The Association Between Frailty and All-Cause Mortality in Community-Dwelling Older Individuals: An Umbrella Review

Background Frailty is associated with multiple adverse health outcomes, including mortality. Several methods have been used to characterize frailty, each based on different frailty scales. These include scales based on a phenotype, multidomain, and decit accumulations. Several systematic reviews have examined the association between frailty and mortality; however, it is unclear whether these different frailty scales similarly predict mortality. This umbrella review aims to examine the association between frailty assessed by different frailty scales and all-cause mortality among community-dwelling older adults. Methods The umbrella review protocol was registered at PROSPERO, and it was conducted following the PRISMA statement. MEDLINE, Embase, PubMed, Cochrane Database of Systematic Reviews, Joanna Briggs Institute (JBI) EBP database, and Web of Science database was searched to identify systematic reviews and meta-analyses examining the association between frailty and all-cause mortality. Methodological quality was assessed using the JBI. Critical Appraisal Checklist and online AMSTAR-2 critical appraisal checklist. For eligible studies, essential information was extracted and synthesized qualitatively. Results Five systematic reviews were included, with a total of 434,115 participants. Three systematic reviews focused on single frailty scales; one evaluated Fried physical frailty phenotype and its modications; another focused on the decit accumulation frailty index. The third evaluated the FRAIL (Fatigue, Resistance, Ambulation, Illness, and Loss of weight) scale. The two other systematic reviews determined the association between frailty and mortality using different frailty scales. All the systematic reviews performed meta-analyses and assessed between-study heterogeneity. All of the systematic reviews found that frailty was signicantly associated with all-cause mortality. Conclusion This umbrella review demonstrates that frailty is a signicant predictor of all-cause mortality, irrespective of the specic frailty scale.


Background
Frailty is a geriatric syndrome characterized by reduced physiological reserves (1). It is associated with increased vulnerability to adverse outcomes such as falls, fractures, hospitalizations, institutionalizations, disability, dementia, lower quality of life, and mortality (2,3). The prevalence of frailty varies across populations, according to age, gender, ethnicity, and socioeconomic status, and ranges from 5.8-35% in community settings (4,5). The wide variation might result from differences in the frailty assessment scale applied, as different tools utilize different criteria to de ne and categorize frailty. The prevalence of frailty is expected to rise worldwide due to the rapidly growing aging population (6).
While the concept of frailty is widely recognized, there is no single explicit criterion to de ne frailty. In 2013, a consensus statement by six major international scienti c societies de ned frailty as a medical syndrome with multiple causes and contributors that is characterized by diminished strength, endurance, and reduced physiologic function and increases an individual's vulnerability for developing disability, dependency, or death (7,8). In this de nition, frailty is viewed as rstly, a clinical entity different from disability, sarcopenia, or multimorbidity; secondly, it affects a person's physical or cognitive domains; and nally, it is considered as a dynamic state, which can improve or deteriorate over time (7). Besides, an intermediate or 'prefrail' stage has also been recognized (3,9,10).
Several frailty scales have been developed to characterize frailty in older adults, described in three broad categories. The rst category includes focused physical scales, which most notably contains the Fried physical frailty phenotype from the Cardiovascular Health Study, as well as adaptations derived from this original scale (3). It consists of ve components: unintentional weight loss, muscle weakness, exhaustion or low energy level, slowness or slow gait, and low physical activity. Persons are frail if three or more of the ve criteria are met. The second category of frailty scales is a multidomain scale (11), which describes multidimensional characteristics of frailty containing more than one medical, physical, cognitive, or environmental factors. The third type of scale is known as a de cit accumulation frailty index (12). It consists of an inventory of various de cits covering multiple domains or body systems and the percentage of de cits calculated.
These three types of scales capture different aspects of the frailty syndrome and, therefore, there may be differences in their association with health outcomes. Understanding these differences is important because it could inform the context in which the different measures are best applied. All three categories of frailty assessment scales have some limitations. For example, the phenotype scale does not cover all dimensions of frailty, such as cognition or affect (3). The multidomain scale and the de cit accumulation model-based scales are comprehensive but time-consuming. Clinical information may not be readily available for all individual components, and consequently, it may be challenging to integrate this scale into regular healthcare practice (13).
Few systematic reviews have explored different frailty scales and determined whether frailty assessed by these scales is predictive of all-cause mortality. Furthermore, it remains unclear whether a particular frailty scale is a better predictor of mortality of community-dwelling older adults (3,6). Therefore, the objective of this umbrella review is to qualitatively synthesize and evaluate the association between frailty determined by different frailty scales and all-cause mortality in community-dwelling older people.

