Association of frailty and mortality risk in community-dwelling older adults: a systematic review and meta-analysis of 59 international prospective cohort studies

Abstract

Background

The global aging population presents significant health challenges. This study aimed to examine the association between frailty and mortality among community-dwelling older adults.

Methods

We searched five databases up to April 29, 2025, for prospective cohort studies evaluating frailty as a predictor of mortality. Study quality and risk of bias were assessed using the ROBINS-E tool. A random-effects meta-analysis was conducted to estimate pooled mortality risks.

Results

A total of 59 studies were included in the systematic review, of which 53 were eligible for meta-analysis, comprising 185,355 participants and over 28,616 deaths. Follow-up periods ranged from 1 to 15 years. Compared to robust individuals, non-robust older adults had a significantly higher risk of all-cause mortality (hazard ratio [HR] = 1.83, 95% confidence interval [CI]: 1.70–1.97). This association was consistent across frailty domains: including physical frailty (HR = 1.74, 95%CI: 1.53–1.99), multidimensional frailty (HR = 2.25, 95%CI: 1.97–2.57), and phenotype-based models accessed frailty (HR = 1.71, 95%CI: 1.48–1.97). Subgroup analyses based on phenotype-based models showed that frail older adults had a higher risk of all-cause mortality (HR = 2.18, 95%CI: 1.72–2.75) than those who were prefrail (HR = 1.51, 95%CI: 1.37–1.67), with a significant difference across frailty levels (Q = 7.88, p < .01). Regarding cause-specific mortality, the risk of death from non–lifestyle-related diseases (e.g., dementia, cancer) (HR = 1.97, 95%CI: 1.36–2.84) was slightly higher than that from lifestyle-related diseases (e.g., cardiovascular and respiratory diseases) (HR = 1.94, 95%CI: 1.52–2.46), though the difference was not statistically significant (Q = 0.008, p > .05).

Conclusions

These findings highlight the need for comprehensive assessment and management of frailty in community-dwelling older adults. Future research and health policy should prioritize targeted strategies according to frailty severity to reduce mortality risk and improve health outcomes.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-26082-w.

Introduction

The rapid aging of the global population has profound implications for geriatric care, particularly concerning frailty, a concept widely acknowledged in gerontological research. Frailty is a biological syndrome commonly associated with disability [1], comorbidity [2], and aging-related decline [3]. It reflects a state of reduced physiological reserve and resilience, leading to diminished capacity to cope with stressors [4–8]. As multiple physiological systems deteriorate, frailty increases vulnerability and is associated with higher risks of disability, hospitalization, and mortality [9].

Frailty is highly prevalent among older adults. In the United States, approximately 70% of older adults are classified as non-robust, including 15% considered frail and 55% prefrail, with prevalence rising steeply with age [10]. In the United Kingdom, more than 43.7% of adults aged 65 and older are identified as frail [11], while a meta-analysis conducted in Asia estimated a prevalence of 20.5% among community-dwelling older adults [12].

Although previous studies have suggested an association between frailty and mortality, many recent meta-analyses have focused on individual frailty instruments or specific domains [13–17], resulting in fragmented evidence and limited comparability across studies. Additionally, some primary studies have classified frailty into multiple levels—such as mild, moderate, or severe [18, 19]—or used hierarchical categories such as first-, second-, or third-class frailty [20]. However, no existing meta-analysis has comprehensively synthesized these diverse classifications into a unified framework.

To address this gap, the present study adopts an integrated conceptualization of “non-robust” status, encompassing a wide spectrum of frailty levels and classifications. This approach facilitates synthesis across heterogeneous definitions and enhances the applicability of findings for both clinical and public health planning. Moreover, given the steady increase in large-scale cohort studies published in the past decade, an updated meta-analysis is warranted to reflect contemporary evidence and improve generalizability across global populations.

Therefore, this study aimed to conduct a systematic review and meta-analysis of international prospective cohort studies to assess the association between frailty—defined as non-robust status—and both all-cause and cause-specific mortality among community-dwelling older adults. A secondary objective was to explore how this association varies across different frailty domains and assessment models. We hypothesized that individuals classified as non-robust or frail would exhibit significantly elevated mortality risks, regardless of the domain or measurement method used.

Methods

Search strategy and selection criteria

This systematic review and meta-analysis was registered in the PROSPERO (CRD42023420948) and conducted in accordance with the Cochrane Handbook for Systematic Reviews of Interventions [21]. It also followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [22].

Published studies examining the association between frailty and all-cause or cause-specific mortality were identified and independently screened by two authors (LYL and FX). Discrepancies were resolved through discussion, and a third reviewer (SR) was consulted when necessary. A comprehensive literature search was conducted in PubMed, Web of Science, Cochrane Library, CINAHL, and EMBASE databases. Searches were limited to original research articles published in English, from inception to April 29, 2025 (see Supplementary Tables 1–5). Search strategies included both free-text terms and Medical Subject Headings (MeSH), as detailed in Supplementary Table 6. Additionally, reference lists of relevant articles and review papers were manually screened to identify further eligible studies.

The inclusion criteria for this study encompassed prospective cohort studies that evaluated frailty as a distinct predictor of mortality in community-dwelling older adults, independent of baseline combined impact factors, and correlated with all-cause or cause-specific mortality. According to the World Health Organization (WHO), older adults were primarily defined as individuals aged 60 and above [23]. Research assessing various types or domains of frailty were included, and in cases where the same cohort of older populations was used in multiple publications, the selection of the study was determined based on the largest sample size or longest follow-up.

