Understanding the Causes of Frailty Using a Life-Course Perspective: A Systematic Review

(1) Background: Few studies have examined risk factors of frailty during early life and mid-adulthood, which may be critical to prevent frailty and/or postpone it. The aim was to identify early life and adulthood risk factors associated with frailty. (2) Methods: A systematic review of cohort studies (of at least 10 years of follow-up), using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (PRISMA). A risk of confounding score was created by the authors for risk of bias assessment. Three databases were searched from inception until 1 January 2023 (Web of Science, Embase, PubMed). Inclusion criteria were any cohort study that evaluated associations between any risk factor and frailty. (3) Results: Overall, a total of 5765 articles were identified, with 33 meeting the inclusion criteria. Of the included studies, only 16 were categorized as having a low risk of confounding due to pre-existing diseases. The long-term risk of frailty was lower among individuals who were normal weight, physically active, consumed fruits and vegetables regularly, and refrained from tobacco smoking, excessive alcohol intake, and regular consumption of sugar or artificially sweetened drinks. (4) Conclusions: Frailty in older adults might be prevented or postponed with behaviors related to ideal cardiovascular health.


Introduction
Numerous potential biomarkers of aging have been proposed in the scientific literature, including molecular, imaging, and clinical data [1].Frailty is a composite aging biomarker characterized by a condition of decreased physiological reserve that leads to a vulnerable state and increases the risk of adverse health outcomes when exposed to a stressor [1].In 2001, two definitions of frailty were introduced in the geriatric literature (although more definitions can be found in the scientific literature).The phenotypic model of Fried et al. [2] is based on the presence of three (or more) of the following characteristics: (a) an involuntary weight loss, (b) self-reported exhaustion in daily life activities, (c) a low level of physical activity, (d) habitual slow walking speed, and (e) muscular weakness.Another frailty definition is the accumulation of deficits model (or frailty index) of Mitnitski et al. [3], which includes deficiencies of functional, sensory, and clinical nature.In a recent systematic review, the prevalence of frailty in older adults in 62 countries and territories was 12% (Fried phenotype) or 24% (frailty index) [4].Women and individuals of a low socioeconomic status level are more likely to become frail, according to a narrative review of Taylor et al. [5].Frailty remains an important public health problem because frail individuals (versus non-frail) are at higher risk of physical disability [2], falls [6], fractures [7], hospitalizations [8], institutionalization [9], and death [10].
In a recent systematic review of older adults (at least 65 years old) [11], authors identified a large number of lifestyle factors and characteristics associated with frailty.Information was mainly derived from cross-sectional studies or cohort studies with a short follow-up.However, it is well established that a life-course perspective offers a more suitable approach to understanding how the aging processes and their consequences emerge during the lifetime [12].Lowering the accumulation of harmful exposures throughout the life course or changing unhealthy behaviors during adulthood may lead to more favorable trajectories of aging [12].However, many statistical associations found in observational studies of risk factors of frailty could reflect bias (reverse causality, selection bias, and measurement errors), confounding, or chance [13].To reduce the risk of bias due to preexisting diseases in the synthesis of evidence, some epidemiologists recommend following some analytical approaches [14], such as (1) excluding (or adjusting for) participants with major noncommunicable diseases (NCDs) at baseline (CVDs, stroke, cancer, and respiratory diseases); (2) including only cohort studies with a minimum of 10 years of follow-up in meta-analysis; and (3) excluding death cases occurring in the first 5 years of follow-up.
To sum up, to the best of our knowledge, no systematic review until now has evaluated how early-life and middle-life risk factors are associated with frailty or studied their epidemiological validity using risk of bias assessments.The main objective of this systematic review was to identify early-and middle-life risk factors associated with frailty in older adults.We also aimed to examine whether authors included appropriate analytical methods to deal with confounding due to pre-existing diseases.

