Identifying modifiable factors and their joint effect on frailty: a large population-based prospective cohort study

Abstract

A thorough understanding and identification of potential determinants leading to frailty are imperative for the development of targeted interventions aimed at its prevention or mitigation. We investigated the potential determinants of frailty in a cohort of 469,301 UK Biobank participants. The evaluation of frailty was performed using the Fried index, which encompasses measurements of handgrip strength, gait speed, levels of physical activity, unintentional weight loss, and self-reported exhaustion. EWAS including 276 factors were first conducted. Factors associated with frailty in EWAS were further combined to generate composite scores for different domains, and joint associations with frailty were evaluated in a multivariate logistic model. The potential impact on frailty when eliminating unfavorable profiles of risk domains was evaluated by PAFs. A total of 21,020 (4.4%) participants were considered frailty, 192,183 (41.0%) pre-frailty, and 256,098 (54.6%) robust. The largest EWAS identified 90 modifiable factors for frailty across ten domains, each of which independently increased the risk of frailty. Among these factors, 67 have the potential to negatively impact health, while 23 have been found to have a protective effect. When shifting all unfavorable profiles to intermediate and favorable ones, overall adjusted PAF for potentially modifiable frailty risk factors was 85.9%, which increases to 86.6% if all factors are transformed into favorable tertiles. Health and medical history, psychosocial factors, and physical activity were the most significant contributors, accounting for 11.9%, 10.4%, and 10.1% respectively. This study offers valuable insights for developing population-level strategies aimed at preventing frailty.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11357-024-01395-7.

Introduction

Frailty, as internationally and conceptually defined, is characterized by being a multidimensional syndrome involving diminished physiological reserves and reduced resistance to stressors [1]. Frailty is a prevalent condition characterized by physical decline, slow walking speed, low body weight, and deficits in endurance and resistance, increasing an individual’s vulnerability to stressors and impairing their ability to manage daily demands [2]. Findings from observational studies suggested that frailty has several health effects, including an increased risk of falls [3], disability [4], dementia [5], and hospitalization [6], emphasizing its importance in clinical and public health. Gaining a deeper understanding of the impact of age-related syndromes, particularly frailty, on health outcomes and identifying interventions that can alter the trajectory of aging are critical efforts to effectively address the challenges posed by an aging population [7].

Understanding the determinants of risk can serve as a guiding principle for designing effective public health initiatives and preventative measures, particularly when these factors are amendable through targeted interventions. Non-genetic risk factors such as obesity, loneliness, and hormonal deficits have been revealed through hypothesis-driven observational studies [2]. Nonetheless, while probing individual risk factors is crucial, it comes with inherent limitations. The intricate interplay among these factors introduces the potential for an increased likelihood of type I errors, underscoring the necessity for a more integrated and multivariate analytical approach in elucidating the etiology of frailty [8]. Furthermore, due to the multifactorial nature of frailty, focusing on one or a few risk factors at a time may not fully reflect the synergistic effects of exposures. Isolated examinations of single exposures may not provide a comprehensive understanding of the collective impact of risk factors on frailty. To address these inherent limitations, the implementation of an exposure-wide association study (EWAS) has emerged as a comprehensive and robust approach in unraveling risk factors associated with complex phenotypes. Unlike traditional single-exposure analyses, EWAS considers a broader spectrum of potential exposures, allowing for a more encompassing evaluation of their collective influence on the phenotype of interest. This method has been utilized to methodically evaluate diabetes [9], depression [8], and obesity [10], validating previous findings and identifying new risk factors. The EWAS approach allows for comparing the effect sizes of different risk factors in the same population and assessing their combined effect on frailty. Analyzing data from over 7000 phenotypes, we conducted the largest frailty EWAS to date, enabling a comprehensive assessment of the association between exposures and frailty.