Exclusion criteria were as follows: [1] studies that did not report Hazard Ratios (HR), Risk Ratios (RR), or Odds Ratios (OR) with 95% confidence intervals (CI) for mortality; [2] prospective cohort studies not published in peer-reviewed journals or not written in English; and [3] grey literature, including unpublished theses, reports, or non-peer-reviewed sources, was excluded to ensure the methodological quality and reproducibility of findings.

Definitions of main outcome

Non-Robust and robust

Given the heterogeneity in frailty definitions across the included studies, we used the term ‘non-robust’ to describe individuals classified as frail, prefrail, moderately frail, severely frail, or any other equivalent indicator of compromised frailty status. This operational definition allowed us to unify diverse frailty levels and domains into a single analytical group. In contrast, individuals labeled as ‘robust’ were those reported to have no signs of frailty and to maintain adequate functional capacity across assessments.

Frail and prefrail

These terms are commonly used within specific frailty models, such as the Cardiovascular Health Study (CHS) physical frailty phenotype [24]. They typically indicate varying stages of frailty severity.

Data analysis

Using a standardized data-extraction form, two authors (LYL and FX) independently extracted data from all eligible articles. The quality and risk of bias of the included studies were assessed using the Risk of Bias In Non-randomized Studies of Exposures (ROBINS-E) tool [25] by two authors (LYL and SJS), and a third author (SR) resolved any discrepancies (See Supplementary Sect. 3).

Meta-analyses were performed using Comprehensive Meta-Analysis (CMA V4.0, Biostat, USA). A two-sided p-value of < 0.05 was considered statistically significant [26], and random-effects models were applied to account for between-study heterogeneity [27]. Hazard ratios (HR) and 95% confidence intervals (CI) were used to estimate the pooled effect size, with risk ratios (RR) treated as equivalent to HR. When only odds ratios (OR) were available, they were converted to RR using the formula: RR = OR/[(1-P₀)+(P₀*OR)], where P₀ represents the mortality rate of the robust group [28]. For a study that reported ORs stratified by gender but not for the overall population [29], a fixed-effects model was used to calculate a pooled OR representing the total group.

We first compared the HR of all-cause mortality between non-robust and robust older adults. Additionally, cause-specific mortality was examined by classifying diseases according to their established association with modifiable behavioral and lifestyle factors. Specifically, diseases such as cerebrovascular disease [30], heart disease [31], cardiovascular disease [30], and respiratory disease [32] were categorized as lifestyle-related. Dementia [33] and cancer [34] were considered as non-lifestyle-related diseases.

To further explore heterogeneity, subgroup analyses were conducted based on specific frailty domains (e.g., physical and multidimensional) and frailty assessment models. Frailty instruments were categorized into phenotype-based models (e.g., CHS and its variations). Instruments that could not be clearly classified into this category were excluded from the analysis due to insufficient data for meaningful synthesis. Although some instruments aligned with deficit accumulation-based models (e.g., Frailty Index and its variations), a separate subgroup analysis was not performed for this category. This decision was based on the considerable conceptual and methodological overlap between deficit accumulation-based tools and those already classified under the multidimensional domain, which would have made an additional subgroup redundant and potentially misleading.

Statistical heterogeneity across studies was assessed using Tau-squared (τ²), Cochran’s Q test, and the I² statistic. A Q test with p <.10 was considered indicative of significant heterogeneity, while I² values of 25%, 50%, and 75% were interpreted as low, moderate, and substantial heterogeneity, respectively [35]. The prediction interval (PI) was also reported to reflect the expected range of true effects in future studies.

Meta-analyses were conducted only when five or more datasets were available for a given subgroup. For subgroups with fewer than five datasets, a narrative synthesis was performed to qualitatively summarize the evidence. Publication bias was evaluated using Egger’s regression test by plotting the logarithm of hazard ratios (log HRs) against their standard errors. Funnel plots were generated when ten or more studies were available; for subgroups with fewer than ten studies, Egger’s test results were reported in the text instead. The trim-and-fill method was applied when applicable to estimate the potential impact of unpublished studies and to adjust for funnel plot asymmetry. Adjusted effect sizes derived from this method are presented in the Supplementary Materials (Sect. 5). Sensitivity analyses were conducted using a leave-one-out approach, in which each study was sequentially omitted to assess its influence on the overall pooled estimate.

Results

Literature search results

Our initial review identified 3,054 publications, with four additional publications added after manually searching in-text citations (Fig. 1). A total of 59 articles met the inclusion criteria [18–20, 24, 29, 36–89] and were included in the qualitative analysis; of these, 53 were eligible for meta-analysis. The remaining six studies were excluded due to asymmetric confidence intervals (n = 1), lack of mortality data for the robust group (n = 1), or failure to adjust for confounders (n = 4) (see Supplementary Sect. 4).

Fig. 1.

Flow diagram of the study selection process

Altogether, the 59 studies encompassed 185,355 participants and reported over 28,616 deaths, with follow-up durations ranging from 1 to 15 years. The quality assessment showed that 9 were low risk (15.3%), 6 had some concerns (10.2%), 39 were high risk (66.0%), and 5 were considered very high risk (8.5%) (See Supplementary Table 7). The primary characteristics of the publications included in this study are summarized in Table 1, and detailed risk of bias assessments using the ROBINS-E tool are available in the Supplementary file (Sect. 3, Table 7).

Table 1.

