Associations between Body Composition Patterns, Brain Diseases and Injuries, and Risk of Epilepsy: A Prospective Cohort Study

Abstract

Background

Different body compositions may exert varying effects on epilepsy. We aimed to prospectively examine the associations between body composition and epilepsy risk while exploring the potential mediation of brain-related injuries.

Methods

We constructed a large-scale cohort study within the UK Biobank (UKB), deriving 7 body composition patterns via principal component analysis that captured variation in muscle strength, bone density, lean mass (LM), and fat distribution. Multivariable Cox proportional hazards models were employed to assess associations between these patterns, individual body measurements, and epilepsy risk. We performed stratified analyses by polygenic risk score (PRS) and mediation analyses to evaluate the indirect contributions of brain injuries.

Results

Among the 475,960 participants, 3,026 epilepsy cases were identified over an average follow-up of 10.9 years. Patterns of “LM,” “muscle strength,” “bone density,” and “leg-dominant fat distribution” were associated with reduced risk of epilepsy (hazard ratios [HRs]: 0.68–0.97), whereas “fat-to-LM,” “central obesity,” and “arm-dominant fat distribution” patterns were associated with increased incidence (HRs: 1.01–1.34). Similar trends were noted for corresponding individual body measurements. These associations were consistent across PRS strata. Falls, stroke, and neurodegenerative diseases mediated 17.9%, 27.2%, and 31.0% of effects for “muscle strength,” “bone density,” and “arm-dominant fat distribution,” respectively.

Conclusions

Body composition patterns involving muscle strength, bone density, and fat distribution show robust associations with epilepsy, partly mediated by neurological disorders. Optimizing body composition and preventing neurological insults may help reduce epilepsy risk in middle-aged and older adults.

Introduction

Epilepsy is a noncommunicable neurological disorder that affects over 50 million people worldwide [1], with its incidence mounting steadily beyond age 50 and peaking in infants and the elderly [2]. Genetic abnormalities have been historically acknowledged as a culprit of generalized epilepsy (GE) [3], while other common risk factors vary by age group. Cerebrovascular and neurodegenerative diseases are prevalent risk factors among older adults, whereas epilepsy-related traumatic brain injuries, infections, and tumors might develop at any age [2, 4]. Still, approximately 50% of cases report unknown causes, probably resulting from complex interactions among genetic predispositions, environmental, and behavioral factors [4].

Cross-sectional studies have revealed that patients with epilepsy tend to be more sedentary, higher in fat [5] and at increased risk for sarcopenia [6], osteoporosis, and fractures [7, 8]. However, it is unclear whether these features serve as underlying risk factors preceding diagnosis or emerge as adverse effects of antiepileptic drug use and lifestyle shifts. Notably, limited studies exploring the association between body mass index (BMI) and epilepsy have yielded inconsistent findings [9–11]. Although BMI is easy to measure and widely used, it neither distinguishes between various body components, such as fat, lean mass (LM) and skeletal structure, nor captures the fat accumulation in different depots, which may drive diverse health outcomes [12].

Leveraging comprehensive body measurements data from the UK Biobank (UKB), we sought to clarify the associations between predefined patterns of body composition representing muscle strength, bone density, LM, and fat mass (FM) storage in different depots, specific body measurements, and risk of epilepsy. Additionally, we examined the modification effect of common variants quantified by polygenetic risk score (PRS), and whether relevant brain disorders and accidental injuries, separately or as a group, play a potential mediating role.

Methods

Study Design and Population

The present study was derived from the UKB, a large-scale survey that enrolled over 500,000 residents aged 40–69 years across England, Scotland, and Wales. Individual information on sociodemographic, lifestyle, body metrics, and genetic factors was collected through touchscreen questionnaires, physical measurements, and biological samples at baseline. The UKB study was approved by the North West Multicenter Research Ethical Committee and all participants provided informed consent. For our cohort, we excluded individuals who withdrew consent (n = 1,298), had a documented or self-reported history of epilepsy before recruitment (n = 5,256), and those with missing or extreme body measurements (i.e., exceeding ±5 standard deviations) or key covariates (n = 19,897), resulting in a study population of 475,960 participants (online suppl. Fig. S1; for all online suppl. material, see https://doi.org/10.1159/000551068).

