Individuals with mental disorders face a substantially higher risk of mortality and are more likely to be lonely, socially isolated, and with low social support compared to those without mental disorders. We aimed to qua...Individuals with mental disorders face a substantially higher risk of mortality and are more likely to be lonely, socially isolated, and with low social support compared to those without mental disorders. We aimed to quantify the extent to which the observed mortality gap associated with mental disorders could be explained by these social factors. This cohort study included 162,483 participants from the Danish National Health Survey in 2013 and 2017 who were followed for six years after survey participation. Survey data on social disconnection (loneliness, social isolation, social support in the form of perceived emotional support, and a composite measure) was linked with register data on hospital-diagnosed mental disorders and mortality. We applied G-computation-based causal decomposition to compare the sex-specific relative risk of mortality associated with mental disorders under a natural course to a counterfactual scenario in which all individuals had a distribution of social disconnection similar to individuals without mental disorders. We found that social disconnection and the distribution of loneliness, social isolation, and social support accounted for 10-34% of the mortality gap associated with mental disorders among men and 2-20% among women, assuming a causal effect of social disconnection on mortality. The largest contributions were found for social isolation and loneliness, whereas the smallest were found for social support. Our results highlight the possibility that different aspects of social disconnection, especially social isolation and loneliness, could explain part of the mortality gap associated with mental disorders, with larger contributions among men than women.
He YQ, Xue WQ, Diao H
… +46 more, Zhan JY, Ji MF, Yang DW, Zhao Y, Deng CM, Wu ZY, Zhou T, Liao Y, Zheng MQ, Zhang WL, Jia YJ, Yuan LL, Luo LT, Li DH, Wang TM, Tong XT, Du Y, Tang LL, Huang JW, Huang CL, Zhao ZY, Wu YX, Cao LJ, Dong SQ, Wang F, Jiang CT, Xiao RW, Zhang WB, Chen XY, Wang QL, Liu QY, Zhao YZ, Tang CL, Ma L, Zheng XH, Zhang PF, Li XZ, Zhang SD, Hu YZ, Yu X, Wu BH, Li FG, Wu JH, Deng BS, Liang XJ, Jia WH
Eur J Epidemiol
· 2026 Feb · PMID 41528651
·
Full text
The global rise of non-communicable diseases (NCDs) presents an urgent public health challenge, particularly in regions undergoing rapid economic and demographic transitions. Guangdong Province, China's most populous and...The global rise of non-communicable diseases (NCDs) presents an urgent public health challenge, particularly in regions undergoing rapid economic and demographic transitions. Guangdong Province, China's most populous and economically advanced region, is experiencing a substantial and accelerating burden of NCDs. However, large-scale, population-based cohorts from this region remain scarce, limiting insights into region-specific disease determinants and prevention strategies. The Guangdong Biobank Cohort (GDBC) was established in 2017, enrolling 35,081 participants aged 40-84 years from urban and rural areas of Zhongshan City in the Pearl River Delta. At baseline, comprehensive data on 346 variables-including lifestyle, environmental exposures, medical histories, physical examinations, and laboratory profiles-were collected via a cloud-based member management information system (MMIS), alongside blood and saliva samples for biobanking. A sub-cohort underwent genome-wide genotyping (N = 2,530) and oral microbiome profiling via 16 S rRNA sequencing (N = 2,049). During dynamic follow-up, 44.2% (N = 15,499) completed Phase I resurvey with repeated measurements and updated biospecimens. Disease outcomes, including hypertension, diabetes, and cancer, were ascertained through active surveillance and regional registry linkage until December 2023. Baseline prevalence of hypertension, diabetes, and cancer was 25.3%, 8.0%, and 3.6%, respectively. Over follow-up, 1,767 hypertension cases, 814 diabetes cases, and 558 cancers were recorded, yielding crude incidence rates of 1,804.6, 649.7, and 423.1 per 100,000 person-years, respectively. The GDBC provides a comprehensive, dynamically updated resource to dissect gene-microbiome-environment interactions and develop precision prevention strategies to inform public health policies.