The characteristics of the included studies

Study	location	sample size	Age, y, Mean ± SD (range)/ Median (IQR)	Frailty assessment	domain	criteria (cutoff points)	outcome assessment	mortality (number)	follow-up period	adjusted variables
Xue et al. (2021)[53]	USA	2557	≥ 65 73.2 ± 6.9	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	interview	all 325	8y	age, race, sex, edu, baseline ds, incident ds
Shin et al. (2021)[55]	Korea	2923	≥ 70 76.0 ± 3.9	KFS	multi	robust (0) prefrail (1–2) frail (≥ 3)	NI	all 38	4y	age, sex, edu, income, BMI, smoking, alcohol, PA, marital, employ, residence
Bartosch et al. (2018)[75]	Sweden	1044	75	OPRA-specific Frailty Index	physiological	low frailty Q1 frailty Q2 frailty Q3 highly frail Q4	Swedish National Population Register	all 221	10y	NI
Fried et al. (2001)[24]	USA	4735	65–101	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	interview	all 864	8y	age, sex, indicator for minority cohort, income, smoking, BP, FBG, albumin, Cr, CS, CHF, cognition, ECG abnormality, diuretics, IADL, health, depression.
Beauchet et al. (2022)[52]	Canada	1504	74.4 ± 4.2 (67–84)	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	interview	all 62	3y	NI
				CARE frailty e-health scale	multi	robust (0–1) prefrail (2–4) frail (> 5)
Langholz et al. (2018)[72]	Norway	712	≥ 70 77.4 ± 2.4	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	death registry	all 501	mean 10.1y	age
John et al. (2016)[77]	Canada	1751	≥ 65 77.5 ± 7.1	FI-40	multi	robust (< 0.25) frail (> 0.25)	interview, death certificates, admin records	all 417	5y	age, sex, edu
Cawthon et al. (2007)[89]	USA	5969	≥ 65	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	interview, death certificates	all 669	mean 4.7y	age, functional limitation, DM, PD, IHD, CHF, COPD, cancer, balance, health, smoking, cognition
Jacobs et al. (2011)[86]	Israel	840	85	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	death certificate	all 194	5y	sex, cognition, edu, IHD, DM, HTN, smoking, ADL, health
Josep et al. (2013)[84]	Spain	875	≥ 74 81.7 ± 4.8	Specified PF	physical	frail (≥ 4)	death registry	all 52	4y (mean 3.6 y)	sex, age, marital
				Specified MF	cognitive	frail (≥ 2)
				Specified SF	social	frail (≥ 2)
Jung et al. (2014)[83]	Korea	693	≥ 65 75.9 ± 8.9	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	death registry	all 97	mean 5.57y	age, Hb, Chol, Cr, ESR, ALT, ALP
				SOF-FI	physical	robust (0) prefrail (1) frail (2–3)
				KLoSHA-FI	multi	prefrail (≥ 0.2) frail (≥ 0.35)
Juan et al. (2014)[82]	Cuba	2813	≥ 65	CHS-PFP + cognitive deterioration	multi	frail (≥ 3)	NI	all 608	mean 4.1y	age, sex, edu
								Dementia 131; CVD 448; DM 129; depression NA
Roman et al. (2014)[29]	UK	4309	≥ 75	SHARE-FI	physical	robust (0) prefrail (1–2) frail (≥ 3)	NI	all 297	2y	age, chronic ds, BADL
				SHARE-FI 75+
Anna et al. (2016)[80]	Russia	306	≥ 75	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	death registry	all 120	5y	sex, age, comorbidities
Crow et al. (2018)[74]	USA	4984	≥ 60 71.1 ± 0.19	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	NDI	all 1901	median 98.5 m	age, sex, race, edu, smoking, DM, heart failure, cancer, CAD, arthritis
							ICD code	CVD 521
Tanaka et al. (2023)[41]	Japan	2031	≥ 65 73.1 ± 5.6	OF-5	oral	robust (0) prefrail (1–2) frail (≥ 3)	LTCI	all 137	9y	age, sex, BMI, edu, living alone, income, cognition, exercise, daily food diversity, alcohol, smoking, chronic ds
				CHS-PFP	physical	frail (≥ 3)	LTCI	all 15
Yamada et al. (2018)[70]	Japan	6603	≥ 65 75.2 ± 6.6	SFSI	social	robust (0) prefrail (1) frail (≥ 2)	ICD code	all 2283	6y	age, sex, BMI, medication, comorbidities
				FSI	multi	robust (0) prefrail (1–2) frail (≥ 3)
Yuki et al. (2018)[69]	Japan	841	65–88	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	population dynamics survey	all 113	mean 7.9y	age, sex, BMI, edu, PA, TCI, alcohol, smoking, income, CESD, MMSE, comorbidities
							ICD code	cancer 45		age, sex
Hao et al. (2019)[68]	China	736	≥ 90 93.6 ± 3.4	FI-22 lab variables	multi	frail (>0.21)	interview	all 394	4y	age, sex, edu, smoking, alcohol, PA, HTN, CVD, CeVD, DM, RD, DD, CRD, OA
Shi et al. (2019)[65]	China	1788	≥ 70 75.4 ± 3.9	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	death registry	all 149	3y	age, sex, marital, edu, smoking, alcohol, BMI, HTN, DM, and MCI
				FI-45	multi	robust (≤ 0.1) prefrail (0.1–0.21) frail (>0.21)
Thompson et al. (2019)[64]	Australia	909	≥ 65 74.4 ± 6.2	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	official death records	all 292	10y	age, sex, edu, income
				FI-34	multi	robust (≤ 0.1) prefrail (0.1–0.21) frail (>0.21)
Wang et al. (2019)[63]	Taiwan	921	65–99	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	death registry	all 160	mean 6.62y	age, sex, edu, marital, BMI, smoking, alcohol, PA, exercise, HTN, DM, heart ds, HLD, gout, HUA, arthritis, osteoporosis, stroke, cataract, falls, sleep, cognition
Sachs et al. (2021)[56]	Canada	146	≥ 90 93.7 ± 2.7	Self-rated CFS	multi	Very fit (1–2) Well (3) apparently vulnerable (4) mildly-severely frail (5–7)	interview	all 59	3y	age
Andrew et al. (2008)[88]	Canada	3707	≥ 65 mean 77.9	SVI-40	social	NA	interview	all 930	5y	age, sex
		2648	≥ 65 mean 73.4	SVI-23				all 761	8y
Graham et al. (2009)[87]	USA	1996	≥ 65 74.5 ± 6.1	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	interview and NDI	all 892	10y	age, male, marital, BMI, smoking, MI, HTN, cancer, hip fracture, DM, ADL, IADL, Depression, cognition, health
Satake et al. (2017)[78]	Japan	5542	≥ 65 M 72.1 ± 5.7 F 72.5 ± 6.1	t-KCL	multi	robust (0–3) pre-frail (4–7) frail (≥ 8)	death registry	all 170	3y	age, sex
Adabag et al. (2018)[76]	USA	3135	≥ 65 76.4 ± 5.6	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	interview	all 828	mean 9.2y	site, age, race
							ICD code	CVD 445
Diniz et al. (2018)[18]	Brazil	515	≥ 65 75.4 ± 7.3	EFS	multi	robust (0–4) vulnerable (5–6) mild frailty (7–8) moderate frailty (9–10)	death certificates, interview, SIM	all 127	mean 5.6y	sex, age, marital, number of ds
Liu et al. (2017)[71]	Taiwan	678	≥ 65 73.3 ± 5.3	Specified CF	cognitive	dynapenia plus cognitive impairment in any domain	NI	all NI	median 28 m	age, sex
