Physical Frailty, Self-Rated Health, and All-Cause Mortality: Implications for Understanding Resilience in Aging

Abstract

Frailty and poor self-rated health (SRH) as potential indicators of reduced resilience are associated with all-cause mortality. However, it is lesser known whether these associations are independent of each other, physical disability or disease burden, and whether SRH remains predictive of mortality among frail older adults.

We leveraged the National Health and Aging Trends Study of U.S. Medicare beneficiaries, followed annually from 2011 to 2019. Baseline assessments included physical frailty and SRH (excellent/very good, good, or fair/poor). Cox models examined associations with all-cause mortality, adjusting for demographic characteristics and comorbidities. Sensitivity analyses excluded participants with probable or possible dementia at baseline.

Of the 7,425 participants at baseline, 14% were physically frail, and 25% reported fair/poor SRH. Although frailty was positively correlated with fair/poor SRH, 12% of frail participants reported excellent/very good health, and 6.6% of those with fair/poor SRH were non-frail. Over a median follow-up of 4.3 years, 29% died. Compared to non-frail participants, pre-frailty and frailty were associated with 1.4- and 2.0-fold increased hazard of mortality, respectively, after adjusting for SRH (p<0.001); reporting good and fair/poor SRH was associated with 29% and 59% higher hazard of mortality compared to excellent/very good SRH, respectively, after adjusting for frailty (p<0.001). Fair/poor SRH was associated with higher mortality among the frail. Sensitivity analyses yielded similar results.

Physical frailty and SRH, while related, capture distinct aspects of health. Each independently and jointly predicts mortality, with SRH retaining prognostic value in frail individuals. Incorporating both measures may improve risk stratification and guide tailored interventions.

Introduction

In the past half-century, numerous studies have investigated the relationship between self-rated health (SRH) and all-cause mortality. As per the Centers for Disease Control and Prevention, approximately 25% of adults aged 65 or above reported fair/poor health in 2019 [1]. Research has established that poor SRH is a reliable and valid predictor of all-cause mortality, often outperforming more complex clinical tools created for the same purpose [2–4].

Although SRH is a widely used, low-cost tool for evaluating morbidity and mortality risk, it is frequently oversimplified as a direct proxy for objective health status [5]. Unlike objective health measures, which typically include clinical diagnosis, biomarkers, or physical function tests [6], SRH captures a broader, subjective appraisal of overall health, including physical, mental, social, and functional dimensions [7]. While many studies have linked SRH to objective indicators of health, findings show that they are not always aligned [8–10]. For example, individuals with multiple chronic conditions may report good SRH, whereas others with few conditions report poor SRH [11–13]. These discrepancies suggest that SRH reflects more than disease burden, shaped by personal expectations, adaptation, psychological resilience, and sociocultural context [14]. Notably, most studies use comorbidity indices or disease counts as proxies for objective health, with less attention to other objective health constructs, such as physical frailty.

Physical frailty is a clinical syndrome characterized by aging-related physiological decline across multiple systems, increasing vulnerability to stressors, and heightened risk of adverse health outcomes in older adults [15]. In the United States, 4-16% of the community-dwelling adults aged 65 or older are frail, and 28-44% are pre-frail [16]. Like SRH, frailty strongly predicts all-cause mortality, with frail individuals facing up to a threefold increased risk of death [17]. A recent Chinese cohort study found that poor SRH and physical frailty independently predicted mortality risk [18]. However, less is known about how these two constructs interact, particularly whether SRH retains its predictive value among frail older adults, and whether discordant profiles, such as high SRH despite frailty, may signify underlying resilience, and thus confer lower mortality risk compared to low SRH without frailty.

To better understand the relationship between SRH and all-cause mortality in older adults, we analyzed data from 7,425 community-dwelling Medicare beneficiaries in the National Health and Aging Trends Study (NHATS) with measures of physical frailty and SRH. Our objectives were: (1) to assess the associations of frailty and SRH jointly with all-cause mortality, (2) to assess the association between SRH and mortality within the frail subset, and (3) to explore whether discrepancies between SRH and frailty status are associated with differential mortality risk. We hypothesize that SRH and frailty are independently associated with mortality, and that SRH remains a strong predictor of mortality even among frail older adults. Clarifying these relationships could support SRH as a simple and low-cost tool to identify at-risk frail older adults, potentially before the emergence of clinical sequelae.

