Quantifying and Classifying Physical Resilience Among Older Adults: The Health, Aging, and Body Composition Study

Abstract

Background

The concept of resilience has gained increasing attention in aging research; however, current literature lacks consensus on how to measure resilience. We constructed a novel resilience measure based on the degree of mismatch between persons’ frailty level and disease burden and examined its predictive validity. We also sought to explore the physiological correlates of resilience.

Methods

Participants were 2,457 older adults from the Health, Aging, and Body Composition Study. We constructed the resilience measure as the residual taken from the linear model regressing frailty on age, sex, race/ethnicity, 14 diseases, self-reported health, and number of medications. Participants were classified into three groups—adapters, expected agers, and premature frailers—based on residuals (less than, within, or above one standard deviation of the mean). Validation outcomes included years of able life (YAL), years of healthy life (YHL), years of healthy and able life (YHAL), disability, hospitalization, and survival.

Results

The average YHAL was 5.1, 7.7, and 9.1 years among premature frailers, expected agers, and adapters, respectively. Compared with premature frailers and expected agers, adapters had significantly lower rates of disability, hospitalization, and mortality and higher proportion surviving to 90 years. The likelihood of surviving to 90 years was 20.4%, 30.6%, and 39.7% among premature frailers, expected agers, and adapters.

Conclusions

We developed and validated a novel approach for quantifying and classifying physical resilience in a cohort of well-functioning white and black older adults. Persons with high physical resilience level had longer healthy life span and lower rates of adverse outcomes.

Frailty—a clinical state of physiological vulnerability to stressors—results from aging-related decline in reserve and function across multiple organ systems (1,2). For older adults, their frailty level matches their extent of disease; these individuals may be termed “expected agers.” In contrast, some individuals with marked organ system impairments do not exhibit corresponding level of frailty; these individuals may be termed “adapters.” In addition, individuals whose frailty level is higher than expected from their biological features and disease burden may be termed “premature frailers.” We theorized that adapters have the intrinsic ability to adapt to and mitigate the consequences of cumulative, intrinsic damage in organ systems. In this sense, adapters may serve as a candidate aging group for capturing physical resilience, defined as “the ability to recover or optimize function in the face of age-related losses or diseases” and considered a characteristic at the whole-person level (3).

Using data from the Cardiovascular Health Study (CHS), Sanders and colleagues (4) showed that adapters had longer years of healthy life (YHL) without disability and lower mortality than expected agers and premature frailers. Notably, this approach could potentially quantify an older adult’s physical resilience before encountering a stressor (eg, hip fracture). However, whether this novel approach is useful for capturing resilience in other cohorts is not known.

The purpose of the present study was twofold. First, we developed a resilience measure based on three groups—adapters, expected agers, and premature frailers—within the Health ABC cohort and validated its predictive validity by examining whether adapters have longer survival and YHL and lower incidence of disability, hospitalization, and mortality than expected agers and premature frailers. Second, we sought to explore the physiological correlates of this novel resilience measure. Having a better understanding of why some older adults can maintain relatively high function in the presence of organ impairments, and others do not, can help identify risk and protective factors and design targeted interventions to promote resilience.

Methods

Data and Study Population

The Health ABC Study is a longitudinal cohort designed to examine age-related changes in health and body composition and functional limitations in initially well-functioning older adults. Between March 1997 and July 1998, 3,075 Black and White individuals aged 70–79 years were recruited from a list of Medicare beneficiaries provided by the Health Care Financing Administration at two study sites across the United States, Pittsburgh, Pennsylvania, and Memphis, Tennessee. Eligibility criteria were (i) free of life-threatening illness, (ii) self-reported ability to walk a quarter of a mile, to climb 10 steps without resting, and to perform basic activities of daily living without assistance, and (iii) no intention to move out of the current geographic area for at least 3 years. These inclusion criteria resulted in a study population that was healthier than an age-matched general population at enrollment. Details about the Health ABC study design have been described elsewhere (5). All participants provided written informed consent. The study protocol was approved by the institutional review boards of the two clinical sites (University of Pittsburgh and University of Tennessee) and the Data Coordinating Center at the University of California, San Francisco.