Lee et al. (2020)[112]	Korea	11,241	≥ 65 72.9 ± 6.7	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	interview	all 735	3y	age, sex, smoking, alcohol, comorbidity, depression, cognition
Lohman et al. (2020)[61]	USA	10,490	≥ 65	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	NDI	all 2148	7y	age, race, gender, edu, marital, cognition, smoking, chronic ds, ADL, heart ds, cancer, DM, RD, CeVD, stroke
							ICD code	Heart ds 626 cancer 490 RD 265 Dementia 131 CeVD 112
Salminen et al. (2020)[60]	Finland	1152	≥ 64 72.7 ± 6.2	FS	physical	robust (0) prefrail (1–2) frail (≥ 3)	death registry	all 776	18y	age and sex
				FI-36	multi	robust (≤ 0.08) prefrail (0.08–0.25) frail (≥ 0.25)
Gobbens et al. (2021)[58]	Netherlands	479	≥ 75 80.3 ± 3.8	TFI	multi	total (5) physical (3) psychological 2) social (2)	death registry	all 162	7y	age and sex
Strandberg et al. (2021)[54]	Finland	2286	≥ 66	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	death registry	all 920	18y	age, BMI, smoking, memory disturbance, mental health, ADL, regular drug use
Verghese et al. (2021)[19]	USA	681	≥ 65 74.6 ± 6.1	FI-41	multi	relatively stable mild frail moderate frail severly frail	interview, NDI	all 57	12y	age, sex, edu, comorbidities
		515	≥ 70 79.3 ± 5.1					all 82	11y
Diniz et al. (2022)[51]	Brazil	1340	≥ 60	FI-41	multi	robust (≤ 0.2) prefrail (0.20–0.35) frail (> 0.35)	interview, death certificates	all NA	10y	age, health, cognition, sex, edu, and serum IL-6
Li et al. (2022)[50]	Germany	3825	≥ 65 68.9 ± 2.9	FI-30	multi	non-frail (≤ 0.11) prefrail (0.11–0.35) frail (≥ 0.35)	death registry	all NA	14y	age, sex, alcohol, smoking
							ICD code	cancer NA CVD NA
Morkphrom et al. (2022)[49]	Thailand	8195	≥ 60 69.2 ± 6.8	30-item TFI	multi	fit (≤ 0.10) Pre-frail (0.10–0.25) Mildly frail (0.25–0.45) Severely frail (> 0.45)	death registry	all1284	7y	age, smoking, income
Sun et al. (2023)[42]	China	359	≥ 65	SFI	social	non-frail (0–1) frail (2)	interview, local government records	all 63	6y	NI
Albala et al. (2017)[79]	Chile	1946	≥ 60	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	death certificate	all NA	15y	age, sex, MCI, dementia
Aguilar-Navarr et al. (2014)[81]	Mexico	5644	≥ 60 68.7 ± 6.9	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	interview	all 1807	11y	sex, edu, smoking, alcohol, chronic ds, health, depression, cognition, and three types of disability
MA et al. (2019)[67]	China	1724	≥ 60	FSQ	physical	robust (0) prefrail (1–2) frail (≥ 3)	interview	all 573	8y	age, sex
Cella et al. (2021)[111]	Italy	407	≥ 65 77.9 ± 4.5	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	administrative data	all 53	5y	age, sex
				MPI	multi	MPI-1 < 0.33 MPI-2 0.33–0.66 MPI-3 > 0.66
Rath et al. (2021)[57]	India	834	≥ 60	EFS	multi	frail (> 6)	NI	all 53	30 m	age, sex, income, chronic ds, smoking, alcohol
Rosas et al. (2022)[48]	Brazil	689	≥ 65	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	NI	all 80	8y	SES, multimorbidity, depression, insomnia, stress
Daniels et al. (2012)[85]	Netherlands	532	≥ 70	GFI	multi	frail (≥ 4)	death registry	all 15	1y	sex, age, income, edu, disability
				TFI		frail (≥ 5)
				SPQ		frail (≥ 2)
Gilardi et al. (2018)[73]	Italy	1335	≥ 65	FGE	multi	robust (> 49) prefrail (11–49) frail (< 11)	death registry	all 52	1y	age, sex, comorbidity
Watanabe et al. (2022)[47]	Japan	10,276	≥ 65 73.9 ± 6.8	FSI	physical	frail (≥ 3)	death registry	all 1257	median 5.3y	age, sex, population density, living alone, SES, edu, smoking, alcohol, sleep, medication use, chronic ds
				KCL	multi	frail (≥ 7)
Ouvrard et al. (2019)[66]	France	1586	≥ 65 74.1 ± 4.8	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	death registry	all 665	14y	age, sex, dependency, comorbidities, PSPS, frailty status
Carolina et al. (2023)[43]	Costa Rica	1790	≥ 65 72.65 (95CI: 72.37–72.95)	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	Costa Rican Death Index	all 661	8y	frailty, age, sex, edu, income, living in a metropolitan zone, marital, smoking
	USA	5936	≥ 65 74.55 (95CI: 74.36–74.73)	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	interview	all 1670		frailty, age, sex, edu, income, living in a metropolitan zone, marital, smoking, race, interactions between race and frailty
Li et al. (2023)[44]	Taiwan	1904	≥ 65	FS	physical	robust (0) prefrail (1–2) frail (≥ 3)	National Register of Deaths	all 239	5y	age, sex, edu, marital, smoking, PA, depression, falls, hospitalization, emergency department visits during the past year
Ekram et al. (2023)[46]	Australia & USA	19,114	≥ 65 in USA ≥ 70 in Australia 74.0 ± 6.1	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	clinical notes, hospitalization records	all 203	median 4.7y	age, gender, ethno-racial origin, smoking history, hypertension, diabetes mellitus and dyslipidemia
				FI-66	multi	non-frail (≤ 0.10) prefrail (0.10–0.21) frail (> 0.21)
Kawai et al. (2024)[45]	Japan	3381	72.7 ± 5.5 (65–85)	TMIG-IC	multi	1. Physical/psychological frailty 2. Psychological frailty 3. Social frailty	database overseen by the ward office	all 91	8y	sex, age, chronic ds, living alone, perceived financial status
Molina et al. (2024)[39]	Chile	1174	≥ 65	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	death certificates	all 147	5y	age, sex, MMSE, belonging to groups, socio-economic status
Shi et al. (2024)[38]	China	1197	≥ 60 74.8 ± 8.6 (60–101)	FI-36	multi	frailty (> 0.2)	family members, local neighborhood committees, or local public security agencies	all 443	11y	NI
Stolz et al. (2024)[37]	Austria	2561	≥ 65 72.1 ± 5.8	FI-41	multi	non-frail (≤ 0.10) prefrail (0.10–0.21) frail (> 0.21)	ICD code	all 405 CVD 148 cancer 127	8y	sex, living alone, edu
Wu et al. (2024)[36]	China	5300	≥ 60	CHS-PFP	physical	robust (0) prefrail (1–2) frail (≥ 3)	exit interviews	all 846	median 6.22y	age, sex, edu, urban residence, marital status
Yang et al. (2025)[20]	Japan	609	≥ 65 74.93 ± 7.64	KCL	multi	1 class: high risk of cognitive impairment 2 class: moderate risk of cognitive, physical, and oral dysfunction 3 class: high risk of cognitive, physical, and functional decline	Local government office	all 52	3y	age, sex, smoking, alcohol consumption, physical activity, chronic ds
Zhao et al. (2023)[40]	China	8642	≥ 60 85.6 ± 11.3	FI-42	multi	non-frailty (< 0.25) frailty (≥ 0.25)	NI	all 4458	7y	NI