Methods

Study Population

Of 7,600 community-dwelling older adults in NHATS at baseline (year 2011) with follow-up data (eFigure 1 in the Supplement), we excluded participants with missing information on physical frailty (n=169, 2.2%) and SRH (n=6, 0.1%), leaving 7,425 NHATS participants for analysis. Participants were assessed annually through 2019 or until death or drop-out to examine associations of baseline SRH and physical frailty with all-cause mortality.

Physical Frailty

Physical frailty was assessed at baseline using the Physical Frailty Phenotype (PFP), which comprises five binary criteria: exhaustion, low physical activity, weakness, slowness, and shrinking. These were operationalized as follows: self-reported low energy that limited activities (exhaustion); self-reported never walked or performed vigorous activities for exercise (low physical activity); measured average hand grip strength across two trials at or below the 20th percentile by sex and body mass index (BMI) (weakness); average walking speed across two trials at or below the 20^th percentile by sex and height (slowness), and BMI less than 18.5 kg/m² or unintentional weight loss of 10 or more pounds in the past 12 months (shrinking). Frailty status was then classified as frail (meeting ≥3 criteria), pre-frail (1-2 criteria), or non-frail (0 criteria) [19].

Self-rated health

In NHATS, participants were asked to rate their overall health using a five-point scale: excellent, very good, good, fair, or poor. We categorized SRH into three groups - excellent/very good health, good health, and fair/poor – to account for skewed response patterns and ensure sufficient sample sizes within each category for meaningful analysis.

All-cause mortality

All-cause mortality was assessed in every NHATS round from baseline in 2011 through 2019. Survival time was defined as the interval between the baseline interview date and the date of death; participants without a recorded death were censored at their last available interview date.

Covariates

Baseline covariates included age (in years: 65–69, 70–74, 75–79, 80 or above), gender, race (non-Hispanic White, non-Hispanic Black, Hispanic, and others), education (8^th grade or less, 9^th-12^th grade (no diploma), high school graduate or higher), number of comorbidities, mobility disability (fully able for all activities, modification in any activity, difficulty in any activity, assistance in any activity) [20], anxiety, depression, healthy relationships, community engagement, and smoking status. Smoking status was classified into current, past, and never smokers, based on self-reported history of regular smoking and current smoking status [21]. Comorbidities was determined based on participants’ self-reported physician diagnoses (1= “Yes”; 0= “No”) of myocardial infarction, any heart disease, high blood pressure, arthritis, osteoporosis, diabetes, lung disease, stroke, dementia, or cancer, and categorized as none, one, two, three, or four or more.

Depression was measured using the two-item Patient Health Questionnaire (PHQ-2), which asks how often participants experienced little interest or pleasure in activities and felt down, depressed, or hopeless [22]. Each item was scored from 1 (not at all) to 4 (nearly every day) and then converted from 0 to 3 scale. Scores for both items were summed, with totals ≥3 classified as depressed and <3 as non-depressed. Similarly, anxiety was measured using the Generalized Anxiety Disorder-2 questionnaire (GAD-2), which asks how often felt nervous, anxious, or on edge, and been unable to stop or control worrying. Scores for both items were summed, with totals ≥3 classified as anxiety and <3 as no anxiety. Healthy relationships were assessed using a self-reported question asking whether visits from friends or family occurred or were prevented because of health restrictions [23]. Response options ranged from 0 to 3: 0-no visit and without restriction, 1-no visit due to restriction, 2-visit despite restriction, and 3–visit and without restriction. Participants were dichotomized as having no visits versus any visits. Community engagement was defined using four items: attending religious services, participating in clubs/classes/activities, going out for enjoyment, and doing volunteer work. A summed score ranging from 0 to 8 was created, with 0 indicating the lowest engagement and 8 indicating the highest.