Analytic Sample

We used data from the second annual clinic visit, when direct calculation of weight loss—one component for measuring physical frailty—between two consecutive visits was possible. Of the 3,075 participants enrolled at baseline, 2,457 (79.9%) with complete data to construct the resilience variable were included in the present analysis.

Frailty

The level of frailty was measured by the Scale of Aging Vigor Epidemiology (SAVE) (6), a 10-point frailty scale that was developed to remove the ceiling effect of the original Fried’s physical frailty phenotype and to achieve better differentiation of frailty. Tertiles were considered for each of the five components: walk time, grip strength, exhaustion, physical activity, and weight change (Table 1). The best tertile received a score of 0, the middle tertile received a score of 1, and the worst tertile received a score of 2. The total score was the sum of the five components, ranging from 0 (least frail) to 10 (most frail). The predictive validity of the SAVE has been previously demonstrated in multiple cohorts (6,7).

Table 1.

Cut-Points of Frailty Components for the Scale of Aging Vigor Epidemiology)

	Best Tertile (score: 0)	Middle Tertile (score: 1)	Worst Tertile (score: 2)
Walk time (seconds to walk 20 m)
Mena	≤16	>16 to ≤18	>18
Women with height <160 cm	≤17	>17 to ≤20	>20
Women with height ≥160 cm	≤17	>17 to ≤19	>19
Grip strength (kg)
Men with BMI <24 kg/m2	>38	>32 to ≤38	≤32
Men with BMI ≥24 and <26 kg/m2	>41	>35 to ≤41	≤35
Men with BMI ≥26 and <29 kg/m2	>42	>35 to ≤42	≤35
Men with BMI ≥29 kg/m2	>42	>35 to ≤42	≤35
Women with BMI <23 kg/m2	>24	>20 to ≤24	≤20
Women with BMI ≥23 and <27 kg/m2	>24	>21 to ≤24	≤21
Women with BMI ≥27 and <31 kg/m2	>26	>20 to ≤26	≤20
Women with BMI ≥31 kg/m2	>26	>21 to ≤26	≤21
Usual energy levelb	8–10	6–7	0–5
Physical activityc (kcal/kg/wk)
Men	≥43	>11 to <43	≤11
Women	≥31	>7 to <31	≤7
Weight change in the past year (pounds)
Men	>1.5	> −3 to ≤1.5	≤ −3
Women	>1.5	> −2 to ≤1.5	≤ −2

Note: BMI = Body mass index.

^aTertiles for walk time did not differ for men based on average height.

^bUsual energy level was measured by the question, “Do you feel full of energy?” in the Geriatric Depression Scale.

^cDid not collect data for kcal/kg/wk of exercise/recreation, therefore total physical activity is quantified based on the kcal/kg/wk doing major chores, walking and climbing stairs, working, volunteering and caregiving.

Clinical Disease

We used algorithms based on self-reported physician diagnoses, recorded medication use, and laboratory test to define the presence of cancer (excluding nonmelanoma skin cancer), coronary heart disease (angina, myocardial infarction, bypass surgery, and angioplasty), heart failure, hypertension, cerebrovascular disease (stroke, transient ischemic attack, and carotid endarterectomy), diabetes mellitus, osteoporosis, osteoarthritis, kidney disease, lung disease (chronic bronchitis, chronic obstructive pulmonary disease, and emphysema), and Parkinson’s disease. Depression was defined as a score greater than 16 on a 20-item Center for Epidemiology Study-Depression scale (8).