The development of frailty is influenced by a mix of genetic and environmental factors. However, the role of genetic predisposition in modifying the relationship between modifiable factors and frailty risk remains largely unexplored. Like many common aging syndromes and diseases, frailty risk is polygenic. Research suggests that frailty has a genetic basis, with heritability rates ranging from 6 to 45% [11, 12]. The use of polygenic risk scores (PRS) shows promise in refining precision frailty prevention. As PRS accuracy improves, they could serve as valuable clinical tools for risk assessment in targeted screening and chemoprevention [13]. PRS aids in identifying high-risk individuals, optimizing interventions, and allocating resources where they are most needed. This not only improves the efficiency of frailty prevention but also enables more precise and effective interventions, ultimately leading to better health outcomes in at-risk populations.

Our study aimed to comprehensively explore the modifiable risk factors of frailty using UK Biobank (UKB) data. We conducted an extensive EWAS analysis, revealing potential unmodifiable and modifiable factors linked to frailty. We generated multidomain risk factor composite scores and quantified the population-attributable fractions (PAFs) for each domain and overall frailty to evaluate the potential impact of preventative measures. Additionally, we categorized the population into low, moderate, and high PRS groups based on the frailty phenotype’s PRS. This was done to investigate modifiable risk factors among different PRS classification groups, facilitating targeted prevention strategies.

Methods

Study population

The individuals included in our study were sourced from the UK Biobank, a vast population-based cohort established between 2006 and 2010 [14]. This assessment encompassed demographic details, physical metrics, a touchscreen questionnaire, genomic information, and the collection of biological specimens. All participants provided written agreement, and ethical approval was conferred by the National Research Ethics Service Committee North West Multi Centre Haydock (committee reference: 11/NW/0382). Analyses were carried out under UK Biobank application No. 19542.

Frailty assessment

We selected the frailty phenotype as the principal tool for evaluating physical frailty, deliberately choosing it over the frailty index or frailty scale due to its superior clinical practicality and widespread utilization as a prominent epidemiological metric [15]. Fried phenotype takes into account five self-reported or objectively measured components, including weight loss, exhaustion, low grip strength, physical inactivity, and slow walking pace [16]. Using validated phenotypic version definitions for UK biobank, we adjusted the criteria to be estimated based on available data [17]. Detailed explanations of these definitions can be found in Table S1. Participants had complete data for all five components which were categorized as robust (not meeting any of the Fried criteria), pre-frailty (meeting 1–2 criteria), or frailty (meeting ≥ 3 criteria) [16].

Modifiable factors

We performed a comprehensive pre-processing and quality control procedure to prepare the dataset for the systematic analyses of modifiable factors of frailty. After quality control and preprocessing (Supplementary Methods), data of remaining 276 modifiable factors measured or derived at baseline were gathered. Figure 1 provides an overview of the variable screening pipeline. The detailed coding and procession of the 276 factors are shown in Table S3. All variables were divided into 11 categories: biochemical measurements (e.g., urate); circadian (e.g., snoring); dietary (e.g., cereal intake); environment (e.g., air pollution); family factors (e.g., adopted as a child); health and medical history (e.g., myocardial infarction); lifestyle (e.g., time spent watching television); physical activity (e.g., summed days activity); physical measures (e.g., waist circumference); psychosocial factors (e.g., loneliness, isolation); sociodemographics (e.g., household income). All modifiable variables were listed in Table S4.

Fig. 1.

Overview of modifiable risk factors analytic design. Analytical procedure to identify modifiable risk factors associated with frailty in the UK Biobank. Abbreviations: PAF, population attributable fraction

Covariates

During the initial assessment, demographic information such as age, gender, and assessment center (categorized into 22 categories) were obtained collected through a self-completed touchscreen questionnaire. The Townsend Deprivation Index, which considers factors such as unemployment, ownership of vehicles, household overcrowding, and occupation, was derived from national census data. Higher score reflects more socioeconomic deprivation. Information on education (whether the participant had a college or university degree), ethnicity (categorized as white or non-white, including Mixed, Asian, Black, Chinese, and Other), smoking status (categorized as never, former, or current), and alcohol use (categorized as never, former, or current) was collected through a touchscreen questionnaire.