Body Composition

Body composition was assessed by trained staff following a standard protocol. Height, weight, waist, and hip circumference were obtained manually. Regional (i.e., arm, leg, and trunk) and total FM and LM were evaluated through the Tanita BC-418MA body composition analyzer. Grip strength was measured using the Jamar J00105 Hydraulic Hand Dynamometer. Bone density was measured at the calcaneus using the Norland McCue Contact Ultrasound Bone Analyzer.

These measurements and their derivations, a total of 28 items, were analyzed using principal component analysis (PCA), as described in previous studies [13, 14]. Seven principal components with eigenvalues >1.0 were retained (Figure 1; online suppl. Fig. S2), accounting for over 90% of the total variance (online suppl. Table S3). In light of the substantial similarity between these patterns in males and females, we applied male-derived loading coefficients to ensure consistency, comparability, and interpretability across sexes, particularly for downstream analyses involving sex-stratified and pooled comparisons [13, 14]. The resulting components were interpreted and labeled based on dominant loading variables: “fat-to-LM” (indicating FM relative to LM), “LM,” “central obesity,” “muscle strength,” “bone density,” “arm-dominant fat distribution” (highlighting FM primarily distributed in the arms), and “leg-dominant fat distribution” (underscoring FM mainly storage in the legs).

Fig. 1.

Sex-specific principal component loadings of body composition patterns derived from 28 body composition measures. Heatmaps display rotated factor loadings (absolute value >0.4) from sex-specific principal component analyses (PCA) conducted separately in males (a) and females (b). Seven PCs, shown on the y-axis, were retained in each sex to capture distinct body composition patterns derived from 28 body measurements shown on the x-axis. Color intensity reflects the magnitude and direction of factor loadings, with blue indicating negative loadings and red indicating positive loadings. The strong concordance in loading structures across sexes supports the application of male-derived loading coefficients to both sexes in downstream analyses. PC, principal component; WC, waist circumference; HC, hip circumference; WTH, waist-to-hip ratio; WTHR, waist-to-height ratio; WWI, waist-to-weight ratio; BSI, body shape index; FM, fat mass; LM, lean mass; Grip, grip strength; BMI, body mass index; BMD, bone mineral density.

Ascertainment of Epilepsy

Epilepsy cases were identified based on self-reported conditions during subsequent visits, linked data from primary care records, hospital admissions, and death registry records, according to the corresponding International Classification of Diseases codes (ICD) (online suppl. Table S4). The follow-up duration was calculated from the recruitment date to the date of first diagnosis, death, or December 31, 2019, whichever came first.

PRS Construction

Details of genotyping and imputation for UKB samples have been described elsewhere [15]. We curated 20 single-nucleotide polymorphisms (SNPs) with minor allele frequency >0.01 and p < 5 × 10⁻⁸ from the latest multi-ancestry genome-wide association study (GWAS) by the International League Against Epilepsy Consortium (ILAE) [16] (online suppl. Table S5). We extracted published effect estimates (β_i) of each variant and recoded the SNP_i as additive risk allele counts (0, 1, 2). The PRS was calculated as the weighted sum of risk alleles using the following formula, where n denotes the number of selected SNPs:

$$\text{PRS}={\displaystyle \sum }_i^n{\beta }_i\times {\text{SNP}}_i$$

Covariates and Mediators

The covariates included age at recruitment, sex, ethnicity, educational level (college/university degree or others), Townsend deprivation index, smoking status (never, former, or current smoker), average weekly alcohol intake (none, low-risk, or high-risk drinking), physical activity (low, moderate, or high level), and comorbidities (i.e., hypertension, diabetes, disorders of lipoprotein metabolism, and other lipidemia) (online suppl. Table S6). In this study, we considered the occurrence of brain-related diseases and injuries after enrollment and within 5 years before epilepsy diagnosis as potential mediators of the studied associations. This was based on the premise that most cases of secondary epilepsy typically emerge within a few years after the primary condition [17–19]. The potential mediators included (1) stroke; (2) neurodegenerative diseases, including Alzheimer’s disease, Parkinson’s disease, and Huntington’s disease; (3) inflammatory encephalopathies, including meningitis, encephalitis, myelitis, and encephalomyelitis; (4) hospitalizations attributable to falls; and (5) hospitalizations attributable to traffic accidents. In addition, a multiple mediator model was constructed to estimate (6) the combined mediating effects of these five conditions.