NI No information, y years, m months, all all-cause mortality, CVD cardiovascular disease, CHS-PFP Cardiovascular Health Study physical frailty phenotype, KFS Korean Frailty Scale, FI-40 Frailty Index-40 accumulation of deficits, SOF-FI Study of osteoporotic fracture frailty index, KLoSHA-FI Korean Longitudinal Study on Health and Aging Frailty Index, SHARE-FI Frailty Instrument for Primary Care of the Survey of Health, Ageing and Retirement in Europe, SHARE-FI75+ Frailty Instrument for primary care for those aged 75 years or more: findings from the Survey of Health, Ageing and Retirement in Europe, a longitudinal population-based cohort study, OF-5 oral frailty five-item checklist, SFSI Social Frailty Screening Index, FSI Frailty Screening Index, FI-22 lab variables Frailty Index-22 laboratory variables, FI-45 Frailty Index-45 accumulation of deficits, FI-34 Frailty Index-34 accumulation of deficits, FI-36 Frailty Index-36 accumulation of deficits, FI-32 Frailty Index-32 accumulation of deficits, FI-41 Frailty Index-41 accumulation of deficits, FI-30 Frailty Index-30 accumulation of deficits, Self-rated CFS Self-rated Clinical Frailty Scale, SVI-40 social vulnerability index-40 items, SVI-23 social vulnerability index-23 items, t-KCL total Kihon Checklist, EFS Edmonton Frail Scale, Study-Specified CF Study-Specified Cognitive Frailty, FS FRAIL scale, TFI Tilburg Frailty Indicator, 30-item TFI 30 items Thai Frailty Index, SFI social frailty index, FSQ Frailty Screening Questionnaire, MPI Multidimensional Prognostic Index, GFI Groningen Frailty Indicator, SPQ Sherbrooke Postal Questionnaire, FGE functional geriatric evaluation, FSI self-administered frailty screening index, KCL self-administered Kihon Checklist, Specified PF Study-Specified physical frailty, Specified MF Study-Specified mental frailty, Specified SF Study-Specified social frailty, edu education, ds diseases, BMI body mass index, PA physical activity, BP Blood Pressure, FBG Fasting Blood Glucose, Cr Creatinine, CS Carotid Stenosis, CHF Congestive heart failure, ECG abnormality Electrocardiographic abnormality, IADL Instrumental Activities of Daily Living, DM diabetes mellitus, PD Parkinson’s Disease, IHD Ischemic heart disease, CHF congestive heart failure, COPD chronic obstructive pulmonary disease, HTN Hypertension, ADL Activity of daily living, Hb Hemoglobin, Chol Cholesterol, ESR Erythrocyte sedimentation rate, ALT Alanine aminotransferase, ALP Alkaline phosphatase, BADL Basic Activities of Daily Living, CAD coronary artery disease, TCI total caloric intake, CESD Center for Epidemiological Studies Depression, MMSE Mini-Mental State Examination, CeVD Cerebrovascular Disease, RD Respiratory Disease, DD Digestive Disease, CRD Chronic Renal Disease, OA Osteoarthritis, MCI mild cognitive impairment, HLD Hyperlipidemia, HUA Hyperuricemia, MI Myocardial Infarction, Serum IL-6 Serum Interleukin-6, SES Socioeconomic status, PSPS Psych-Socioeconomic Precarious Score, ICD code International Classification of Disease codes, NDI National Death Index, SIM Brazil Mortality Information System, LTCI long-term care insurance system, IQR interquartile range, SE stand error, TMIG-IC Tokyo Metropolitan Institute of Gerontology Index of Competence