Polygenic risk score

The UK Biobank project outlines procedures for genetic data processing (http://www.ukbiobank.ac.uk/scientists-3/genetic-data/). Given limited studies beyond UK Biobank conducting frailty GWAS, we independently analyzed both discovery and validation cohorts, identifying various genetic risk variants. PRSice software (www.PRSice.info) was used to calculate PRS in the discovery cohort using validation cohort GWAS results. Genetic investigations focused on Caucasian participants (British, Irish, and other Europeans). SNPs with call rates < 95%, minor allele frequencies < 0.1%, and P < 1E-10 for Hardy–Weinberg equilibrium were excluded. The study employed P-informed aggregation with R2 = 0.01 in a 250-kb window. Higher PRS scores indicated elevated genetic frailty predisposition based on all relevant SNPs. Participants were categorized by frailty PRS into three groups: low (lowest quintile), moderate (quintiles 2–4), and high (highest quintile).

Statistical analyses

Following the lead of prior EWAS, we conducted an exhaustive examination of the links between 276 modifiable risk factors and frailty through an EWAS, utilizing logistic regression models in the discovery and replication data sets. Given that frailty is a dynamic and potentially reversible condition [18], we validated our findings using a longitudinal dataset, consisting of individuals with available frailty phenotype data in instance 2 (n = 26,683). Participants were classified as robust, pre-frail, or frail only if they were diagnosed as such at both baseline and instance 2 assessment. Variables significantly associated with frailty in both datasets underwent longitudinal analysis.

To establish the most robust associations, a stringent Bonferroni-corrected significance threshold of P < 1.443 E-05 was employed for identifying top hits. Numeric variables were standardized to z-scores. All associations were initially unadjusted (model 0), and were subsequently adjusted for age, gender, and assessment center (model 1), and for additional covariates including education, ethnicity, Townsend Deprivation Index, smoking status, and alcohol use in model 3. Stratified analyses were performed based on age at baseline (< 60 and ≥ 60 years), sex (male and female), and PRS (low, median, and high). To ensure the robustness of our findings, we conducted a sensitivity analysis by repeating the original analysis prior to interpolation.

We categorized the continuous variables into tertiles. Protective factors (OR < 1) in EWAS were reversed and coded to indicate unfavorable effects. The weighted standardized score for each domain was generated based on the β coefficient of each variable in the logistic model, adjusting for other risk factors within the same domain and for factors in model 3. The initial binary variables were multiplied by their respective β coefficients, aggregated, normalized by the total β coefficient, and then multiplied by 100 [19]. The likelihood of exposure to multiple risk factors increases with higher score values. Subsequently, we categorized the scores into three quartiles: favorable (lowest risk), intermediate, or unfavorable (highest risk). We employed a logistic regression model to explore the associations between incident frailty and ten variable domains, adjusting for the factors in model 3 and accounting for mutual adjustments among the ten domains. The analysis was performed twice, initially using the non-frailty population (robust and pre-frailty) as the reference and then with only the robust population as the reference.

Understanding the role of modifiable factors in frailty is crucial for clinical and public health recommendations. PAF quantifies the proportion of preventable diseases by substituting a specific risk factor with a more favorable alternative [20]. The individual-adjusted PAF indicates the highest proportion of preventable frailty cases achievable by eliminating a particular risk factor, regardless of other factors. Conversely, the combined adjusted PAF provides insight into the proportion of frailty cases avoidable by removing all risk factors simultaneously [21]. Using data from UKB, the stdReg R package generated PAF for each domain in univariate logistic regression models adjusted for model 3 variables [22]. Considering interdependence among ten domains, PCA-estimated PAF weights used to calculate combined-weighted PAF and individually weighted PAF. To produce conservative results (eliminating the worst 1/3 risk factors), intermediate and favorable profiles of ten domains were merged.

All P values were two-sided. The data were analyzed using R 4.0.3, PLINK 1.9, and Python 3.9.

Results

The status of frailty and the corresponding baseline characteristics of participants can be found in Table 1. Among the 469,301 participants with complete data for all five frailty indicators, 256,098 (54.6%) were classified as robust, 192,183 (41.0%) as pre-frail, and 21,020 (4.4%) as frail. The incidence of pre-frailty and frailty showed a more pronounced prevalence in females compared to males, and both conditions exhibited an upward trend with increasing age. Frailty prevalence increased from 2.8% in the 37–45 age group to 3.8% in the 45–55 group, reaching 5.6% among individuals aged 65 and above. Age-related decline was notable in grip strength and walking speed, while self-reported fatigue and weight loss were more frequently reported among younger participants (Table S2).