Literature Search and Screening
To formulate a clear and concise research question, a description of the population, intervention/exposure, comparison, and outcomes are provided.We searched cohort studies that examined associations between any risk factor (Exposure) during adulthood, adolescence, childhood, or natal factors (Population) associated with frailty (Outcome), using Web of Science, Embase, and PubMed (from inception until 1 January 2023).One researcher (A.S.) was in charge of producing the first database for the identification of relevant scientific literature.Details of the list of keywords are included in Table S1.We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines to report the results of this systematic review [15].Two authors, A.B. and J.P.R.-L., screened independently articles by title and abstract and, in a later stage, reading full-text articles using the website covidence.orgIn case of discrepancies, a third author (C.J.K.) made a final decision.
The eligibility criteria of this systematic review were any cohort study in humans (both sexes, healthy at baseline), published in English language, that evaluated associations of risk factors associated with frailty status in humans.Studies with a retrospective study design were eligible studies, as they inform about early-in-life risk factors.In addition, we considered eligible those studies cited in references from selected studies.Exclusion criteria: studies whose cohort studies had a follow-up lower than 10 years were excluded because they did not inform of early-or middle-life risk factors.Studies published in non-English language or abstracts of conferences were excluded.

Data Extraction
We retrieved the first author's name, year of publication, country, sample size, participant's sex, age at baseline, exposure variable/s, frailty definition, average length of follow-up, and fully adjusted hazard ratio (HR) or odds ratio (OR) or relative risk ratio (RR) and 95% confidence intervals (CIs) for frailty, comparing for each exposure variable (using 1-unit increase or comparations between categories with the highest and lowest values; Comparison).We also extracted data about the covariates used in the fully adjusted model and whether authors included sensitivity analyses in their publications.Data extraction was performed by one researcher (A.B.) and double-checked by another (J.P.R.-L.).

Risk of Confounding Due to Pre-Existing Diseases
Three methodological characteristics defined the risk of confounding based on subject matter expertise instead of a mechanistic risk of bias assessment [16]: average age at baseline of 70+ years, authors did not exclude participants with diseases/conditions at baseline and did not adjust for diseases/conditions in the fully adjusted model.A risk of confounding score was created with the three mentioned characteristics (ranging from zero to three).We defined a high risk of confounding due to pre-existing diseases when studies had two or three points in the confounding score (total of three).Scores were calculated by one senior researcher (J.P.R.-L.) with prior experience in risk-of-bias assessments of epidemiological studies.A comprehensive meta-analysis of each risk factor identified was initially planned, taking into account the risk of confounding scores, but it was finally discarded due to the scarce number of studies identified.

Results
Figure 1 shows the PRISMA 2020 flow diagram used in the present systematic review.A total of 7425 records were initially identified.After screening 5681 articles by title and abstract, 84 articles were retrieved for eligibility analyses through a full-text reading.Of these, 33 articles were finally selected .Each section may be divided by subheadings.It should provide a concise and precise description of the experimental results, their interpretation, as well as the experimental conclusions that can be drawn.
Healthcare 2023, 11, x FOR PEER REVIEW 3 of 17 1-unit increase or comparations between categories with the highest and lowest values; Comparison).We also extracted data about the covariates used in the fully adjusted model and whether authors included sensitivity analyses in their publications.Data extraction was performed by one researcher (A.B.) and double-checked by another (J.P.R.-L.).

Risk of Confounding Due to Pre-Existing Diseases
Three methodological characteristics defined the risk of confounding based on subject matter expertise instead of a mechanistic risk of bias assessment [16]: average age at baseline of 70+ years, authors did not exclude participants with diseases/conditions at baseline and did not adjust for diseases/conditions in the fully adjusted model.A risk of confounding score was created with the three mentioned characteristics (ranging from zero to three).We defined a high risk of confounding due to pre-existing diseases when studies had two or three points in the confounding score (total of three).Scores were calculated by one senior researcher (J.P.R.-L.) with prior experience in risk-of-bias assessments of epidemiological studies.A comprehensive meta-analysis of each risk factor identified was initially planned, taking into account the risk of confounding scores, but it was finally discarded due to the scarce number of studies identified.