Missing data for education (n=24), number of comorbidities (n=72), anxiety (n=34), depression (n=51), healthy relationships (n=6), community engagement (n=17), and smoking status (n=6) were addressed using multiple imputation, which imputes covariates using models that are compatible with the specified substantive model, i.e., Cox model [24]. Imputation was performed using the SMCFCS command in STATA [25].

Statistical Analysis

Baseline characteristics were summarized using frequencies and survey-weighted percentages for categorical variables to generate nationally representative estimates of Medicare beneficiaries. Group comparisons used χ² tests, and polychoric correlation was used to assess the correlation between physical frailty and SRH. Kaplan-Meier curves were constructed to compare crude all-cause mortality by SRH and physical frailty levels. Survey-weighted Cox proportional hazards regression models were performed to determine the independent and joint association of mortality with physical frailty and SRH after covariate adjustment, with Wald’s test for interaction. A separate Cox model was performed to assess the SRH-mortality association within the frail subgroup. Two multivariable models were estimated. Model 1 adjusted for all covariates except psychosocial variables, while Model 2 additionally included psychosocial factors including anxiety, depression, healthy relationships, and community engagement.

To examine discordant patterns, we defined a “resilient” group as frail individuals with excellent or very good SRH, and a “non-resilient” group as non-frail individuals with fair or poor SRH. Propensity score matching (1:1, caliper=0.2) using all baseline covariates was performed to achieve covariate balance between these groups, which reduced the sample size from 379 to 156. Further, Cox models were used to examine the relationship between resilience and all-cause mortality.

For sensitivity analyses, we repeated all primary analyses after excluding individuals with probable dementia, as well as those with either probable or possible dementia at baseline [26] to ensure the robustness of the results.

All analyses were performed using STATA 19 (Standard Edition) and SAS 9.4.

Results

Baseline Characteristics

The study sample included a total of 7,425 community-dwelling NHATS participants at baseline in 2011. Among them, 53% were aged 65 to 74 years, 56% were females, 82% were non-Hispanic White, and 78% had at least a high school education (Table 1). A high prevalence of comorbidities was observed at baseline, with 70% reporting two or more chronic conditions. About 13% of participants reported anxiety and 15% reported depression. From a social connection perspective, 87% reported receiving visits from family or friends outside the household, and the mean community engagement score of 4 indicated that participants were neither isolated nor highly engaged with the community. Additionally, 13% reported a disability requiring assistance with daily activities, and 9% were current smokers. Over the nine-year follow-up period (2011-2019), 2,164 (29.1%) died, with an incidence rate of 6.2 per 100 person-years. At baseline, about 10% and 11% of individuals had probable and possible dementia, respectively.

Table1.

Baseline characteristics of study population (N=7,425)

Predictors	Overall (N=7,425)
	N (weighted %)
Age
65-69 years	1,386 (28.09)
70-74 years	1,551 (25.03)
75-79 years	1,478 (19.08)
80 or above	3,010 (27.80)
Gender
Male	3,113 (43.57)
Female	4,312 (56.43)
Race
Non-Hispanic White	5,124 (81.56)
Non-Hispanic Black	1,634 (8.18)
Hispanic	440 (6.69)
Others	227 (3.57)
Education
8th grade or less	949 (10.08)
9-12th grade (no diploma)	1,039 (11.33)
High school graduate or higher	5,413 (78.31)
Missing	24 (0.28)
Number of comorbidities
No Comorbidities	575 (9.03)
One Comorbidity	1,299 (19.67)
Two Comorbidities	1,882 (25.52)
Three Comorbidities	1,717 (21.88)
Four or more Comorbidities	1,880 (22.94)
Missing	72 (0.96)
Mobility Disability
Fully able for all activities	4,456 (67.14)
Modification in any activity	1,274 (14.89)
Difficulty in any activity	427 (4.85)
Assistance in any activity	1,268 (13.12)
Smoking Status
Current Smoker	582 (8.58)
Past Smoker	3,177 (44.14)
Never Smoker	3,660 (47.16)
Missing	6 (0.12)
Dementia
Probable Dementia	979 (9.65)
Possible Dementia	954 (10.61)
No Dementia	5,492 (79.74)
Anxiety
Yes	996 (12.47)
No	6,395 (87.09)
Missing	34 (0.44)
Depression
Yes	1,181 (14.45)
No	6,193 (84.92)
Missing	51 (0.63)
Healthy relationships
Yes	6,331 (87.39)
No	1,088 (12.55)
Missing	6 (0.06)
Community engagement, weighted mean (95% CI)	4.23 (4.15-4.31)
Missing	17 (0.16)