Age-related macular degeneration (AMD) is a leading cause of irreversible vision loss worldwide. However, evidence regarding the relationship between air pollution and AMD is limited, and the modifying effect of genetic...Age-related macular degeneration (AMD) is a leading cause of irreversible vision loss worldwide. However, evidence regarding the relationship between air pollution and AMD is limited, and the modifying effect of genetic susceptibility on this association remains unknown. A total of 445,237 participants without AMD at baseline were included from the UK Biobank. The concentrations of nitrogen dioxide (NO), nitrogen oxides (NO), particulate matter (PM, PM, PM) were collected by using land-use regression models. Air pollution score (APS) was constructed through summing each pollutant weighted by the regression coefficients with AMD from single-pollutant model. Cox proportional hazard models were used to evaluate hazard rations (HRs) and 95% confidence intervals (95%CIs) of associations between air pollutants and polygenic risk score (PRS) with incident AMD. During a median follow-up of 13.83 years, we observed 9,635 incident AMD events. The HR (95%CI) of incident AMD for each standard deviation increase in NO, NO, PM, PM, and APS were 1.04(1.02, 1.06), 1.03(1.01, 1.05). 1.04(1.02, 1.07), 1.02(1.00, 1.04), and 1.04(1.02, 1.06), respectively. Significant additive interaction effects of NO, NO, PM, APS and PRS with incident risk of AMD were observed, with the relative excess risk due to the interaction (RERI), attributable proportion (AP), and their 95% CIs of 0.10(0.01, 0.18) and 0.05(0.01, 0.11) for NO, 0.11(0.01, 0.19) and 0.05(0.02, 0.10) for NO, 0.15(0.06, 0.23) and 0.08(0.03, 0.13) for PM, and 0.12(0.03, 0.20) and 0.06(0.01, 0.11) for APS, respectively. Compared with participants exposed to low level of above air pollutants and low PRS, those exposed to high air pollution and high PRS had almost double incident risk of AMD [HR(95%CI) ranged from 1.83(1.68, 1.99) to 2.03(1.86, 2.21)]. Long-term exposure to air pollutants NO, NO, PM, and PM showed positive associations with increased risk of AMD, which could be further enhanced by genetic susceptibility.
Muszynska-Spielauer M, Di Giulio P, Minagawa Y
… +2 more, Di Lego V, Luy M
Eur J Epidemiol
· 2026 Mar · PMID 41524888
·
Full text
This study tests the "longevity hypothesis," which posits that women's greater number of years spent in poor health is primarily a direct consequence of their longer survival. We analyse gender differences in unhealthy l...This study tests the "longevity hypothesis," which posits that women's greater number of years spent in poor health is primarily a direct consequence of their longer survival. We analyse gender differences in unhealthy life years (ULY) at age 50 across 22 European countries in 2015-2017. ULY was estimated using three approaches-the Sullivan method, the cross-sectional average length of healthy life, and multistate life tables-applied to four health indicators of varying severity: chronic diseases, functional limitations, self-rated health, and disability. Data were drawn from the Human Mortality Database and the Survey of Health, Ageing and Retirement in Europe. We decomposed the gender gap in ULY into a "mortality effect" (ME), reflecting differences in life years lived, and a "health effect" (HE), reflecting differences in morbidity prevalence. Women at age 50 lived more unhealthy years than men across almost all health indicators and countries. In most cases, more than half of the gender gap in ULY was attributable to the ME, indicating that women's longer survival primarily explains their greater number of years spent in poor health. The HE showed greater variation across indicators and countries. Results were most consistent for chronic diseases and self-rated health, while functional limitations and disability yielded smaller and less consistent differences. Findings support the longevity hypothesis: women's higher life expectancy is the main driver of their longer lifetime spent in poor health. The variation across health dimensions highlights the importance of distinguishing between them when studying gender inequalities in health.