Primary results: all-cause mortality in the non-robust group

A meta-analysis of 56 studies (see Supplementary Sect. 4 for rationale) using a random-effects model revealed that older adults classified as non-robust had a significantly higher risk of all-cause mortality compared to those classified as robust (pooled HR = 1.83, 95% CI: 1.70–1.97; 95% PI: 1.12–2.99; Fig. 2). Given the substantial heterogeneity (I² = 97.3%, Q = 2008.95, p <.001; τ² = 0.059), we further evaluated publication bias using funnel plots (Fig. 3). Potential publication bias was indicated by the Egger regression intercept of 5.23 (p <.001), showing an asymmetric pattern suggesting publication bias.

Fig. 2.

Pooled effect of the all-cause mortality in the non-robust group

Fig. 3.

Funnel plot: Pooled effect of the all-cause mortality in the non-robust group

Sensitivity analysis demonstrated consistent results when omitting any single study at a time, with no change in the overall pooled risk estimate. Given that more than half of the studies included were rated as high risk of bias, we conducted a further sensitivity analysis to examine the robustness of results across risk-of-bias levels. No significant difference was found between studies rated as having ‘low risk of bias’ and those with ‘some concerns’ (Q = 0.04, p =.84). However, comparisons between studies with ‘low risk of bias’ and those with ‘high risk of bias’ revealed a statistically significant difference (Q = 9.04, p <.01). Nonetheless, the association between non-robust status and all-cause mortality remained significant even after excluding studies classified as ‘high risk of bias’ (HR = 1.46, 95% CI: 1.27–1.68; PI = 0.84–2.55).

Primary results: cause-specific mortality in the non-robust group

A meta-analysis using a random-effects model showed that the non-robust group had a significantly higher risk of cause-specific mortality compared to the robust group (pooled HR = 1.95, 95% CI: 1.59–2.38; 95% PI: 0.91–4.18; Fig. 4). Sensitivity analyses indicated that the overall results were not substantially influenced by the exclusion of any single study.

Fig. 4.

Pooled effect of the cause-specific mortality in the non-robust group

Further pairwise comparisons revealed that the risk of mortality from non-lifestyle diseases (pooled HR = 1.97, 95% CI: 1.37–2.85, PI = 0.56–6.98) was slightly higher than that from lifestyle diseases (pooled HR = 1.94, 95% CI: 1.52–2.46, PI = 0.83–4.52). However, the difference in effect sizes between the two categories of cause-specific mortality was not statistically significant (Q = 0.008 p =.93). Considerable heterogeneity was observed in both comparisons (non-lifestyle diseases: I² = 92.27%, Q = 64.69, p <.001, τ² = 0.172; lifestyle diseases: I² = 92.18%, Q = 89.54, p <.001, τ² = 0.105). Egger’s test results (p =.54 and p =.62, respectively) suggested that publication bias was unlikely.

Subgroup analysis of frailty domains

A total of five frailty domains were identified: physical frailty (n = 35), multidimensional frailty (n = 26), social frailty (n = 4), cognitive frailty (n = 2), and oral frailty (n = 1), and in accordance with the recommendations of the Cochrane Handbook [21], meta-analyses were performed for domains with more than five datasets available, namely physical frailty and multidimensional frailty. For the remaining domains, a narrative synthesis was conducted.

For physical frailty, the pooled HR was 1.74 (95% CI: 1.53–1.99, PI = 0.82–3.72; Fig. 5a), indicating a statistically significant association. However, heterogeneity was substantial (I² = 96.17%, Q = 887.827, p <.001, τ²=0.134). Visual inspection of the funnel plot (Fig. 6a) revealed asymmetry, and Egger’s regression intercept was 4.51 (p <.001), suggesting the presence of publication bias. Sensitivity analysis showed that the overall pooled estimate remained stable after excluding any single study.