Methods
A protocol was developed, and the review was conducted following the 'Preferred Reporting Items for Systematic reviews and Meta-Analyses' (PRISMA ) statement (14). The protocol was registered at the International Prospective Register of Systematic Reviews or PROSPERO (ID: CRD 42020155407).

Data sources and search strategy
The search strategy aimed to nd published systematic reviews and meta-analyses that evaluated the association between frailty and all-cause mortality in community-dwelling older populations. Systematic and comprehensive searches were conducted in electronic databases: MEDLINE, Embase, PubMed, Cochrane Database of Systematic Reviews (CDSR), Joanna Briggs Institute Evidence-Based Practice (JBI EBP) Database, and Web of Science. The search was conducted in October 2019 and updated in July 2020. The search strategy and search terms are provided in Appendix I. Studies conducted on humans and articles published in English were considered eligible for this review, and duplicates were excluded. The searches were independently performed by two authors (ARMSE and CB). Any discrepancies were resolved by discussion.

Inclusion and exclusion criteria
We included systematic reviews and meta-analyses which have reported the association between frailty and mortality among community-dwelling older adults using any of the frailty scales, e.g., Fried physical frailty phenotype or modi cations, de cit accumulation frailty index, and multidomain frailty index. We excluded systematic reviews and meta-analyses. The review included only hospitalized and institutionalized older adults or examined disease-speci c outcomes (i.e., falls, fractures, heart failure, etc.) rather than mortality.

Study selection and data extraction
Two reviewers (ARMSE and CB) independently searched titles. They screened abstracts before retrieving the full texts, assessed eligibility for the type of participants, study design, and outcomes. Data were extracted using a standardized form including author and year of publication, location, population characteristic, sample size, the proportion of female participants, age range, frailty scales used, number of de cits used to create the frailty scales, and follow-up period. We also noted the quality and bias assessment, effect sizes, and measure of variance, most commonly hazard ratios with 95% con dence intervals and heterogeneity assessments.

Methodological quality assessment
Manuscripts were assessed for methodological quality before inclusion in the review. The quality assessment of the included ve systematic reviews were performed by ARMSE and CB. We used JBI. Critical Appraisal Checklist for Systematic Reviews and Research Syntheses (15) ( Table 1) and online 'A MeaSurement Tool to Assess systematic Reviews Version 2' (AMSTAR 2) checklist (16).

Data synthesis and analysis
The studies were combined using qualitative best evidence synthesis, as statistical pooling could not be done due to the high heterogeneity of meta-analyses of the included studies. We extracted and reported the pooled effect sizes of the outcomes meta-analyzed within the reviews ( Table 2).