Table 1.

Baseline characteristics of the study population by frailty status

Variables	Robust (n = 256,098)	Pre-frailty (n = 192,183)	Frailty (n = 21,020)	P value
Age at baseline, years, mean (SD)	56.05 (8.08)	56.95 (8.09)	58.06 (7.63)	< 0.001
Sex, N (%)
Male	125,034 (51.2%)	81,386 (57.7%)	7734 (63.2%)	< 0.001
Female	131,064 (48.8%)	110,797 (42.3%)	13,286 (36.8%)
Education, N (%)
≥ College or University degree	106,392 (41.5%)	67,935 (35.3%)	5984 (28.5%)	< 0.001
< College or University degree	149,706 (58.5%)	124,248 (64.7%)	15,036 (71.5%)
Ethnicity
British	232,904 (90.9%)	167,911 (87.4)	17,326 (82.4)	< 0.001
Others	23,194 (0.09%)	24,272 (12.6%)	3694 (17.6%)
Townsend Deprivation Index, mean (SD)	− 1.74 (2.84)	− 1.05 (3.17)	0.44 (3.55)	< 0.001
Smoking status
Never	144,904 (56.6%)	102,453 (53.3%)	9607 (45.7%)	< 0.001
Previous/current	111,194 (43.4%)	89,730 (46.7%)	11,413 (54.3%)
Alcohol drinker status
Never	7399 (2.9%)	9970 (5.2%)	2,260 (10.8%)	< 0.001
Previous/current	248,699 (97.1%)	182,213 (94.8%)	18,760 (89.2%)

Data are n (%) and mean (SD). P values are derived using ether Student’s test, Mann–Whitney U test, or chi-square test

SD standard deviation

Modifiable factors in EWAS

Potentially modifiable factors may serve as promising targets for preventing and treating frailty. Our analysis revealed that out of the 223 variables initially linked to frailty in the baseline discovery dataset (Fig. 2A), a substantial 213 variables (96%) were confirmed in the baseline replication dataset (Fig. 2B). Further analysis of the 213 replicated variables over time revealed that 90 variables (40%) showed a significant correlation with frailty, which can be classified into ten categories (Fig. 2C). All associations were in the expected direction (Table S4). Among these factors, 67 have the potential to negatively impact health, while 23 have been found to have a protective effect. Five most significant protective factors are overall health rating (excellent, OR [95% CI] = 0.002 [0.001–0.002]), getting up in morning (very easy, 0.044 [0.041–0.048]), leisure/social activities (sports club or gym, 0.154 [0.145–0.164]), average total household income before tax (52,000 to 100,000, 0.162 [0.148–0.177]), and above moderate/vigorous/walking recommendation (yes, 0.183 [0.175–0.192]). The five most significant risk factors are daytime dozing/sleeping (narcolepsy) (often/all the time, 9.450 [8.688–10.278]), heart failure (9.034 [7.576–10.773]), falls in the last year (more than one fall, 8.926 [8.415–9.469]), pain type(s) experienced in last month (back pain, 7.114 [6.760–7.486]), and nap during the day (usually, 6.726 [6.253–7.236]).

Fig. 2.

The figure displays an association plot illustrating the relationship between modifiable risk factors and the incidence of frailty, with the x-axis categorized by conceptual domains and the y-axis depicting the statistical significance as − log10 of the p-value. A horizontal line is depicted to indicate the significance threshold corrected for multiple testing. To enhance readability, a subset of the most significant factors is annotated. The full set of association results is provided in the supplemental materials