Resilience

Resilience was measured based on a residual approach that was previously developed in the CHS cohort (4). We used linear regression to regress the level of frailty (score: 0–10) on each clinical disease, disease burden (indicated by self-rated health and number of medications), age, age², age³, race/ethnicity (Black and White), and sex. We included the polynomial forms of age to allow for nonlinear relationship between age and frailty. Inclusion of the quadratic and cubic terms of age also allows for better comparison of our findings to previous studies. Subsequently, residuals from the regression were used to define three aging groups based on values less than, within, or above 1 SD (1.89) of the mean residual value. Participants whose observed frailty scores were at least 1.89 points higher than their regression-estimated scores were considered premature frailers, those whose observed frailty scores were at least 1.89 points lower than their regression-estimated scores were considered adapters, and those whose frailty scores were within 1 SD of their regression-estimated scores were considered expected agers.

Alternatively, we defined three resilience groups using residuals imported from the regression model developed in the CHS cohort (Figure 1). The SD for classifying these groups was 1.85 in the original model. Participants whose observed frailty scores were more than 1.85 points higher than, within 1.85 points, and more than 1.85 lower than their regression-estimated scores were defined as “premature frailers,” “expected agers,” and “adapters,” respectively.

Figure 1.

Histogram of residuals by resilience group.

Outcomes

Outcomes include all-cause mortality, survival to 90 years, disability, hospitalization, years of able life (YAL), YHL, and years of healthy and able life (YHAL). All outcomes were assessed starting after the second clinic visit.

Mortality and survival to 90 years

Participants alive at the time of analysis were censored at the date of last contact or by the end of the follow-up period (April 30, 2010 for the Memphis site and May 30, 2010 for the Pittsburgh site), whichever came first. Deaths were ascertained by review of local obituaries, by reports to the clinical centers by family members, or by means of the semiannual telephone contacts. For the analysis for survival to 90 years, only participants who died or were at least 90 years of age by the end of the study were included (N = 1,907).

Disability

Disability was defined as self-reported difficulty in getting in and out of bed or chairs, bathing or showering, or dressing. Time to incident disability was defined as the time from the date of second annual clinic visit to the date of visit when disability was first reported. Persons who were disabled at baseline (second clinic visit) were excluded from the analysis for incident disability.

Hospitalization

Participants were asked to report any hospitalizations (an illness episode resulting in overnight admission to an acute care hospital) every 6 months through telephone contacts (9). Hospitalization data were adjudicated through April 30, 2010 for the Memphis site and May 30, 2010 for the Pittsburgh site. Time to incident hospitalization was defined as the time from the date of the second clinical visit to the date of first hospital admission.

YAL, YHL, and YHAL

YAL was calculated as the number of years the participant did not have disability. YHL was calculated as the number of years the participant reported good or better health on a scale of excellent, very good, good, fair, or poor. YHAL was calculated as the number of years the participant was in good health and had no disability.

Physiological Measures

Fasting blood samples were collected at the second annual clinic visit using standardized protocols and quality assurance (9). Physiological measures included serum glucose (mg/dL), total cholesterol (mg/dL), high-density lipoprotein (HDL) cholesterol (mg/dL), low-density lipoprotein (LDL) cholesterol (mg/dL), and triglycerides (mg/dL); C-reactive protein (CRP; μg/mL), interleukin-6 (IL-6; pg/mL), and IL-18 (pg/mL); cystatin C (mg/L); blood pressure (BP; mmHg) and ankle-arm index (AAI).

Covariates

Sociodemographics included study site (Pittsburgh and Memphis), age (years), race (Black and White), and education (less than high school, high school or equivalent, and more than high school). Behavioral characteristics included smoking status (current, former, and never), and body mass index (BMI) defined as body weight (kilograms) divided by height (meters) squared. Self-rated health was classified as excellent, very good, good, fair, or poor. Participants were asked to bring all prescription and over the counter medications used in the previous 2 weeks.