Statistical Analyses

Baseline demographic characteristics were presented according to the ultimate epilepsy diagnosis. We conducted sex-stratified analyses to examine the associations between identified body composition patterns, individual components, and epilepsy risk with Cox proportional hazard models. The proportional hazards assumption was assessed using Schoenfeld residuals, and no violations were observed. Exposures were initially modeled as continuous variables using restricted cubic splines with knots at the 5th, 35th, 65th, and 95th percentiles to flexibly characterize the shape of associations. Nonlinearity was assessed via likelihood ratio tests comparing models with linear terms only versus models including both linear and cubic terms. In light of the observed nonlinearity in most associations, we further categorized exposures into sex-specific tertiles for phenotypic associations. Cox models with attained age as the time scale were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for epilepsy incidence per 1 standard deviation increase in continuous variables and by tertile for categorical variables (referenced to the lowest tertile). All models were adjusted for sex, ethnicity, age at recruitment, Townsend deprivation index, smoking status, average weekly alcohol intake, physical activity, and comorbidities. To disentangle whether associations varied by genetic susceptibility to epilepsy, we constructed PRS based on common variants and performed stratified analyses by PRS tertiles. We employed single mediator models with and without exposure-mediator interaction terms to evaluate the separate intermediary role and introduced a multi-mediator model to assess the combined contributions with the R package “CMAverse” [20]. Indirect effects were defined on the hazard ratio scale as the contrast between the counterfactual outcome under the exposed versus unexposed mediator distribution, while holding the exposure fixed. Parametric bootstrapping (n = 400 times) was used to calculate 95% CIs and p values.

We excluded participants who self-reported taking antiepileptic drugs and applied a 2-year lag time to minimize reverse causality. We also repeated the main analyses with patterns determined based on sex-specific loading coefficients. Furthermore, we investigated the associations between body composition patterns and two main epilepsy subtypes, GE and focal epilepsy (FE), indexed by ICD codes. To address the potential influence of ethnicity on body composition and genetics, we limited our analyses to solely European descent. Lastly, Fine-Grey models accounting for the competing risk of all-cause mortality were constructed. All analyses were performed using R (version 4.3.2, R Foundation for Statistical Computing, Vienna, Austria), with two-sided p values <0.05 considered statistically significant.

Results

Baseline Characteristics

Among the 475,960 participants included, 3,026 incident cases of epilepsy were identified during an average follow-up of 10.9 years. Compared with non-cases, those who developed epilepsy were more likely to be male, older, less educated, exhibit adverse health behaviors, and present a higher prevalence of common metabolic comorbidities (online suppl. Table S7).

Phenotypic Associations of Body Composition and Epilepsy

Restricted cubic spline models revealed an L-shaped association between the “bone density” pattern and epilepsy risk and largely linear associations for “LM,” “central obesity,” and “arm-dominant fat distribution” patterns across sexes (Fig. 2). Alternatively, associations for “fat-to-LM,” “muscle strength,” and “leg-dominant fat distribution” patterns differed between men and women.

Fig. 2.

a–g Multivariable adjusted hazard ratios for epilepsy risk according to identified body composition patterns on a continuous scale, with separate analyses for females (red) and males (blue). The solid lines represent the multivariate adjusted HRs, and the shaded areas indicate the 95% CIs. Reference lines for no association are marked by the dashed lines at a hazard ratio of 1.0. Restricted cubic splines were constructed with 4 knots at the 5th, 35th, 65th, and 95th percentiles for each pattern. Analyses were adjusted for sex, ethnicity, age at recruitment, Townsend deprivation index, smoking status, average weekly alcohol intake, physical activity and comorbidities (denoted as “P overall”). We tested the potential nonlinearity by using the likelihood ratio test comparing models with linear terms only versus models including both linear and cubic terms (denoted as “P nonlinearity”). HR, hazard ratio; CI, confidence interval; PC, principal component.