Frailty Status and All-Cause Mortality

Out of 7,425 individuals, 40% were non-frail, 46% were pre-frail, and 14% were frail at baseline (eTable 1 in the Supplement). A comparison of baseline characteristics across these groups (eTable 2 in the Supplement) showed that physically frail participants tended to be older, with 48% aged 80 years or older, 51% had four or more comorbidities, and 51% reported disability requiring assistance with all activities, compared to 31%, 25%, and 12% among pre-frail participants and 17%,11%, and 1% among non-frail participants, respectively.

Kaplan-Meier curves showed a markedly faster decline in survival probability among frail participants compared to pre-frail and non-frail participants (Figure 1A). After adjusting for baseline covariates and SRH, physical frailty was associated with a twofold higher hazard of all-cause mortality (adjusted hazard ratio [aHR]= 2.10, 95% confidence interval [CI]: 1.78-2.48; aHR=2.01, 95% CI: 1.70-2.37 after further adjusting for psychosocial variables), compared to the non-frail group (Table 2).

Figure 1.

Kaplan-Meier survival curve of all-cause mortality by Frailty status (A), by self-rated health status (B)

Table 2.

Adjusted Hazard Ratios (aHR) with 95% Confidence interval (CI) of all-cause mortality (N=7,425)

Predictors	Unadjusted for Psychosocial measures		Adjusted for psychosocial measures
	aHR (95% CI)	p-value	aHR (95% CI)	p-value
Frailty
Non-frail	Ref		Ref
Pre-frail	1.46 (1.26-1.70)	<0.001	1.42 (1.22-1.65)	<0.001
Frail	2.10 (1.78-2.48)	<0.001	2.01 (1.70-2.37)	<0.001
Self-rated Health
Excellent/very good	Ref		Ref
Good	1.31 (1.11-1.53)	0.001	1.29 (1.10-1.52)	0.002
Fair/Poor	1.65 (1.40-1.94)	<0.001	1.59 (1.34-1.89)	<0.001
Age
65-69 years	Ref		Ref
70-74 years	1.48 (1.15-1.91)	0.003	1.47 (1.15-1.89)	0.003
75-79 years	2.39 (1.93-2.96)	<0.001	2.39 (1.93-2.96)	<0.001
80 or above	5.24 (4.30-6.39)	<0.001	5.19 (4.26-6.32)	<0.001
Gender
Male	Ref		Ref
Female	0.69 (0.61-0.77)	<0.001	0.71 (0.63-0.79)	<0.001
Race
Non- Hispanic White	Ref		Ref
Non- Hispanic Black	0.82 (0.71-0.94)	0.007	0.81 (0.70-0.94)	0.005
Hispanic	0.61 (0.50-0.75)	<0.001	0.58 (0.48-0.71)	<0.001
Others	0.74 (0.50-1.10)	0.136	0.70 (0.48-1.04)	0.078
Education
8th grade or less	Ref		Ref
9-12th grade (no diploma)	1.17 (0.93-1.47)	0.171	1.17 (0.94-1.46)	0.151
High school graduate or higher	1.08 (0.90-1.30)	0.415	1.13 (0.94-1.37)	0.198
Number of Comorbidities	1.12 (1.06-1.18)	<0.001	1.12 (1.06-1.19)	<0.001
Mobility Disability
Fully able for all activities	Ref		Ref
Modification in any activity	1.46 (1.22-1.75)	<0.001	1.43 (1.20-1.71)	<0.001
Difficulty in any activity	1.50 (1.21-1.85)	<0.001	1.43 (1.15-1.78)	0.002
Assistance in any activity	2.25 (1.86-2.72)	<0.001	2.19 (1.80-2.66)	<0.001
Smoking Status
Never Smoker	Ref		Ref
Past Smoker	1.18 (1.05-1.34)	0.009	1.17 (1.04-1.33)	0.013
Current Smoker	1.67 (1.33-2.08)	<0.001	1.56 (1.25-1.94)	<0.001
Anxiety
No			Ref
Yes			0.96 (0.80-1.14)	0.629
Depression
No			Ref
Yes			1.09 (0.90-1.32)	0.382
Healthy relationships
No			Ref
Yes			0.89 (0.78-1.01)	0.080
Community engagement			0.95 (0.93-0.98)	<0.001