Fig. 5.

forest plot of physical/multidimensional frailty on all-cause mortality in the non-robust group

Fig. 6.

Funnel plot: Pooled effect of physical/multidimensional frailty on all-cause mortality in the non-robust group

Similarly, multidimensional frailty was significantly associated with mortality, with a pooled HR of 2.25 (95% CI: 1.97–2.57, PI = 1.12–4.13, p <.001; Fig. 5b). Heterogeneity was also high (I² = 87.77%, Q = 204.407, p <.001, τ²=0.082). Egger’s test indicated potential publication bias (intercept = 2.01, p <.05), supported by asymmetry in the funnel plot (Fig. 6b). Sensitivity analysis again confirmed that exclusion of any single study did not materially alter the pooled estimate.

For social frailty, 3 studies reported statistically significant positive associations [58, 70, 84], with HR ranging from 1.04 to 1.27 and confidence intervals that did not cross 1. The remaining two subgroups from the same dataset reported elevated HR = 1.44 [88], but their confidence intervals crossed unity, rendering the associations statistically non-significant.

For cognitive frailty, one study reported a statistically significant association, with a hazard ratio of 3.18 (95% CI: 1.09–9.25), using a mental frailty phenotype approach [84]. In contrast, another study that defined cognitive frailty based on a cognitive screening-based frailty index did not find a significant association (HR = 1.23, 95% CI: 0.64–2.36) [71].

Oral frailty was examined in a single study [41], which reported a statistically significant association with mortality (HR = 1.37, 95% CI: 1.15–1.63, p <.001).

Subgroup analysis based on phenotype-based models

Frailty was assessed using phenotype-based models (CHS and its variations) in 29 studies. Frailty defined by the CHS physical frailty phenotype was significantly associated with increased all-cause mortality risk in the non-robust group compared to the robust group. The pooled HR for all-cause mortality in the non-robust group was 1.71 (95% CI: 1.48–1.97, PI = 0.81–3.60; Fig. 7a).

Fig. 7.

forest plot of phenotype-based models subgroup analysis

Further subgroup analyses comparing the prefrail and frail groups showed that the frail group had a significantly higher all-cause mortality risk (pooled HR = 2.18; 95% CI: 1.72–2.75; PI: 0.67–7.08; Fig. 7b) than the prefrail group (pooled HR = 1.51; 95% CI: 1.37–1.67; PI: 1.00–2.29; Fig. 7b), both compared to the robust group. The difference in effect sizes according to frailty severity (prefrail vs. frail) was statistically significant (Q = 7.88, p <.01).

Heterogeneity was substantial for both the non-robust (I² = 95.64%, Q = 642.865, p <.001, τ²=0.126) and frail groups (I² = 95.59%, Q = 567.125, p <.001, τ²=0.312), and moderate for the prefrail group (I² = 72.35%, Q = 75.960, p <.001, τ²=0.037).

Publication bias was not evident in the prefrail group, as suggested by the funnel plot (Fig. 8b) and Egger’s test (intercept = 0.59, p =.58). In contrast, potential publication bias was identified in the non-robust (Fig. 8a) and frail groups (Fig. 8c), with Egger’s intercept values of 4.11 (p <.001) and 4.10 (p <.001), respectively.

Fig. 8.

Funnel plot: Pooled effect of phenotype-based models’ frailty on all-cause mortality

Sensitivity analyses confirmed the robustness of these findings, as the exclusion of any single study did not meaningfully change the overall pooled estimates.

Discussion

To our knowledge, this is the first study to examine the relationship between frailty and both all-cause and cause-specific mortality among community-dwelling older adults using a macro-level conceptualization of ‘non-robust’ status. Rather than relying solely on isolated frailty tools or domains, our approach emphasizes the need for a broader and more integrative understanding of frailty. We found that older adults classified as non-robust had a significantly higher risk of all-cause mortality compared to their robust counterparts. This association remained significant even after sensitivity analyses that excluded studies deemed at high risk of bias. Although substantial heterogeneity was observed across studies, we addressed this by applying a random-effects model. The high heterogeneity likely reflects variations in frailty assessment tools, population characteristics, and follow-up durations across studies, rather than undermining the validity of the overall association. Therefore, while the findings should be interpreted with caution, they provide robust and meaningful insights based on sound methodological rigor.

The concept of frailty represents a comprehensive paradigm encompassing aspects of biology, psychology, sociology, and a variety of body functions [90, 91]. In this study, the category of ‘non-robust’ has been unified to capture the multifaceted and multidimensional nature of frailty. By avoiding narrow definitions of frailty that may overlook certain individuals, this approach provides a more accurate picture of the connection between frailty and mortality. In addition, our study confirms that frailty cannot be measured and defined by a gold standard. Frailty is perceived subjectively by older adults [92, 93], despite the objectivity of the tools used to measure frailty [94–96]. When a single measurement tool is used, it could result in a narrow perspective, may overlook individuals who are functionally ‘frail’ and in need of assistance. The multidimensional and dynamic aspects of frailty warrant further research.

We found that mortality from non–lifestyle-related diseases was slightly higher among non-robust older adults than from lifestyle-related diseases. In this review, lifestyle-related diseases were defined as chronic conditions strongly influenced by behavioral risk factors (e.g., smoking, diet, and physical inactivity) [97, 98], while non-lifestyle-related diseases referred to conditions less amenable to prevention through lifestyle modification (e.g., dementia and certain cancers) [33, 34]. Although both categories include non-communicable diseases, the distinction reflects differences in preventability and intervention potential.