Search results
A total of 969 records were identi ed from the six databases, and after removing the duplicates, 686 were screened for eligibility based on title and abstract. Twenty-three full-text articles were then reviewed for relevance, out of which 18 were excluded because aspects similar to, but not de ned explicitly as, frailty were assessed, e.g., gait speed (17,18), sarcopenia (19), different health indicators (20) or geriatric syndromes (21); outcomes other than mortality were examined, e.g., trauma (22), fractures (23,24), falls (25), high blood pressure and cardiovascular outcomes (26) or heart failure (27); study population included were from clinical practice (28), nursing home (29) or critical care (30) but not from a community setting; the study involved interventions, e.g., treatment modalities (31); or the article was a systematic review protocol or an umbrella review which evaluated frailty scales for clinical outcomes from community, residential care and hospital settings (32)(33)(34). This left ve eligible systematic reviews and meta-analyses in the umbrella review ( Fig. 1).
All ve reviews were of moderate to high quality, as assessed by the JBI critical appraisal checklist (Table 1) and online AMSTAR-2 checklist. The reviews included 93 studies (some of which were included in multiple systematic reviews), and they assessed a range of outcomes. Of these studies, 77 examined the association between frailty and all-cause mortality over one to sixteen years of follow-up, and are the focus of this systematic review. Of the ve systematic reviews, one review focused only on studies which used exclusively the frailty scale based on the Fried phenotype and its modi cations (11 scales in a total of which four were original and seven modi ed) (35); one examined the FRAIL scale which is a questionnaire-based phenotype scale with ve components, i.e., fatigue, resistance, ambulation, illness, and loss of weight (2); one review included studies assessing the de cit accumulation frailty index, with between 23 and 70 de cit items (36); one review included studies assessing either the Fried phenotype (7 studies) or the de cit accumulation frailty index (8 studies) (37); while the fth review included 25 different scales of which ve were Fried phenotype-based scales, 14 multidomain scales and six were de cit accumulation frailty index containing 23 to 83 de cits (11).  (Table 2). Individual study's frailty outcome was adjusted for a range of two to ten covariates (e.g., age, gender, education, smoking, alcohol intake, socioeconomic conditions) in their analysis. Overall ndings for the association between frailty and all-cause mortality All ve systematic reviews reported a signi cant association between frailty and an increased risk of mortality; however, the effect size between frailty and mortality varied across the included systematic reviews. For example, the meta-analysis that included 24 studies using three types of scales (i.e., Fried physical frailty phenotype, de cit accumulation frailty index, and multidomain frailty index) estimated an overall hazard ratio 2.34 (95% CI:1.77, 3.09) between frailty and all-cause mortality (11). The estimated overall relative risk was 1.83 (95% CI: 1.68, 1.98). In their analysis, comparing the non-frail to frail groups, the risk associated with mortality varied depending on the frailty scales used. The Fried physical frailty phenotype was associated with a 2.6-fold increased risk of mortality (HR: 2.58; 95% CI: 1.83, 3.64; I 2 = 89%, P < 0.001); the multidomain frailty index with a 2.1-fold increased risk (HR: 2.13; 95% CI: 1.38, 3.29; I 2 = 96%, P < 0.001); and the de cit accumulation frailty index a 1.85-fold (HR:1.85; 95%CI: 1.30, 2.63; I 2 = not available, P = not available) (11). Similar effect sizes were reported from the systematic review that included only the phenotype-based frailty index and found that frailty was associated with a two-fold increased risk of mortality than robust or non-frail persons (HR: 2.00; 95% CI: 1.73, 2.32) (35). Direct comparison of effect sizes from the other systematic review was not possible, given they considered the association between a one-unit increase in frailty score using the de cit accumulation frailty index and mortality (random effect model  (36). Only one systematic review included a questionnaire-based FRAIL scale to assess the relationship between frailty and mortality (2). From the eight studies included in this review, it was found that individuals classi ed as frail or prefrail, compared to non-frail individuals, had a 3.5-fold and 1.8-fold increased risk of mortality, respectively, over 2.4 years to 4.3 years of follow-up. The predictive value of mortality remained similar across de nitions of frailty, ranging from 54-70% in the receiver operating characteristic curve areas using a questionnaire-based FRAIL scale (2) and remained around 70% if the Fried physical frailty phenotype or the de cit accumulation frailty index were used (37).

Gender differences
Three of the ve reviews examined potential gender differences in the association between frailty and mortality and yielded some con icting results (35)(36)(37). For example, one review using the Fried physical frailty phenotype and another using the de cit accumulation frailty scale showed that older men with frailty had a higher risk of mortality than older women with frailty (35,36). However, the third review (37) found mixed results depending on the individual study, with some reporting that men had an increased risk of mortality (39)(40)(41)(42). Still, others found that women had an increased risk (41)(42)(43). One study reported a doseresponse association between a more signi cant number of de cits and increased mortality in women across all age categories (41). However, this review (37) did not directly compare the risk between gender. Age Age did not appear to be an effect modi er of the relationship between frailty assessed using the Fried physical frailty phenotype or de cit accumulation index, and mortality. Two of the ve systematic reviews examined the association between frailty and mortality according to age groups (35,36). The pooled estimates showed that the association between the de cit accumulation index and mortality did not vary between those aged below 65 years (HR:1.05; 95% CI: 1.03, 1.07) and those above 65 years (HR:1.04; 95% CI:1.03, 1.05) (36) per unit increase in a frailty index. Likewise, mortality risk was similar for those aged below 80 years (HR:1.62; 95% CI:1.39, 1.89) and above 80 years (HR:1.41; 95% CI:1.17, 1.70) estimated by the Fried physical frailty phenotype (35). Three other systematic reviews did not compare mortality based on age strati cation (2,11,37).

Follow-up duration
The association between frailty and mortality varied according to follow-up duration. The risk of mortality was the lowest when the follow-up period was less than 12 months (HR:1.33; 95% CI:1.11, 1.60) and was the highest when the follow-up period was between two years to ve years (HR: 3.25; 95% CI: 2.14, 4.94) (11).
However, another review using the de cit accumulation frailty index found that the risk of mortality was higher when a shorter follow-up time was examined than a more extended follow-up, but effect sizes are not mentioned (36). Likewise, one systematic review compared a follow-up time of 4 years versus 11 years and observed that the strongest association between frailty and mortality was in the shorter follow-up group.
However, individual values were not provided (37).