Figure 3 and Table S6-12 report the results of an EWAS analysis in different subgroups. The study found that 52 variables were significantly correlated with frailty in males, which could be categorized into eight groups including circadian, dietary, health and medical history, lifestyle, physical activity, physical measures, psychosocial factors, and sociodemographic. Interestingly, the research revealed that women may be more vulnerable to a broader range of factors associated with frailty than men. Specifically, the study identified 83 factors that were significantly associated with frailty in women, indicating that these factors may have a greater impact on female health outcomes. Additionally, the study identified 86 factors that were significantly associated with frailty in middle-aged participants, highlighting their vulnerability to a wider range of frailty-related factors compared to older adults (n = 45). We have identified 90 positive modifiable risk factors specific to the frailty-prone population with high PRS. Notably, this study first reported the joint effect of genetic risk and modifiable phenotype. We found that a higher risk of developing frailty was significantly associated with 90 modifiable factors and higher frailty-related PRS. This indicates that the 90 phenotypes may be a factor attenuating the genetic risk of frailty. Unlike genetic risk, the manifestation of physical frailty can be reversed by changing lifestyle or physical exercise habits. Therefore, delaying the development of frailty by modifying risk factors can be offered as an option. Overall, these findings suggest that certain factors may have a more significant impact on the development of frailty in specific subgroups, and could provide valuable insights for developing targeted interventions to prevent or delay frailty in these populations.

Fig. 3.

Summary heat map for significant factors in EWAS analysis across the full sample and subgroups. The color of cells indicates the effect sizes (OR) between each risk factor and incident risk. Asterisks in cells represent significant associations after correction for multiple testing (Bonferroni-corrected, P < 1.443e-5

Joint effects of identified factors on frailty prevention

Compared with the favorable profile, intermediate and unfavorable profiles of biochemical measurements (intermediate, OR [95% CI] = 1.297 [1.238–1.359]; unfavorable, 1.122 [1.004–1.253]), circadian (intermediate, 1.336 [1.263–1.413]; unfavorable, 2.667 [2.536–2.805]), dietary (intermediate, 1.297 [1.238–1.359]; unfavorable, 1.729 [1.655–1.806]), family factors (unfavorable, 1.122 [1.004–1.253]), health and medical history (intermediate, 1.381 [1.268–1.504]; unfavorable, 5.187 [4.800–5.604]), lifestyle (intermediate, 1.439 [1.355–1.529]; unfavorable, 2.152 [2.034–2.277]), physical activity (intermediate, 1.526 [1.439–1.619]; unfavorable, 3.662 [3.470–3.865]), physical measures (intermediate, 1.585 [1.492–1.684]; unfavorable, 2.797 [2.635–2.969]), psychosocial factors (intermediate, 1.807 [1.685–1.937]; unfavorable, 3.277 [3.067–3.500]), and sociodemographic (intermediate, 1.191 [1.123–1.262]; unfavorable, 2.465 [2.336–2.601]) significantly increased the risk of frailty (Fig. 4). Pattern of results was nearly identical in the sensitivity analysis (Figure S1).

Fig. 4.

Associations between ten domains and frailty. The favorable profile was set as reference in each domain. The associations were estimated applying Logistic model including all ten domains mutually adjusted and with adjustment of age, sex, and assessment center, education, ethnicity, Townsend Deprivation Index, smoking status, and alcohol use. Robust and pre-frailty individuals were set as the control group. Abbreviations: OR, odds ratio

PAF estimates for the ten domains in frailty prevention

When shifting all unfavorable profiles to intermediate and favorable ones, overall adjusted PAF for potentially modifiable frailty risk factors was 85.9%, which increases to 86.6% if all factors are transformed into favorable tertiles (Table S23-26). Under a more conservative setting, the most significant preventive impact was attributed to individuals’ health and medical history, which was estimated to result in an 11.9% decrease in the incidence of frailty. Other domains were responsible for 4.5% (biochemical measurements), 9.8% (circadian), 6.7% (dietary), 3.2% (family factors), 9.9% (lifestyle), 10.1% (physical activity), 9.7% (physical measures), 10.4% (psychosocial factors), and 9.7% (sociodemographic) of frailty cases (Table 2).

Table 2.