Observational studies have linked maternal hormonal contraception use to childhood cancer risk, but findings are inconsistent. A systematic review was conducted of this potential relationship. A systematic search was per...Observational studies have linked maternal hormonal contraception use to childhood cancer risk, but findings are inconsistent. A systematic review was conducted of this potential relationship. A systematic search was performed in PubMed, Embase, Scopus, Cochrane, and Web of Science databases until April 9, 2025. Studies reporting maternal hormonal contraception use before or during pregnancy and childhood cancer risk (0-19 years) were eligible. We included studies providing risk estimates in English or Scandinavian languages. Newcastle-Ottawa Scale was used to assess study quality. Meta-analysis using fixed and random effects was used to pool relative risks (RRs) with 95% confidence intervals (CIs) for childhood cancer according to maternal hormonal contraception use (1) up to or during pregnancy, and (2) exclusively during pregnancy. We included 27 studies (24 case-control and 3 cohort), totaling 11,067 childhood cancer cases. Maternal hormonal contraception use up to and during pregnancy increased risk of any childhood cancer (RR = 1.18; 95% CI = 1.10-1.26), leukemia (RR = 1.24; 95% CI = 1.06-1.45), and lymphoid leukemia (RR = 1.17; 95% CI = 1.06-1.28). Exposures during pregnancy showed higher risk estimate for any cancer (RR = 1.32; 95% CI = 1.12-1.56) and leukemia (RR = 1.63; 95% CI = 1.07-2.49). Most studies were moderate (70%) or high (26%) quality. Maternal hormonal contraception use may increase childhood cancer risk, particularly for leukemia, and during pregnancy. Further prospective studies are needed, focusing on specific hormonal contraception substances and exposure timing.
Machine learning (ML) algorithms are increasingly used to estimate propensity score with expectation of improving causal inference. However, the validity of data-driven ML-based approaches for confounder selection and ad...Machine learning (ML) algorithms are increasingly used to estimate propensity score with expectation of improving causal inference. However, the validity of data-driven ML-based approaches for confounder selection and adjustment remains unclear. In this study, we emulated the device-stratified secondary analysis of the PARADIGM-HF trial among U.S. veterans with heart failure and implanted cardiac devices from 2016 to 2020. We benchmarked observational estimates from three propensity score approaches against the trial results. (1) logistic regression with pre-specified confounders (2), generalized boosted models (GBM) using the same pre-specified confounders, and (3) GBM with expanded covariates and automated feature selection. Logistic regression-based propensity score approach yielded estimates closest to the trial (HR = 0.93, 95% CI 0.61-1.42; 23-month RR = 0.86, 95% CI 0.57-1.24 vs. trial HR = 0.81, 95% CI 0.61-1.06). Despite better predictive performance, GBM with pre-specified confounders showed no improvement over the logistic regression approach (HR = 0.97, 95% CI 0.68-1.37; RR = 0.96, 95% CI 0.89-1.98). Moreover, GBM with expanded covariates and data-driven automated feature selection substantially increased bias (HR = 0.61, 95% CI 0.30-1.23; RR = 0.69, 95% CI 0.36-1.04). Our findings suggest that ML-based propensity score methods do not inherently improve causal estimation possibly due to residual confounding from omitted or partially adjusted variables and may introduce overadjustment bias when combined with automated feature selection. These results underscore the importance of careful confounder specification and causal reasoning over algorithmic complexity in causal inference.
Johansen M, Omsland TK, Laine K
… +2 more, Håberg SE, Magnus MC
Eur J Epidemiol
· 2026 Feb · PMID 41524885
·
Full text
Women with endometriosis have a higher burden of anxiety and depression. Whether they are at increased risk of postpartum depression (PPD) remains unclear. We aimed to compare the risk of PPD between women with and witho...Women with endometriosis have a higher burden of anxiety and depression. Whether they are at increased risk of postpartum depression (PPD) remains unclear. We aimed to compare the risk of PPD between women with and without endometriosis and to explore mediation by previous history of major depression and infertility. In a population-based cohort study, we compared 1,159 singleton pregnancies to women with self-reported endometriosis and 74,590 pregnancies to women without endometriosis. We calculated a djusted risk ratios (aRR) with 95% confidence intervals (CI) using multivariable log-binomial regression, adjusting for age, body mass index, education and income. Mediation analyses assessed the indirect effect of any history of major depression or infertility. Women with endometriosis had a higher risk of PPD (aRR: 1.34, 95% CI: 1.15-1.55). Mediation analyses indicated that a large part of this association was explained by a higher lifetime prevalence of major depression among women with endometriosis (natural direct effect of endometriosis: aRR: 1.17, 95% CI: 1.00-1.36; natural indirect effect of any history of major depression: aRR: 1.14, 95% CI: 1.08-1.20), with 49.3% proportion mediated. Infertility demonstrated a negative natural indirect effect on the association between endometriosis and PPD (aRR: 0.87, 95% CI: 0.81-0.94). Women with endometriosis had an elevated risk of PPD which was largely explained by a higher lifetime prevalence of major depression. Our findings suggest that they constitute a high-risk group and could benefit from closer follow-up to facilitate early identification and intervention.