Results
Figure 1 shows the PRISMA 2020 flow diagram used in the present systematic review.A total of 7425 records were initially identified.After screening 5681 articles by title and abstract, 84 articles were retrieved for eligibility analyses through a full-text reading.Of these, 33 articles were finally selected .Each section may be divided by subheadings.It should provide a concise and precise description of the experimental results, their interpretation, as well as the experimental conclusions that can be drawn.Table 1 describes the main characteristics of all cohort studies selected.The population sample sizes ranged between 323 and 121,700 participants; 24 studies were conducted on both sexes, 4 only in men, and 5 only in women; 8 studies recruited participants in the USA, 7 in Finland, 1 in Australia, 4 in France, 6 in the United Kingdon, 1 in Israel, 1 in Sweden, 3 from China, and 1 in the Netherlands; participants were followed up between 10 and 30 years; the 11 definitions of frailty used across the cohort studies were Fried phenotype [2] (15 studies), Modified Fried phenotype (6 studies), FRAIL scale-Abellan van Kan et al. [50] (1 study), FRAIL scale-Morley et al. [51] (4 studies), Modified FRAIL scale-Morley et al. (1 study), Frailty index-Mitnitski et al. [3] (2 studies), Hospital Frailty Risk Score (HFRS)-Gilbert et al. [52] (1 study), Frailty phenotype-Kucharska-Newton et al. [53] (1 study), Frailty phenotype-Strawbridge et al. [54] (1 study), and Frailty index-Searle et al. [55] (1 study).The prevalence of frailty ranged between 2 and 61%.
The exposure variables of the 33 articles selected were dietary inflammatory index in adulthood (1 study), blood inflammatory markers in adulthood (two studies), alcohol consumption in adulthood (three studies), sitting time in adulthood (one study), multicomponent healthy heart score in adulthood (one study), overweight/obesity or higher BMI in adulthood (four studies), neighborhood-social deprivation in childhood (one study), cardiovascular risk scores in adulthood (two studies), physical inactivity in adulthood (five studies), asthma in adulthood (one study), anemia in adulthood (one study), diabetes in adulthood (one study), high liver enzymes (one study), dietary clusters (pasta or biscuits plus snacking) in adulthood (one study), birth body composition (BMI, weight, length) (one study), children of separated parents in childhood (one study), fruit and vegetable consumption in adulthood (three studies), smoking status in adulthood (two studies), nonsteroidal anti-inflammatory drugs (NSAID) use (one study), neighborhood quality in adulthood (one study), education level achievement (three studies), paternal education (one study), occupation or employment level in adulthood (two studies), low literacy in adulthood (one study), low income in adulthood (one study), malnutrition in adulthood (one study), depression in adulthood (one study), forced expiratory volume or HDL cholesterol or hypertension (one study), sugar-sweetened beverages or artificial sweetened beverage or orange juices or non-orange juices (one study), red meat in adulthood (one study), pain during walking in adulthood (one study), subjective social status (one study), health-related quality of life (one study), and social vulnerability index in adulthood (one study).
Table 2 shows the analytical approaches used to account for confounding due to preexisting diseases in 33 studies examining the association between risk factors and frailty.In 10 of them, authors omitted the inclusion of any type of disease as a covariate in their fully adjusted models; only 2 studies excluded all participants with diseases in main analyses; and another 2 used sensitivity analyses.[48] Legend: None; adjusted in the model, excluded participants with morbidities in main analysis, excluded participants with morbidities in sensitivity analysis.CVD: cardiovascular disease; CPK: chronic kidney disease.
Table 3 shows the scored risk of confounding due to pre-existing diseases in the 33 studies selected.The final score takes into account whether studies included participants younger than 70 years at baseline or not, whether studies excluded participants with diseases/conditions at baseline, and whether studies adjusted for diseases/conditions in the fully adjusted model.A total of 16 studies scored a low risk of confounding due to pre-existing diseases (0 or 1 point) and the rest a high risk of confounding (2 or 3 points).Table S2 shows whether authors included physical activity or nutritional factors (energy intake, quality nutritional indexes, sugar-sweetened beverages, and red meat) as covariates in their regression models.In 20 studies, authors omitted both physical activity and nutritional factors as covariates in their statistical analyses.