Self-rated health and All-cause Mortality

Of 7,425 participants, 44% reported excellent/very good health, 31% reported good health, and 25% reported fair/poor health (eTable 2 in the Supplement). Over nine years, a dose-response relationship was observed between worse SRH and higher all-cause mortality (Figure 1B). Adjusted hazard ratios were 1.65 (95% CI: 1.40-1.94) for fair/poor SRH and 1.31 (95% CI: 1.12-1.52) for good health, compared with excellent/very good health. After additional adjustment for psychosocial measures, the corresponding hazard ratios were 1.59 (95% CI: 1.34-1.89) and 1.29 (95% CI: 1.10-1.52), respectively (Table 2).

Self-rated health and all-cause mortality within the frail subset

SRH remained an independent predictor of all-cause mortality even among the frail subset (n=1,306), with fair/poor health having a higher hazard (aHR=1.36, 95% CI: 1.03-1.81; Table 3) compared to excellent/very good health without adjusting for psychosocial measures. After adjusting for psychosocial measures, the association was attenuated and showed borderline statistical significance (aHR=1.30, 95% CI: 0.99-1.70; p=0.06). Compared to the full-sample model in Table 2, this model showed “good” SRH, current smoking status, limited disability, Black race, and age group “70-74” were no longer significantly associated with all-cause mortality. Additionally, hazard ratios for other covariates were attenuated (Table 3).

Table 3.

Adjusted hazard ratio (aHR) with 95% Confidence interval (CI) of all-cause mortality among the frail participants (n=1,306)

Predictors	Unadjusted for psychosocial measures		Adjusted for psychosocial measures
	aHR (95% CI)	p-value	aHR (95% CI)	p-value
Self-rated Health
Excellent/very good	Ref		Ref
Good	0.99 (0.71-1.38)	0.937	0.96 (0.70-1.33)	0.820
Fair/Poor	1.36 (1.03-1.81)	0.033	1.30 (0.99-1.70)	0.060
Age
65-69 years	Ref		Ref
70-74 years	1.18 (0.73-1.89)	0.489	1.09 (0.66-1.79)	0.742
75-79 years	2.09 (1.37-3.20)	0.001	2.04 (1.29-3.20)	0.003
80 or above	3.98 (2.65-5.97)	<0.001	3.74 (2.41-5.82)	<0.001
Gender
Male	Ref		Ref
Female	0.71 (0.61-0.82)	<0.001	0.74 (0.63-0.86)	<0.001
Race
Non- Hispanic White	Ref		Ref
Non- Hispanic Black	0.79 (0.61-1.02)	0.067	0.77 (0.60-1.00)	0.051
Hispanic	0.49 (0.34-0.71)	<0.001	0.46 (0.32-0.67)	<0.001
Others	0.51 (0.28-0.94)	0.031	0.54 (0.30-0.95)	0.034
Education
8th grade or less	Ref		Ref
9-12th grade (no diploma)	1.14 (0.81-1.59)	0.456	1.15 (0.82-1.60)	0.415
High school graduate or higher	1.21 (0.94-1.56)	0.138	1.30 (1.00-1.68)	0.048
Number of Comorbidities	1.14 (1.02-1.28)	0.027	1.15 (1.03-1.29)	0.018
Mobility Disability
Fully able for all activities	Ref		Ref
Modification in any activity	1.04 (0.75-1.44)	0.819	1.01 (0.74-1.37)	0.939
Difficulty in any activity	1.32 (0.93-1.86)	0.115	1.25 (0.89-1.76)	0.201
Assistance in any activity	1.99 (1.45-2.73)	<0.001	1.95 (1.41-2.69)	<0.001
Smoking Status
Never Smoker	Ref		Ref
Past Smoker	1.10 (0.88-1.37)	0.390	1.09 (0.87-1.36)	0.457
Current Smoker	0.94 (0.69-1.29)	0.704	0.88 (0.64-1.21)	0.420
Anxiety
No			Ref
Yes			0.86 (0.70-1.05)	0.128
Depression
No			Ref
Yes			1.20 (0.95-1.53)	0.129
Healthy relationships
No			Ref
Yes			0.93 (0.75-1.14)	0.474
Community engagement			0.92 (0.88-0.96)	0.001