The Danish National Patient Registry (DNPR) and the Danish Cancer Registry (DCR) are central to registry-based cancer research. This systematic review evaluates studies assessing the quality of cancer-related data in the...The Danish National Patient Registry (DNPR) and the Danish Cancer Registry (DCR) are central to registry-based cancer research. This systematic review evaluates studies assessing the quality of cancer-related data in these registries under their current data structures. PubMed and Embase were systematically searched on January 24, 2025 (PROSPERO: CRD420251005952). Studies validating cancer-related data in the DNPR or DCR against a gold standard were included. Findings were synthesized narratively and categorized by DNPR data, DCR data, or multi-source algorithms. The literature search generated 915 records, of which 50 were included: 23 validated DNPR data, 9 DCR data, and 18 algorithm performance. The quality of DNPR cancer diagnoses and treatment showed positive predictive values (PPVs) of 57-100%, highest for common malignancies and treatments. The quality of DNPR comorbidities and complications varied substantially (PPVs 0-98%). The PPV of a melanoma diagnosis in the DCR was 97%. The DCR staging completeness varied considerably (34-95%). Algorithms presented PPVs of 60-96% for recurrence, active cancer, and recognized metastases, and 28% for unrecognized metastases. The DNPR and DCR provide high-quality data for many cancer diagnoses, treatments, and outcomes, supporting their use in register-based research. While some data elements, including data on complications, exhibit lower quality, algorithmic approaches can enhance utility for less robust data. However, several aspects of cancer-related data remain unvalidated.
Zhou C, Mooyaart AL, Hulscher N
… +9 more, Kerkour T, Ouwerkerk J, Louwman MWJ, Wakkee M, Li Y, Voorham QJM, Bruggink A, Nijsten TEC, Hollestein LM
Eur J Epidemiol
· 2026 Jan · PMID 41524883
·
Full text
There is a high need for accurate prognostic models among stage II melanoma to determine who may benefit from (neo)adjuvant systemic therapy. The Dutch Early- Stage Melanoma (D-ESMEL) study was designed to identify new p...There is a high need for accurate prognostic models among stage II melanoma to determine who may benefit from (neo)adjuvant systemic therapy. The Dutch Early- Stage Melanoma (D-ESMEL) study was designed to identify new prognostic features in a population-based sample of stage I/II melanoma patients in addition to American Joint Committee of Cancer (AJCC) staging. The validation cohort of the D-ESMEL study employs a nested case-control design. Initially, controls were randomly sampled to develop prognostic that included both known and new prognostic factors to assess the additive value of new prognostic factors. As a consequence, most controls had a very thin melanoma (<1.0 mm) while most cases had a thicker melanoma (>2.0 mm). This resulted in insufficient variability and high weights for stage II controls when applying weighted analyses in absolute risk prediction models. Therefore, randomly sampled controls were re-matched on AJCC stage (stage IA, IB, IIA, IIB, IIC), and new stage-matched controls were collected for cases who could not be rematched. The original D-ESMEL validation cohort included 5,815 stage I/II melanoma patients, of whom 154 developed distant metastasis (cases). 98/154 Cases were stage II and only 24 stage II controls were included, while the stage-matched design now includes 153 stage-matched case-control sets of which 97 stage II cases and 97 stage II controls derived from a population-based cohort of 5,785 stage I/II patients. The updated design increased the biological variability among stage II controls, balanced weights in weighted analyses and thereby facilitating reliable subgroup analyses in this clinically important subgroup.
