths over 8 non-consecutive years between 1452 and 1485, 15 861 with age at death available. These included 217 deaths from non-natural causes: 118 accidents, 68 homicides, 7 suicides and 24 deaths of undetermined intent. Accidental deaths accounted for 61.1% of non-natural deaths with determined intent in the LM compared with 79.9% in contemporary data. Among non-natural causes, deaths from homicides were more frequent in the medieval period than in contemporary times (35.2% vs 1.5%, age-adjusted PMR: 7.6). In the medieval period, 5.9% of homicide victims were female compared with 31.5% in recent data, with the female-to-male mortality ratio increasing from 0.06 to 0.46. Suicides were rarely reported in the LM. CONCLUSIONS:Homicide was more frequent in general in medieval times, even though female homicide appears proportionally more represented in contemporary data than in the 15th-century records and represents a key finding of the study.
BACKGROUND:Proportionate universalism aims to improve overall health while decreasing health inequity across the social spectrum. For health visiting in England, the 'universal' offer is five health reviews for children aged under 5 years, with 'proportionate' support provided through additional health visiting contacts when greater need is identified. METHODS:We synthesised the published findings of four mixed-method studies into variation in health visiting services in England to examine how proportionate universalism works in practice. The studies analysed Community Services Dataset data from 2016-2020 and interviews and observations with parents and professionals (collected 2023-2024). RESULTS:Between 2018 and 2020, up to 98% of infants in 57 areas received their universal contacts, even as services experienced reduced funding and staffing, indicating high reach. However, interview data suggest universal reach was widely achieved by offering a differentiated version of universal contacts based on whether families already have known needs or not. This affects health visiting's ability to identify need across the social spectrum. Similarly, additional contacts were found to be widespread. Health visitors typically reported using these to support families with known complex needs that fell short of thresholds for other services. CONCLUSIONS:In an environment of resource constraint and limited specialist services, a proportionate universalist service will be pushed away from identifying unknown needs and pulled towards supporting the most acute known needs. This may reduce the quality of the universal service, even if reach is maintained. Implications for policymakers and service commissioners are discussed.
Background Ethnicity data are essential for understanding health inequalities yet incomplete data can bias analyses and undermine representation. Reporting guidelines recommend that studies report how missing data are handled as methodology varies. However, little is known about trends in ethnicity missingness method use. Patient and Public Involvement (PPI) can shape equitable practices although its role in decisions about handling missing ethnicity remains unclear. This review aimed to describe trends in handling missing ethnicity data, reporting guideline adherence and the role of PPI. Methods A systematic review identified approaches to handling missing ethnicity data in studies using national primary care databases in England and Wales. Medline, Cochrane Library, CINAHL, Scopus, Web of Science and database bibliographies were searched. Approaches to handling missing ethnicity were extracted and grouped (eg, ‘missing/unknown’ categorisation exclusion imputation). Time trends, reporting guidelines and PPI use were summarised. Results 313 studies were included. Over half used Clinical Practice Research Datalink. Nearly half linked to secondary care data; only one in four supplemented ethnicity using secondary sources. The most frequent methods were ‘missing/unknown’ categorisation (85, 27.2%) and exclusion (83, 26.5%); 71 studies (22.7%) used imputation. Nearly one-third of studies (99) provided no or unclear details; 100 (31.9%) cited reporting guidelines. 38 studies (12.1%) mentioned PPI; none detailed PPI involvement in missing data choices. Conclusion Handling of missing ethnicity data in primary care research in England and Wales is inconsistent and poorly reported. Greater transparency is needed to improve interpretability, comparability and inclusivity of research using routine health data.
BACKGROUND:Although avoidable mortality in South Korea has fallen substantially, whether socioeconomic inequality gradients have narrowed in parallel remains unclear. We assessed 13-year trends (2012-2024) in absolute and relative inequality in avoidable, preventable and treatable mortality across approximately 250 districts, weighting by district population rather than equal quintiles. METHODS:We adopted an ecological repeated cross-sectional design. District-level age-standardised mortality rates, calculated from cause-of-death microdata via the Korean Microdata Integrated Service, were linked to the Area Deprivation Index from census microdata. Districts were grouped into deprivation quintiles using ridit scores from actual population shares. The Slope Index of Inequality (SII) and Relative Index of Inequality (RII) were estimated annually via weighted least squares regression. RESULTS:Population-weighted avoidable mortality fell from 204.9 to 138.0 per 100 000 (-32.7%). Q1 (least deprived) held 34-35% of the population; Q5 (most deprived) only 4-5%. The SII for avoidable mortality showed no significant trend (56.6 to 53.5; slope+0.02/year, p=0.925), yet the RII rose from 0.276 to 0.388 (slope+0.011 /year, p<0.001). For treatable mortality, the SII climbed 65.4% (10.4 to 17.2, p<0.001) and the RII more than doubled (0.177 to 0.398, p<0.001). The preventable-to-treatable SII ratio converged from 4.4 to 2.1, and the RII ratio fell below unity for the first time in 2023, indicating that relative inequality in treatable mortality now exceeds that in preventable mortality. CONCLUSION:Population-weighted analysis reveals an inequality paradox-stable absolute inequality (SII) with rising relative inequality (RII)-masked under conventional equal weighting, which yields a spurious SII decline and non-significant RII trend. Treatable mortality showed simultaneous widening of absolute (+65.4%) and relative (+124.9%) inequality, with its RII surpassing that of preventable mortality for the first time in 2023. These findings call for population-weighted monitoring and policies addressing structural barriers to healthcare access beyond financial coverage.