Frailty and SRH jointly with all-cause mortality

Among the 7,425 participants, 66% of non-frail participants rated their health as excellent/very good (eFigure 2A in the Supplement). In contrast, 64% of frail participants reported fair/poor health at baseline. Conversely, 59% of those with excellent/very good SRH were non-frail, and 36% of those with fair/poor health were frail (eFigure 2B in the Supplement). Notably, discordant subgroups comprised 4.3% of the total sample: 1.7% reported excellent/very good health despite being frail, and 2.6 % reported fair/poor health despite being non-frail.

The Kaplan–Meier curves in eFigure 3 suggest greater separation in mortality risk by frailty status than by SRH. Within each frailty stratum, SRH demonstrates a graded association with mortality, particularly among non-frail and pre-frail participants. Among frail individuals, the distinction in mortality risk appears greatest between fair/poor SRH and higher SRH categories. Figure 2 shows results from the Cox model assessing the joint association between frailty and SRH with all-cause mortality. Frail participants with fair/poor SRH showed a significantly higher hazard of all-cause mortality compared with non-frail participants with excellent/very good SRH. This association remained strong after multivariable adjustment, both without (aHR=4.14, 95% CI: 3.27-5.26) and with additional adjustment for psychosocial measures (aHR=3.81, 95% CI: 2.97-4.89). In a fully adjusted model, a closer examination of the interaction between SRH and frailty in relation to all-cause mortality revealed a nuanced pattern. Specifically, the mortality risk associated with poor SRH compared to excellent/very good health was strongest among non-frail individuals (aHR=2.63, 95% CI: 2.01-3.45) and weakest among frail individuals (aHR=1.23, 95% CI: 0.83-1.82). Conversely, the mortality risk associated with frailty compared to non-frailty was most pronounced among participants with excellent/very good SRH (aHR=3.09, 95% CI: 2.29-4.17) and attenuated among those reporting fair/poor SRH (aHR=1.45, 95% CI: 1.00-2.09).

Figure 2.

Adjusted hazard ratios (aHR) with 95% CI of all-cause mortality by combined profiles of self-rated health and frailty status.

*LCL- Lower Confidence interval; UCL- Upper Confidence interval

Discordant groups and all-cause mortality hazard

After comparing the baseline characteristics between the two groups, the “resilient group” (n=162) was older, more likely to be female and white, had higher educational attainment, had a higher burden of disability, and was highly anxious and depressed compared to the “non-resilient group” (n=217) (eTable 3 in the Supplement). Across all racial groups, the proportion of the resilient group was significantly lower with Non-Hispanic Whites (45%), followed by Non-Hispanic Blacks (27%), Others (27%), and Hispanics (19%) compared to the non-resilience group. Similarly, a resilient group was 19-30% among all age groups, except among those 80 or above (64%). Following matching on baseline covariates (eTable 4 and eFigure 4 in the Supplement), individuals classified as “resilient” had a two-fold increased hazard of all-cause mortality compared with the “non-resilient” group (aHR=2.04, 95% CI: 1.23-3.38).