Discussion
There is increasing interest in frailty as a clinically meaningful measure of geriatric health condition, and the number of publications on frailty has increased exponentially over the last few decades (5,44).
Determination of frailty status is emerging as a signi cant predictor of outcomes in various medical disciplines, including cardiology (27,45), neurology (46,47), oncology (48), orthopedics (23), and surgery (49), as well as the senior population in general. Accordingly, the association between frailty and all-cause mortality has been investigated across different settings and populations. We identi ed ve moderate to high quality large systematic reviews examining three major categories of frailty scales and the association with all-cause mortality in community-dwelling older individuals. This umbrella review included 93 primary studies comprising 434,115 participants from Australia, Canada, China, Hong Kong, Mexico, the UK, the USA and multiple other European countries. A total of 80 different frailty scales were used across the different studies, ranging from Fried physical frailty phenotype and various modi cations of this scale to multidomain scales and the de cit accumulation frailty index of included de cits ranging from 23 to 83. The follow-up period ranged from 12 months to 16 years. A large number of community-dwelling participants across the world means that the outcomes of this study can be more widely generalized to older individuals. Heterogeneity in frailty de nitions, scales, adjustment strategies (37), different population groups, different follow-up periods (11), and modi cations of the phenotype scale (35) hampered the synthesis of evidence. Only one review (36) exclusively examined de cit accumulation frailty index and mortality. In that review, many primary studies were excluded because they did not provide effect size for mortality. The effect measures available were based on frailty groups de ned by different cut-off points. The review of studies using the questionnaire-based FRAIL scale also described moderate to high heterogeneity; furthermore, the FRAIL scale was relatively newer and only included a limited number of primary studies in the meta-analysis (2). The extent to which the different frailty scales predict mortality was a consideration of this umbrella review.
However, only two of the ve systematic reviews estimated the predictability of the included frailty scales, showing similar predictive value across de nitions of frailty. One review (37) compared the survival estimates based on age and adjusted relative risk using both the Fried physical frailty phenotype and the de cit accumulation frailty index. They found that frail older adult, compared to non-frail older adults de ned using the Fried phenotype, had an approximate 50% increase in mortality risk while studies using the de cit accumulation index reported an approximate 15% increase in risk. A similar predictive value was observed across de nitions of frailty, averaging around 70% in receiver operating characteristic curve areas (37). It is hard to directly compare predictive value across frailty measures, as only two studies directly assessed this.
However, the results of these studies showed that the receiver operating characteristic curve ranged from 0.54 to 0.70 (2,37).

Strengths And Limitations
The current umbrella review has multiple strengths. The protocol was registered at PROSPERO, and the PRISMA guidelines were followed in completing this review. The search strategy of the review was robust and reproducible and utilized comprehensive search terms in multiple electronic databases. We evaluated ve moderates to high-quality systematic reviews, which examined many participants from different parts of the world. Therefore, the generalizability of the results is high. We included systematic reviews that measured frailty using the commonly available scales, i.e., Fried physical frailty phenotype, multidomain frailty scale (including the questionnaire-based FRAIL scale), and de cit accumulation frailty index, meaning the ndings will be relevant more broadly.
However, there are some limitations. This umbrella review did not include intervention studies, or systematic reviews of frail participants from hospitals or nursing homes were excluded. Thus, the ndings apply to community-dwelling older individuals only.
For researchers, this umbrella review shows that any category of frailty scale has utility for predicting mortality. Finally, this umbrella review focussed on the utility of frailty assessment to predict mortality though it could be considered that delaying mortality is not the only or best objective for the geriatric population. Improving the quality of life before death or extending life free of disability could be considered as the critical outcomes for assessing risk in those who are frail, possibly a future analysis.

Conclusion
The ndings of this review provide satisfactory evidence that frailty is associated with mortality risk, and thus highlight the importance of assessing frailty in primary community settings. This umbrella review has demonstrated that frailty is a signi cant predictor of all-cause mortality regardless of the speci c frailty scale. For example, frailty assessed using ve components that excludes cognition and affect (Fried physical frailty phenotype) predicted mortality to a similar extent as did more comprehensive de cit accumulation frailty indices that included up to 83 items. As such, this implies that researchers and clinicians can use the most appropriate frailty scales given their circumstances, resources, and access to information. Together these ndings emphasize that the assessment of frailty status itself may be more important than the choice of which type of scale is used.