Discussion
The goal of this systematic review was to identify early-life and middle-life risk factors (any exposure variable) associated with frailty.We found evidence that maintaining a normal weight in adulthood, being physically active, not smoking tobacco, refraining from ultra-processed food and beverages, and avoiding excessive alcohol intake may decrease the risk of frailty several decades later.These findings may have important implications for elderly populations because, in theory, most cases of frailty might be prevented if populations remain healthy before older age.For example, staying physically active in adulthood was robustly associated with a lower future risk of frailty [32].The physiological mechanisms underlying the positive influence of physical activity on frailty prevention have been comprehensively reviewed elsewhere [5].On the other hand, the regular consumption of fruits and vegetables and a low consumption of SSBs or ASBs were also associated with a lower risk of becoming frail [46].Therefore, it seems unquestionable that diet and physical activity have a key role in preventing the future risk of frailty.In support of this, we found that obesity [21], diabetes [26], or having worse cardiovascular risk scores [27] or blood inflammatory markers [40] were equally associated with a higher risk of frailty.The mechanisms by which high ultra-processed foods and beverages promote obesity, diabetes, or cardiovascular disease are complex and only partially known nowadays.High ultra-processed foods and beverages may result in unique patterns of gut-brain signals during digestion processes, as they are absorbed more proximally in the gut compared with natural foods and beverages.Altered absorption of nutrients and a low amount of fiber in the diet may play important disruptions in control, leading to a long-term positive energy balance [56].
Nonetheless, a second goal of our review was to evaluate whether authors included appropriate analytical methods to deal with confounding due to pre-existing diseases and we found serious deficiencies in this matter.For example, many (half of the studies selected) epidemiological studies of risk factors of frailty were at high risk of confounding due to pre-existing diseases.Another source of concern was the observation that the majority of studies did not adjust their effect estimates by important confounders such as physical activity or nutritional factors.The framing that frailty is a direct consequence of the normal aging process should be approached with caution, as we also find evidence that frailty was associated with worse socioeconomic markers (income, education, and employment).So far, the mechanisms linking socioeconomic factors with frailty remain unexplored, and multiple factors may be involved (beyond the proven benefits of physical activity or healthy diets).Although the best way to define frailty is still debated among scientists [51], future studies should adopt (at least) the most common definition of frailty (Fried phenotype) to allow a future synthesis of the scientific evidence [57].
To move the field forward, it is important to acknowledge that well-powered randomized clinical trials, although the gold standard of scientific inquiry, are limited in their ability to add valuable insights for prevention because it is unfeasible to test human interventions with a duration of 10 years or more.To illustrate how clinical trials in the elderly do not always offer important insights about how to prevent frailty, see reference [58], where a complex intervention that combined a nutrition plus physical activity intervention over 2 years was ineffective in reducing frailty in older adults.Although it could be argued that the design of interventions was not optimal (for example, short duration or the physical activity intervention only included 1 h per week of strength plus balance instead of exercise programs of aerobic activities based on physical activity recommendations for health in adults), the authors stressed that their interventions mirrored the real world (good external validity).Consequently, we think that future studies on this topic should rely on welldesigned cohort studies with valid methodologies of assessment of exposure variables and robust statistical analysis (including sensitivity analyses).Our systematic review examines, for the first time, what modifiable factors may determine a higher long-term risk of frailty (life course epidemiology) and evaluates the risk of confounding due to pre-existing diseases.Despite employing a comprehensive search strategy, we found very few studies evaluating the same exposure variable, which precluded our ability to perform additional meta-analyses.Therefore, we acknowledge that progress in this field of study is still limited.Nonetheless, facilitating the adoption of cardiovascular healthy behaviors in the general population seems a promising strategy of intervention to prevent frailty.

Conclusions
Maintaining a normal weight in adulthood, being physically active, not smoking tobacco, refraining from ultra-processed food and beverages, and avoiding excessive alcohol intake may decrease the risk of frailty several decades later.The framing that frailty is a direct consequence of the normal aging process should be viewed with caution, as we found clear evidence that more vulnerable socioeconomic groups are more likely to become frail.

Figure 1 .
Figure 1.PRISMA 2020 flow diagram showing 33 cohort studies (with at least 10 years of followup) of early-life and middle-life risk factors of frailty.

Figure 1 .
Figure 1.PRISMA 2020 flow diagram showing 33 cohort studies (with at least 10 years of follow-up) of early-life and middle-life risk factors of frailty.

Table 1 .
Main characteristics of 33 cohort studies included, examining associations between any exposure variable and risk of frailty.

Table 2 .
Analytical approaches to account for confounding due to pre-existing diseases in 33 studies of risk factors and frailty.

Table 3 .
Score risk of confounding due to pre-existing diseases in 33 studies examining the association between risk factors and frailty.