In survival analyses, the “muscle strength,” “bone density,” and “arm-dominant fat distribution” patterns showed robust associations with epilepsy risk, regardless of whether they were modeled as continuous (Fig. 3; online suppl. Table S8) or categorical variables (Fig. 4), with highly comparable estimates across sexes. Compared with the first tertiles, the higher level of “LM” (HR for high level: 0.91 [0.83, 1.00]; HR for continuous: 0.97 [0.95, 0.99]), “muscle strength” (HR for moderate level: 0.77 [0.70, 0.83]; HR for high level: 0.68 [0.62, 0.75]; HR for continuous: 0.89 [0.87, 0.91]), “bone density” (HR for moderate level: 0.87 [0.79, 0.94]; HR for high level: 0.87 [0.79, 0.95]; HR for continuous: 0.96 [0.93, 0.99]), and “leg-dominant fat distribution” (HR for moderate level: 0.90 [0.82, 0.99]) patterns were associated with lower incidence of epilepsy, whereas “fat-to-LM” (HR for continuous: 1.01 [1.00, 1.02]), “central obesity” (HR for high level: 1.13 [1.04, 1.24]; HR for continuous: 1.04 [1.02, 1.07]), and “arm-dominant fat distribution” (HR for high level: 1.34 [1.22, 1.47]; HR for continuous: 1.12 [1.09, 1.16]) patterns were associated with increased risk. These associations were directionally consistent across sexes, except for the “central obesity” pattern, which exhibited a notable inverse correlation in males at a moderate level (HR: 0.88 [0.77, 1.00]) versus a positive association in females at a high level (HR: 1.25 [1.10, 1.43]).

Fig. 3.

Hazard ratios for incident epilepsy according to identified patterns, individual components of body composition on a continuous scale in multivariable Cox proportional hazards models. The associations between identified patterns, individual components of body composition, and epilepsy were evaluated using Cox proportional hazards models with attained age as the time scale after adjusting for sex, ethnicity, age at recruitment, Townsend deprivation index, smoking status, average weekly alcohol intake, physical activity, and comorbidities. The spots represent the multivariate adjusted hazard ratios, and the lines indicate the 95% confidence intervals. *Significant at p < 0.05. PC, principal component; WC, waist circumference; HC, hip circumference; WTH, waist-to-hip ratio; WTHR, waist-to-height ratio; WWI, waist-to-weight ratio; BSI, body shape index; FM, fat mass; LM, lean mass; Grip, grip strength; BMI, body mass index; BMD, bone mineral density.

Fig. 4.

Hazard ratios for incident epilepsy according to body composition patterns as categorical variables in multivariable Cox proportional hazards models. IRs were calculated as the number of cases per 100,000 person-years. HRs and 95% CIs were derived from Cox proportional hazards models with attained age as the time scale after adjusting for sex, ethnicity, age at recruitment, Townsend deprivation index, smoking status, average weekly alcohol intake, physical activity, and comorbidities. IR, incidence rate; HR, hazard ratio; CI, confidence interval; PC, principal component.

As for individual components of body composition with risk of epilepsy, partial similarity to the findings of the above patterns was noted (online suppl. Table S9). Specifically, the leading central obesity indicators were positively correlated with epilepsy onset, while measures of muscle strength and bone density presented inverse associations. Particularly, higher “arm FM/LM” (HR for continuous: 1.06 [1.03, 1.10]) and “arm/whole FM” (HR for high level: 1.14 [1.05, 1.25]; HR for continuous: 1.04 [1.01, 1.08]) were associated with elevated epilepsy risk, collaborating the finding of “arm-dominant fat distribution” pattern.

Modification Role of PRS

Epilepsy incidence was found to increase with higher PRS strata across all exposure categories (Fig. 5). Associations between body composition patterns and epilepsy were reasonably comparable across PRS tertiles, indicating a limited modification effect.

Fig. 5.

Hazard ratios for incident epilepsy according to body composition patterns as categorical variables across polygenetic risk score tertiles. IRs were calculated as the number of cases per 100,000 person-years. HRs and 95% CIs were derived from Cox proportional hazards models with attained age as the time scale after adjusting for sex, ethnicity, age at recruitment, Townsend deprivation index, smoking status, average weekly alcohol intake, physical activity, and comorbidities. IR, incidence rate; HR, hazard ratio; CI, confidence interval; PC, principal component.

Mediation Analyses of Brain Diseases and Injuries

Associations of “bone density,” “muscle strength,” and “arm-dominant fat distribution” with epilepsy were partially mediated by brain disorders and injuries (Fig. 6; online suppl. Table S10). The effect of “bone density” was exclusively mediated by falls, with the mediating effect accounting for 23.5%, which increased to 27.2% when incorporating the interactions of all mediators. “Muscle strength” and “arm-dominant fat distribution” patterns were found to be mediated through multiple conditions encompassing falls, stroke, and neurodegenerative diseases, among which stroke took a predominant role (proportion mediated: 9.1% for “muscle strength”; 31.6% for “arm-dominant fat distribution”), contributing to combined effects of 6.7% and 31.4%, respectively.