It has been suggested that non-lifestyle-related diseases, such as genetic disorders, environmental exposures, and age-related conditions, may pose a slightly greater mortality risk [99, 100]. This may be because they are less easily modified and are more difficult to detect or manage effectively in healthcare settings [101, 102]. Nevertheless, lifestyle-related diseases also contribute substantially to mortality in older adults [103]. These findings emphasize the importance of public health strategies aimed at reducing modifiable risk factors through interventions such as dietary improvement, physical activity, and preventive health screening. These approaches may help delay the onset or progression of frailty, ultimately improving survival outcomes and quality of life among community-dwelling older adults [104, 105].

A previous meta-analysis [13] demonstrated that physical frailty, as assessed by the frailty phenotype [24], was significantly associated with higher all-cause mortality. Our study supports this association and extends the evidence by adopting a broader conceptualization of frailty, referred to as non-robust status, which integrates multiple physical frailty tools. In addition, we also conducted subgroup analyses on physical frailty phenotype models and their variations, which confirmed the elevated mortality risk among physically frail individuals and aligned with previous findings. Physical frailty typically involves a decline in muscle strength, mobility, balance, coordination, and the ability to see obstacles [106]. These factors may heighten the risk of falls, which is one of the leading causes of death among older adults [107, 108]. Falling does not only result in physical injuries but also has a significant impact on the lives of autonomous and independent community-living older adults by limiting their mobility and self-care abilities [109], which in turn increases their mortality risk [110].

Our study found that the pooled HR for multidimensional frailty was significantly higher than that for other domains (e.g. physical domain), indicating a strong association between multidimensional frailty and mortality. This finding is consistent with previous studies, suggesting that multidimensional frailty, which captures various domains of frailty, may serve as a more robust predictor of mortality than assessments based on a single frailty dimension [111]. Accordingly, comprehensive geriatric assessments are recommended for the identification and management of frailty in older adults [112, 113].

Our findings offer additional insights into underexplored frailty domains. For social frailty, the observed associations with mortality were generally modest but statistically significant in three subgroups, with hazard ratios ranging from 1.04 to 1.27. These results support prior suggestions that social disconnection and diminished social roles may adversely affect health trajectories in older adults [114, 115]. However, the presence of non-significant findings within subgroups highlights the variability in measurement, underscoring the need for more consistent operational definitions in social frailty assessment.

Regarding cognitive frailty, mixed findings emerged. While one study using a mental frailty phenotype reported a strong and significant association with mortality, another study using a cognitive screening-based frailty index did not find a significant association. This discrepancy may reflect differences in operational definitions and highlights the conceptual heterogeneity of cognitive frailty. These inconsistencies underscore the need for standardized criteria and longitudinal validation across diverse populations.

Oral frailty, examined in a single study, was found to be significantly associated with increased mortality. Although this result aligns with hypotheses linking poor oral health to systemic inflammation, reduced nutritional intake, and functional decline [116], the evidence remains limited. More studies are required to confirm this association and to explore potential mediating pathways.

The findings of this study should be interpreted with caution due to several limitations. First, frailty and aging are dynamic processes that may be modified through interventions such as medical treatment, nutrition, and exercise. As this study assessed frailty only at baseline, we were unable to account for potential changes in frailty status over time, which may have influenced mortality outcomes during follow-up. Second, some analyses—particularly those related to cause-specific mortality—were based on a limited number of studies, warranting cautious interpretation. Further research is needed to clarify disease-specific mortality risks and disparities, and to inform more precise clinical management strategies. Third, we excluded grey literature, unpublished studies, and trial registry data, which may have introduced a risk of publication bias. This decision was made to ensure methodological rigor and data quality by focusing exclusively on peer-reviewed prospective cohort studies; however, it may have limited the comprehensiveness of the evidence base. Lastly, more than half of the included studies were rated as having high or very high risk of bias, primarily due to inadequate control for confounding variables and missing data. Future well-designed prospective cohort studies are needed to strengthen the evidence base and enhance the reliability of meta-analytic findings.

Conclusion

Frailty—regardless of domain or severity level—is a significant predictor of decreased survival and increased mortality among community-dwelling older adults. As a critical geriatric syndrome, it warrants close attention in clinical practice. Comprehensive assessment and stratified management of frailty should be emphasized to help healthcare providers identify and address frailty at all stages. Prioritizing resources for vulnerable groups should be considered by policymakers when allocating resources for caring for older adults. To prevent or mitigate adverse outcomes associated with frailty, future research should focus on developing personalized, long-term care strategies that can be integrated into daily routines—such as targeted nutritional interventions and safe, effective non-pharmacological approaches.

Abbreviations

HR

Hazard ratios

RR

Risk ratios

OR

Odds ratios

CHS

Cardiovascular health study

PRISMA

Preferred reporting items for systematic reviews and meta-analyses

WHO

World health organization

CI

Confidence intervals

ROBINS-E

Risk of bias in non-randomized studies of exposures

PI

Prediction interval

Authors’ contributions

LYL and RS conceived the research idea and contributed to the methods and search strategy. LYL and FX conducted the initial literature search; LYL, FX, and SJS contributed to reviewing search results and assessing the quality of included studies. LYL, FX, and RS conducted data extraction and completed data analysis with SHJ. The first draft of the manuscript was written by LYL and RS. It was critically revised by SHJ, CYM, and XLJ. All authors contributed to the interpretation of results and critiquing of the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by the National Research Foundation of Korea (NRF-2022R1A2C2011502). The funding bodies had no role in study design, data collection, and analysis, decision to publish, or preparation of the manuscript.

Data availability

Data is provided within the manuscript or supplementary information files.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.