Bychkov G, Engsner N, Bang B
… +6 more, Heyman MM, Barbany G, Nordenvall AS, Tettamanti G, Strannegård C, Nordgren A
Eur J Epidemiol
· 2026 Feb · PMID 41524882
·
Full text
Acute lymphoblastic leukemia (ALL) is the most common childhood malignancy. While space-time clustering of ALL cases has been suggested, only one prior study has examined clustering by genetic subtype. We investigated sp...Acute lymphoblastic leukemia (ALL) is the most common childhood malignancy. While space-time clustering of ALL cases has been suggested, only one prior study has examined clustering by genetic subtype. We investigated space-time clustering of childhood ALL in Sweden, both overall and by genetic subtype. The cohort included 1,629 children age 0-18 years diagnosed with ALL between 1992 and 2017, comprising 1,446 B-cell precursor ALL (BCP-ALL) and 183 T-cell ALL (T-ALL) cases. Two BCP-ALL subgroups were analyzed: high hyperdiploidy (HeH, n = 466) and ETV6::RUNX1 (n = 225). The Unbiased Knox Test and Unbiased Combined Knox Test were used to assess space-time clustering at the municipality level, accounting for multiple testing and population shifts. The Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm was applied to identify significant clusters. Logistic regression was used to evaluate demographic differences between clusters, including age, sex, and birth order. Significant space-time clustering was observed in the HeH subgroup for both place and date of birth (p = 0.005) and place and date of diagnosis (p = 0.011), at space-time thresholds of 40 km/18 months and 30 km/24 months, respectively. No clustering was detected in the overall BCP-ALL group, T-ALL group, or the ETV6::RUNX1 subgroup. Space-time clustering at birth and diagnosis was observed in the HeH subgroup, suggesting potential etiologic heterogeneity in BCP-ALL. These findings support further investigation of environmental and infectious exposures across immunophenotypes and genetic subtypes in larger cohorts.
Mendelian randomization has emerged as a powerful tool for exploring causal relationships in observational studies by using genetic variants as instrumental variables. While multivariable Mendelian randomization extends...Mendelian randomization has emerged as a powerful tool for exploring causal relationships in observational studies by using genetic variants as instrumental variables. While multivariable Mendelian randomization extends this approach to simultaneously address multiple exposures, it faces significant challenges with highly correlated exposures, particularly in high-dimensional settings such as multi-omics data. Conventional MVMR methods, which are primarily based on linear regression models, may suffer from multicollinearity and reduced statistical power when analyzing correlated exposures. The increasing availability of high-dimensional multi-omics data has highlighted the limitations of conventional MVMR approaches in analyzing correlated exposures while maintaining biological interpretability. To address these challenges, we propose integrating latent factor analysis into the MVMR framework, enabling dimension reduction without compromising interpretability. Through extensive simulation studies, we demonstrate that our method maintains a well-controlled false positive rate and offers superior sensitivity compared to conventional MVMR approaches. We apply our method to investigate the causal relationship between DNA methylation and mitochondrial DNA copy number. Our method offers a significant advantage in scenarios with highly correlated exposures driven by common latent factors or shared pathways, especially when individual effects are sparse. By applying our method to correlated multi-omics data, we can uncover new insights into the molecular mechanisms underlying complex phenotypes.
Walters EE, Luczak SE, Beam CR
… +15 more, Ericsson M, Kremen WS, Krueger RF, Markon KE, McGue M, Nygaard M, Panizzon MS, Plassman BL, Reynolds CA, Sachdev PS, Thalamuthu A, Whitfield KE, Pedersen NL, Gatz M, IGEMS Consortium
We reply to the letter to the editor by Soohyeon Ko (Eur J Epidemiol, https://doi.org/10.1007/s10654-025-01305-x ) concerning our article by Walters et al. (Eur J Epidemiol, https://doi.org/10.1007/s10654-025-01286-x )....We reply to the letter to the editor by Soohyeon Ko (Eur J Epidemiol, https://doi.org/10.1007/s10654-025-01305-x ) concerning our article by Walters et al. (Eur J Epidemiol, https://doi.org/10.1007/s10654-025-01286-x ). We reiterate that genetic explanations contribute to understanding why education is protective against dementia, alongside influences reflecting the whole of one's family and societal context. We also caution that genetic explanations should not be misinterpreted as deterministic.
Eur J Epidemiol
· 2025 Nov · PMID 41335397
·
Full text
Screening colonoscopy has been shown to reduce colorectal cancer incidence. However, the magnitude of this effect is debated. There is concern that some trial participants already had colorectal cancer at baseline. The s...Screening colonoscopy has been shown to reduce colorectal cancer incidence. However, the magnitude of this effect is debated. There is concern that some trial participants already had colorectal cancer at baseline. The screening procedure could not prevent disease occurrence in these participants, leading to "prevalence bias". Some authors have argued that the effect of interest is confined to participants without disease at baseline, and failing to exclude prevalent cases supposedly leads to effect underestimation. Yet, the issue is debated, with other authors arguing that conventional randomized trials provide the effects that are most relevant to public health. Here we present new, formal arguments that clarify misconceptions in this debate. We show that, under mild assumptions, the so-called "prevalence bias" is not a concern when researchers are interested in estimating risk differences, rather than risk ratios. This is because of a statistical property of the causal risk difference when outcomes are rare, called "doomed-selection stability".