Fig. 6.

Mediating effect of brain diseases and injuries on the associations between body composition patterns and epilepsy. The results that reached statistical significance (i.e., p < 0.05 using the false discovery rate for adjustment of multiple testing) are marked with color, with blue denoting inverse associations and red denoting positive associations. PC, principal component; ns, not statistically significant.

Sensitive Analyses

These observed associations generally persisted after excluding potentially undiagnosed patients (online suppl. Table S11), for patterns extracted with sex-specific loading coefficients (online suppl. Table S12), among the white only (online suppl. Table S13) and when adjusting for the competing risks of all-cause mortality (online suppl. Table S14), but were markedly attenuated in specific epilepsy subtypes, probably attributable to reduced cases (GE: 874; FE: 868 vs. all epilepsy: 3,026) and wider CIs (online suppl. Table S15–16).

Discussion

Utilizing a prospective cohort of 475,960 participants with an average follow-up of 10.9 years, we investigated the associations between 7 patterns of body composition extracted by PCA, 28 individual body components, and the risk of epilepsy. Analyses of the identified patterns, modeled as either continuous or categorical variables, consistently showed that “muscle strength” and “bone density” were associated with reduced risk of epilepsy, while “arm-dominant fat distribution” patterns were associated with increased incidence, partly supported by the findings of individual measurements. These associations persisted across strata of genetic susceptibility and were significantly mediated by brain dysfunction and traumatic injuries, particularly falls, stroke, and neurodegenerative diseases.

Our findings largely aligned with previous studies delving into the relationship between body composition and epilepsy, while most of them lacked prospective designs and solely concentrated on separate body measurements, disregarding the interactions within various components [10, 21–24]. The prospective design of our study enables temporal inference, providing novel insights that low LM, reduced grip strength, and decreased bone density may contribute to epileptogenesis or reflect early premorbid changes, rather than merely serving as comorbidities. Furthermore, our study expanded existing knowledge by demonstrating that these associations were independent of genetic susceptibility of common variants.

The health implications of adiposity are heterogeneous and depend strongly on fat distribution, with central or abdominal adiposity conferring markedly different risks compared with peripheral or subcutaneous fat depots [25, 26]. Central obesity is an established risk factor of metabolic syndrome [27] and cardiovascular diseases [28], whereas gluteal-femoral fat has been suggested to confer metabolically protective effects. A previous Mendelian randomization study reported a positive association between hip circumference, waist-to-hip ratio, and juvenile myoclonic epilepsy, which aligns with our observation that the “central obesity” pattern was associated with increased epilepsy risk [29]. We further identified gender differences in this association, characterized by an inverse relationship at moderate levels in males versus a positive association at high levels in females, indicating underlying sex-specific susceptibility that merits further investigation.

Building on this distinction between central and peripheral adiposity, we further demonstrated that peripheral fat distribution itself is heterogeneous with respect to epilepsy risk. In our study, the “leg-dominant fat distribution” pattern was associated with a lower risk of epilepsy, probably benefiting from lowering lipid overflow and ectopic fat and guarding against insulin resistance and systemic inflammation [30, 31]. However, the “arm-dominant fat distribution” pattern, collaborating with measurements “arm FM/LM” and “arm/whole FM,” was associated with a higher risk of epilepsy. The pattern, where fat is stored in the arms and lean tissue is distributed in the hips and legs, tends to exhibit a higher abdomen fat ratio and increased muscle fat infiltration and thus was considered a passive loading effect of excessive adiposity [13].

The underlying mechanisms of body composition patterns with epilepsy have not been clarified, but our study suggests that brain insults play an important mediating role. It is well established that stroke and traumatic brain injuries are common causes of acquired epilepsy, with post-stroke epilepsy accounting for approximately 50% of cases in the elderly [32] and post-traumatic epilepsy contributing to 20% of the cases in the general population [33]. People with neurodegenerative diseases also reported a 7.5-fold increased risk of developing epilepsy later in life [34]. It is plausible that body composition affects epilepsy risk by altering susceptibility to these intermediary neurological conditions, either through increasing the likelihood of secondary seizures or by contributing to cumulative structural and functional brain damage over time.