Sensitivity analysis

In sensitivity analyses excluding individuals with probable dementia (n=6,466) and those with either probable or possible dementia (n=5,492), results were consistent with the main findings (Table 2). Frailty remained associated with an approximately twofold higher hazard of all-cause mortality. The association between poor SRH and all-cause mortality was stronger in both restricted samples (aHR=1.85, 95% CI: 1.52-2.25 and aHR=1.94, 95% CI: 1.58-2.40, respectively), with additional adjustment for psychosocial variables having minimal impact (aHR=1.81, 95% CI: 1.48-2.22 and aHR=1.91, 95% CI: 1.54-2.37; eTable 5).

Among the frail, excluding those with probable dementia reduced the sample from 1,306 to 862, and excluding those with either probable or possible dementia further reduced the sample to 640. In these restricted samples, poor SRH was associated with a higher, but marginally significant hazard of mortality compared with excellent/very good SRH, with effect estimates slightly larger than in the main analyses (aHR=1.42, 95% CI: 1.00-2.02, and aHR=1.43, 95% CI: 0.94-2.16, respectively; eTable 6). Additional adjustment for psychosocial variables only slightly attenuated these associations (aHR=1.37, 95% CI: 0.96-1.95 after excluding probably dementia and aHR=1.37, 95% CI: 0.90-2.09 after excluding both probable and possible dementia).

For the discordant groups, the matched sample decreased from 156 to 128 and 106 in the two restricted samples, respectively. In Cox model, the resilient group showed higher hazards of mortality than the non-resilient group (aHR=1.52, 95% CI: 0.88-2.61 after excluding probably dementia and aHR=1.34, 95% CI: 0.71-2.53 after excluding both probable and possible dementia), though these associations were not statistically significant. Overall, sensitivity analyses did not change the direction or interpretation of the findings.

Discussion

Physical frailty and SRH were independently and jointly associated with nine-year all-cause mortality in a nationally representative sample of community-dwelling Medicare beneficiaries in the United States. When mutually adjusted, physical frailty doubled the hazard of mortality compared with non-frail status, and fair/poor SRH increased the hazard by 59% compared with excellent/very good health. Frail participants with fair/poor SRH showed a fourfold higher mortality hazard than non-frail participants with excellent/very good health. Within the frail subgroup, fair/poor SRH remained significant, increasing mortality hazard by 30%.

SRH and physical frailty are distinct yet related constructs, each contributing unique insights into an individual’s overall health. SRH is a subjective and multidimensional measure that reflects an individual’s perception of physical, mental, and social health, whereas physical frailty encompasses both subjective experiences (e.g., low energy) and objective indicators (e.g., weakness) within the physical health domain [18, 27, 28]. To preserve conceptual separation between perceived health and objective vulnerability, we operationalized frailty using Fried’s frailty phenotype, which is hypothesized as a distinct clinical entity, rather than a cumulative deficit index (e.g., Rockwood’s Frailty Index; 29) that partially embeds disease-related and psychosocial components overlapping with SRH. Consistent with prior studies, both poor SRH and frailty were independently associated with increased mortality [18,28]. Notably, poor SRH remained predictive of mortality risk even among physically frail older adults, extending prior work by directly evaluating the prognostic significance of SRH within this population. This finding underscores its value as simple and low-cost prognostic tool in this highly vulnerable subgroup.

A longitudinal study among community-dwelling older adults in China found that SRH and physical frailty were independently associated with all-cause mortality, but no significant interaction was observed [18]. In contrast, our study revealed a similar joint association with a statistically significant interaction between SRH and physical frailty. However, the pattern did not support a positive synergistic effect – namely, that the mortality hazard associated with frailty would be amplified among individuals with poor SRH. Instead, the combined effect appeared weaker than expected, with the influence of each factor reduced when both were present. One possible explanation for this unexpected interaction is a ceiling effect. Among individuals already at heightened risk of mortality due to either poor SRH or physical frailty, having both may not further proportionally increase their hazards. Nevertheless, these findings underscore the importance of considering both SRH and physical frailty as complementary indicators of overall health in older adults.