Weighted and unweighted PAF for ten domains

	Model 1			Model 2
Domains	Un weighted PAF	Communality	Weighted PAF	Un weighted PAF	Communality	Weighted PAF
Biochemical measurements	0.115	0.344	0.045	0.117	0.537	0.045
Circadian	0.248	0.130	0.098	0.250	0.105	0.097
Dietary	0.169	0.240	0.067	0.184	0.235	0.071
Family factors	0.082	0.172	0.032	0.082	0.042	0.032
Health and medical history	0.301	0.245	0.119	0.301	0.246	0.117
Lifestyle	0.252	0.184	0.099	0.265	0.145	0.103
Physical activity	0.257	0.212	0.101	0.263	0.198	0.102
Physical measures	0.247	0.118	0.097	0.256	0.205	0.100
Psychosocial factors	0.264	0.236	0.104	0.288	0.184	0.112
Sociodemographics	0.246	0.119	0.097	0.222	0.103	0.087
Overall weighted PAF			85.9			86.6

The calculation of the weighted PAF considered the overlap between risk factors. The unfavorable profiles were reclassified as intermediate or favorable categories in model 1. In model 2, all risk factors were reclassified into the favorable tertile. Robust and pre-frailty individuals were set as the control group

PAF population attributable fraction

Discussion

This cohort study of 469,301 participants found 4.4% were frail and over 41.0% had pre-frailty using an adapted definition. Compared to a previous systematic review, the lower prevalence here may be attributed to the younger age of participants [23]. This largest EWAS to date found 90 modifiable risk factors for frailty across ten domains. Poor profiles of each domain independently increased frailty risk. Optimizing them can prevent 85.9% of cases. Health and medical history, psychosocial factors, and physical activity were significant contributors (11.9%, 10.4%, and 10.1% respectively). Based on our data-driven investigation, we conclude that such modifications could contribute to preventing frailty and lead to substantial reductions in specific morbidities and mortality over time.

Our SNP-based frailty heritability estimate was 5%, lower than previous family or twin-based estimates (30 to 45%). This difference may be due to our method not accounting for rare, non-additive, or non-autosomal genetic factors. While low heritability suggests non-genetic factors may have a greater impact on frailty, it does not rule out potential gene involvement through other mechanisms. We found 90 factors from 10 domains. Consistent with existing research, modifiable factors involving leisure/social activities [24], household income [25], walking [26], heart failure [27], back pain [28], and nap during the day were among the top correlates. Frailty was also associated with relatively unexplored factors, such as waking up in the morning, experiencing daytime drowsiness or narcolepsy, and experiencing falls. Those indicate that most frailty-related factors identified can be modified at an individual level. Promoting these modifiable risk factors to the public encourages proactive measures toward preventing or delaying frailty and improving the overall quality of life. This approach could potentially reduce the burden on governments and public health by empowering individuals to take charge of their own health. Furthermore, previous researches aimed at preventing and addressing frailty have mainly focused on physical activity, exercise, and nutritional interventions, with disease commonly studied as outcome of frailty [29–31]. However, through an examination of the synergistic effects of related exposures, it was found that poor health and medical history are the most critical risk factors for frailty. The estimation of PAF also highlights the importance of intervening or modifying health and medical history as a priority to promote effective management of frailty. If lifestyle interventions cannot be fully implemented, public health programs should prioritize other coexisting illnesses.

This study conducted extensive data analysis, identifying potential risk factors and frailty outcomes such as epilepsy, arrhythmias, and pancreatitis with robust statistical support. Additionally, our EWAS integrated various self-reported psychosocial factors as indicators, revealing a significant association between subjective experiences of irritability, anxiety, loneliness, and frailty. By addressing only adverse conditions and adhering to moderate and favorable interventions in psychosocial factors, we observed the second most substantial potential for reducing frailty incidence. These findings align with previous cohort studies, underlining the importance of a comprehensive understanding of the interplay between physical frailty and mental behaviors [32]. Prior studies and meta-analyses have also reported a bidirectional relationship between physical frailty and mental behaviors. The MR analysis utilizing genetic variations also discovered a causal effect of frailty on mental and behavioral disorders [33, 34]. Specifically, compromised mental and behavioral conditions elevate the risk of physical frailty, while individuals with physical frailty face an increased risk of mental and behavioral disorders. This perpetuates a detrimental cycle, exacerbating both physical and mental health challenges [35, 36]. To interrupt this cycle resulting from the synergistic impact of frailty and psychosocial factors, it is imperative to incorporate both physical and mental dimensions into treatment and prevention strategies.