Beyond their role in preventing falls and related injuries, muscle strength and bone density may also influence neurological health through direct musculoskeletal-brain signaling pathways. Grip strength is a robust marker of overall neuromuscular integrity and has been consistently associated with lower risks of adverse aging-related outcomes [35], including stroke [36, 37] and dementia [38]. Experimental and epidemiological evidence suggests that skeletal muscle functions as an endocrine organ, releasing myokines such as irisin and brain-derived neurotrophic factor, which support neuroplasticity, synaptic maintenance, and neuroprotection [39, 40]. In parallel, emerging evidence supports the existence of a bone-brain axis, whereby osteokines such as osteocalcin exert regulatory effects on cognitive function, stress response, and neuronal signaling [41, 42]. Therefore, reduced muscle strength or bone density may reflect broader dysregulation of musculoskeletal-neural interactions, increasing vulnerability to neurological injury and epileptogenesis.

The “arm-dominant fat distribution” pattern identified in our study likely represents an adverse metabolic and inflammatory phenotype. This pattern, characterized by increased abdominal adiposity and muscle fat infiltration, has been linked to elevated secretion of adipocytokines and pro-inflammatory cytokines [30]. These processes promote insulin resistance, systemic and neuroinflammation, and the accumulation of amyloid-β pathology [43], increasing the risk of cerebrovascular [30] and neurodegenerative diseases [44]. Such processes are also recognized contributors to neuronal hyperexcitability and seizure susceptibility.

This study has several notable strengths. First, we leveraged a large, nationally representative cohort with robust longitudinal health data, facilitating a prospective investigation of these associations, which substantially reduces the likelihood of reverse causality and strengthens the identification of mediators. Additionally, through PCA, we generated 7 patterns representing muscle strength, bone density, LM, and fat accumulated in different depots. This approach moves beyond traditional metrics such as BMI and enables a more nuanced investigation of the relationship between various body components, interactions among them, and epilepsy risk, with direct implications for public health. However, FM and LM were measured at baseline via bioelectrical impedance, which may introduce greater variability compared with the quantitative MRI and dual-energy X-ray absorptiometry. We were also unable to cross-validate the associations with body measurements from diverse sources. Moreover, body composition was assessed only at baseline, and we were therefore unable to account for longitudinal changes during follow-up. Despite conducting subtype-specific sensitivity analyses for GE and FE indexed by ICD codes, we lacked sufficient clinical detail to distinguish spontaneous and provoked seizures or to classify finer subtypes. Due to the transient nature of seizures, which can sometimes present with subtle or focal manifestations, some cases may have been missed, leading to incomplete case exclusion. To address this, we conducted lagged analyses excluding cases diagnosed within the first 2 years of follow-up, and the associations remained robust. Finally, this observational study identifies associations rather than causal relationships, despite extensive adjustment and sensitivity analyses.

Conclusions

In this large prospective cohort, higher levels of body composition patterns characterized by “muscle strength” and “bone density” were associated with reduced risk of epilepsy, while the “arm-dominant fat distribution” pattern was linked to elevated risk. These associations were independent of genetic susceptibility and were partially mediated by falls, stroke, and neurodegenerative diseases. Our findings highlight the importance of optimizing body composition and implementing early interventions targeting brain injuries to reduce epilepsy risk in middle-aged and older adults.

Acknowledgments

This research was conducted using the UK Biobank Resource (Application No. 78559). The authors thank all participants and staff for their cooperation and assistance in the study.

Statement of Ethics

This study was derived from the UKB study (Application No. 78559), which was ethically approved by the North West Multicenter Research Ethics Committee (REC reference: 21/NW/0157). Informed written consent was obtained from all participants in the UKB study.

Conflict of Interest Statement

The authors have no relevant financial or nonfinancial interests to disclose.

Funding Sources

This research was supported by High-Performance Computing Public Platform (Shenzhen Campus) of Sun Yat-Sen University. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Author Contributions

Q.Y. and Y.Q. conceptualized and designed the study. Q.Y. performed the statistical analyses. Q.Y., L.Q., X.Y., X.G., and Q.Q. extracted the data and performed the validation. Q.Y. drafted, and R.Q., D.C., Y.Z., X.W., and Y.Q. revised the manuscript. X.W., Y.Z., and Y.Q. supervised the data analysis and interpretation. All authors provided feedback and approved the final version of the manuscript submitted for publication.

Funding Statement

This research was supported by High-Performance Computing Public Platform (Shenzhen Campus) of Sun Yat-Sen University. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data Availability Statement

The data used in the study can be accessed through the application and approval from the UK Biobank.