Rietz Liljedahl E, Nielsen K, Engfeldt M
… +2 more, Saxne Jöud A, Nielsen C
Eur J Epidemiol
· 2025 Dec · PMID 41284183
·
Full text
The incidence of cutaneous melanoma (CM) has risen sharply over the past 30 years, coinciding with the rapidly growing tattoo trend. In Sweden, 20% of the population is tattooed. Repeated reports of the presence of carci...The incidence of cutaneous melanoma (CM) has risen sharply over the past 30 years, coinciding with the rapidly growing tattoo trend. In Sweden, 20% of the population is tattooed. Repeated reports of the presence of carcinogenic chemicals, such as polyaromatic hydrocarbons, aromatic amines and heavy metals in tattoo ink justifies the investigation of CM risk in relation to tattooing. We aimed to investigate the potential association between tattoo exposure and CM. We identified 2880 individuals who were diagnosed with CM at age 20-60 years, in the Swedish National Cancer Register. For each case, we sampled three random age- and sex-matched controls from the Swedish Total Population Register. Exposure data and data on potential confounders were collected through a questionnaire in 2021. We estimated the relative risk of CM in tattooed compared with nontattooed individuals using multivariable logistic regression, rendering incidence rate ratios (IRR). Of the participants, 22% of the cases (354/1598) had a tattoo before the index date, vs. 20% of the controls (815/4097). We observed an adjusted relative risk of CM in tattooed compared to nontattooed individuals of 1.29 (95% confidence interval [CI]: 1.07-1.56). The results suggested that tattoos may be a risk factor for CM, but further studies are needed to establish causality.
Little MP, Mai JZ, Barnholtz-Sloan JS
… +8 more, Linet MS, Fang M, Chernyavskiy P, Kennerley V, Cahoon EK, Cockburn MG, Kendall GM, Kimlin MG
Eur J Epidemiol
· 2026 Mar · PMID 41284182
·
Full text
Brain tumour is the second most common type of childhood cancer and the most common solid tumour in children, but its aetiology is largely unknown. Some previous studies have suggested that elevated ultraviolet radiation...Brain tumour is the second most common type of childhood cancer and the most common solid tumour in children, but its aetiology is largely unknown. Some previous studies have suggested that elevated ultraviolet radiation (UVR) exposures decrease brain tumour risk, but the evidence is inconsistent. We conducted a cross-sectional study (with census-derived population counts) to assess age < 20 malignant/non-malignant brain tumour incidence overall and for major categories in Surveillance, Epidemiology and End Results 2000-2021 data, using ground-based UVR-irradiance measures, via quasi-likelihood models accounting for over/under-dispersion, adjusted for age, sex, race/ethnicity and other socioeconomic variables. There were 29,088/18,585 cases of malignant/non-malignant brain tumour, with generally significant decreasing trends of both types of tumour with UVR irradiance [relative risk (RR) = 0.754/mW/cm (95% CI 0.659, 0.862, p < 0.0001) for malignant brain tumour, RR = 0.466/mW/cm (95% CI 0.382, 0.567, p < 0.0001] for non-malignant brain tumour), although there was significant heterogeneity by histopathologic subtype, race/ethnicity, and sex. Equally, there is a highly significant decreasing trend of both types of tumour with UVR-cumulative radiant exposure (p < 0.0001). These trends are also significant in many malignant/non-malignant brain tumour histopathological subtypes and racial/ethnic groups. However, there are certain non-malignant brain tumour subtypes, for example tumours of the pineal region and meningeal tumours, where RR significantly exceed 1 in relation to UVR irradiance (p = 0.0330, p = 0.0024 respectively). Our finding, of a generally protective effect of UVR on brain tumour risk is not clear-cut, and warrants large studies of specific histopathological pediatric/adolescent brain tumours using individual-level data on solar exposures and key effect modifiers and potential confounders.