While the initial criteria for defining frailty phenotype were limited to individuals over 60 years old, it has become evident that frailty is also experienced by younger individuals [37, 38]. This condition is not solely determined by chronological age but reflects biological and phenotypic factors [38]. In the subgroup analyses, our extensive exploration of risk factors related to frailty revealed that middle-aged individuals are more affected by a greater number and variety of modifiable factors compared to older adults. Middle-aged individuals exhibited a significantly higher number of frailty risk factors across a wider range of domains, with 86 factors identified across 9 domains, compared to only 45 factors across 8 domains in older adults. Most epidemiological studies on frailty excluded individuals younger than 60 years old [39]. Investigations into its risk factors not adequately incorporate this population, with limited studies having small samples or incomplete analyses. Our research suggests shifting focus to middle-aged individuals for earlier identification of frailty or pre-frailty and intervention.

Our study also has limitations. First, assessments of self-reported frailty criteria have indicated their comparability with, or in certain scenarios, superiority to alternative measures obtained through direct evaluation [40]. However, frailty assessment using a combination of objective measurements and self-reported characteristics is susceptible to reporting bias. Second, despite the advantage, some meaningful associations may be difficult to discern due to multiple testing corrections. Several variables failed to meet the “top” factor criteria. We emphasized robustly linked factors, but complete findings in Supplementary Tables should be reviewed.

In conclusion, research offers evidence-based guidance for clinical practitioners and policymakers to intervene in frailty effectively and reduce disease and mortality. The complexity of multimorbidity, resulting from the accumulation of diverse diseases, presents a tough prevention target when undertaken on an individual basis. However, the strong association between frailty and a wide range of systems including 65 common health conditions underscores the importance of frailty as a critical, modifiable target for disease prevention. To assess and manage frailty, we identified 90 potential underlying causes, such as pain, disease, loneliness, worry, or sleep disturbances. This comprehensive approach encompasses 10 areas, including psychosocial, lifestyle, and circadian rhythms, among others, empowering individuals to improve in these aspects and reduce their risk of frailty and chronic disease.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contribution

L Tan, W Cheng, and JT Yu designed the study. LZ Ma and YJ Ge conducted the main analyses and drafted the manuscript. YJ Ge and BS Wu conducted the genetic analyses. SD Chen and Y Zhang contributed the interpretation of the results. Y Zhang and LZ Ma contributed to data collection. JF Feng, W Cheng, and JT Yu critically revised the manuscript. All authors reviewed and approved the final version of the manuscript and all authors had full access to the data in the study and accept responsibility to submit for publication.

Funding

This study was supported by grants from the Science and Technology Innovation 2030 Major Projects (2022ZD0211600), the National Natural Science Foundation of China (82071201, 81971032), Shanghai Municipal Science and Technology Major Project (No. 2018SHZDZX01), Research Start-up Fund of Huashan Hospital (2022QD002), Excellence 2025 Talent Cultivation Program at Fudan University (3030277001), Shanghai Talent Development Funding for The Project (2019074), and ZHANGJIANG LAB, Tianqiao and Chrissy Chen Institute, and the State Key Laboratory of Neurobiology and Frontiers Center for Brain Science of Ministry of Education, Fudan University. Wei Cheng was supported by grants from the National Natural Sciences Foundation of China (no. 82071997) and the Shanghai Rising-Star Program (no. 21QA1408700).

Data availability

The data used in the present study are available from UKB with restrictions applied. Data were used under license and are thus not publicly available. Access to the UKB data can be requested through a standard protocol (https://www.ukbiobank.ac.uk/register-apply/).

Declarations

Ethical approval

All participants gave written informed consent prior to data collection. UK Biobank has full ethical approval from the NHS National Research Ethics Service (16/NW/0274).

Statement

The manuscript has been read and approved by all the authors, that the requirements for authorship as stated earlier in this document have been met, and that each author believes that the manuscript represents honest work.