Physical frailty and SRH are not always concordant, as found in previous studies [30–33], and we hypothesized that this discordance may provide insight into resilience in aging. Prior work, including Wu et al. [18], has articulated the hypothesis that frail individuals reporting excellent SRH may represent a resilient subgroup despite marked physical vulnerability. We share this conceptual framework and view our work as building directly on this hypothesis. Resilience is widely conceptualized as a multidimensional construct shaped by one’s capacity to recover from adversity, with higher resilience linked to better coping capacity, quality of life, and lower mortality [5, 34–36], whereas negative self-perceptions are associated with lower self-efficacy and worse health outcomes [37]. Rather than redefining resilience or modeling it as a separate construct, our study operationalized one observable manifestation of resilience: the ability to maintain a positive self-perception of health despite measurable physical vulnerability. Within this framework, discordance between SRH and frailty serves as a pragmatic, empirically observable proxy for resilience expressed through incongruence between subjective and objective health measures in population-based data. Individuals who perceive their health as excellent or very good despite being frail may possess adaptive resources that buffer the impact of physical deficits, whereas non-frail individuals reporting poor SRH may reflect vulnerability despite preserved physical function. We recognize that such discordance may also reflect misperception or incomplete awareness of health status; nevertheless, it represents a meaningful pattern through which adaptive or maladaptive responses to vulnerability can be examined empirically rather than inferred conceptually. By explicitly examining mortality risk across discordant SRH–frailty groups, our study provides a preliminary empirical test of whether favorable SRH among frail individuals is protective when contrasted with other discordant patterns. Unexpectedly, we found higher mortality risk among frail individuals with favorable SRH compared with non-frail individuals with poor SRH. This finding does not contradict the resilience hypothesis per se, but rather highlights the complexity of interpreting SRH–frailty discordance. One possible explanation is delayed care-seeking or under-recognition of health limitations among those with optimistic self-appraisal, potentially postponing early disease prevention or detection. On the other hand, individuals with poor SRH despite preserved physical function may seek healthcare earlier due to their perceived poor health, thereby facilitating earlier identification and management of health risks not captured by frailty measures. Unmeasured social, psychosocial, or cultural factors may also contribute to these findings. Together, these findings highlight the need to attend to discordant groups, as reliance on physical frailty or SRH in isolation may conceal clinically meaningful risks and adaptive processes relevant to aging and resilience.

Our study has several limitations. First, physical frailty and SRH were measured only at baseline. Because both measures can change over time, relying on baseline values may not fully capture their dynamic nature. Nevertheless, baseline assessments remain meaningful from a predictive standpoint, as they capture initial health status and are widely used in epidemiological studies to estimate future risk. Second, the small sample size of the discordant group might have limited statistical power and generalizability. Third, SRH was assessed using a single-item question, which may have introduced reference bias, as older adults from different racial/ethnic backgrounds, genders, or sociocultural contexts may differ in how they interpret and report health [38,39]. Such measurement variability can complicate comparisons across the groups. Fourth, loss to follow-up may have introduced selection bias, as those who dropped out were likely at higher mortality risk, leading to underestimation. Despite these limitations, our study used a validated measure of frailty in a nationally representative sample. The use of survey weights enhanced generalizability of our findings to all community-dwelling older individuals in the United States. Further, the extended nine-year follow-up allowed for robust evaluation of long-term mortality outcomes associated with baseline frailty and SRH.

In conclusion, our study revealed the complex interplay between SRH and physical frailty in predicting all-cause mortality. These findings highlight the importance of incorporating both subjective and objective health measures into risk assessment, health education, prevention, and intervention efforts. While our study did not examine socio-environmental factors, healthcare literacy, or access to care, which may be particularly relevant for understanding discordant SRH-frailty profiles, future research should explore these dimensions. We also recommend future studies include open-ended questions to capture individuals’ perceptions and reasoning behind their SRH ratings, which may provide deeper insight into the mechanisms linking SRH and physical frailty to mortality risk.