Statistical Analyses

We first calculated the Pearson’s correlation between residuals derived from regression model we developed using data from the Health ABC and those derived from applying the regression coefficients imported from the model developed in the CHS. Then, we examined the agreement between two models with regard to the categorical definition of resilience (premature frailer, expected ager, and adapter) using the percentage of absolute agreement and Cohen’s kappa statistic.

We conducted several sensitivity analyses. First, we dropped the age squared and age cubic terms from the model to evaluate the robustness of resilience classification to specification of age. Second, we fit two models separately for men and women to assess the sensitivity to sex-specific specification.

We compared characteristics between three groups—premature frailer, expected ager, and adapter—using a Pearson chi-square test for categorical variables and analysis of variance F test or nonparametric equivalent (eg, Kruskal–Wallis test) for continuous variables. We described each outcome (YAL, YHL, YHAL, disability, hospitalization, mortality, and survival to 90 years) using mean and SD, count and percentage, or rates per 1000 person-years by each resilience group. We conducted a series of linear regressions to compare YAL, YHL, and YHAL between premature frailers, expected agers, and adapters. We used Cox proportional hazard models to compare the hazard of incident disability, incident hospitalization, and all-cause mortality, respectively, between the three groups. Survival to 90 years (yes/no) was modeled using logistic regression to estimate odds ratio. Study site (Pittsburgh and Memphis), years of education, and smoking status (current, former, and never) were included in all multivariable adjusted models. To evaluate the value of resilience beyond frailty, we examined the association between resilience and survival to 90 years after adjustment of frailty and other covariates (study site, education, and smoking).

To evaluate the relationship of physiological measures and resilience, continuous physiological measures were classified into categories based on quartiles of the sample distribution or clinically meaningful cut-points and Pearson chi-square tests were used. Because resilience was defined as a three-level categorical outcome, we used the multinomial regression to simultaneously examine the physiological factors that were associated with resilience at a significance level of p < .10 in the univariate analysis. Relative risks were estimated for adapters and premature frailers with respect to expected agers (reference group). We included study site, years of education, and smoking status in all multivariable adjusted models.

All tests were two-sided with a significance level of p < .05. All analyses were conducted using Stata 15.0 (Stata Corp, College Station, TX).

Results

Sample Characteristics

The average age of the cohort was 75.1 years (SD = 2.9); 49.7% were women and 39.0% were Black (Table 2). In the final model, age, sex, race, and clinical diseases explained 17.0% of the variability in frailty. Of the 2,457 study participants, 1,645 (67.0%) were expected agers, 415 (16.9%) were premature frailers, and 397 (16.2%) were adapters based on the regression model developed in the Health ABC. When defining resilience groups using residuals derived from the model developed in the CHS, 1,624 (66.1%), 421 (17.1%), and 412 (16.8%) were classified as expected agers, premature frailers, and adapters, respectively (Supplementary Table S1). The Pearson’s correlation between two sets of residuals was 0.96. The percentage of absolute agreement and Cohen’s kappa between two classifications of resilience was 88.0% and 0.76%, respectively. Resilience classification was robust to alternations of the regression model (Supplementary Tables S2 and S3).

Table 2.

Adjusted Association of Demographics and Clinical Disease with the SAVE Scores (N = 2,457)

	Value, N (%)	Beta (95% CI) Adapted from the CHS	Beta (95% CI) Derived Using the HABC Data	p-Value
Self-rated health
Excellent	363 (14.8)	−0.97 (−1.19, −0.75)	−0.96 (−1.19, −0.72)	<.001
Very good	784 (31.9)	−0.53 (−0.66, −0.41)	−0.56 (−0.74, −0.37)	<.001
Good	958 (39.0)	1.00 (reference)	1.00 (reference)	<.001
Fair	343 (14.0)	0.90 (0.76, 1.05)	0.54 (0.30, 0.79)	.030
Poor	9 (0.4)	1.78 (1.48, 2.08)	1.41 (0.13, 2.68)	<.001
Clinical disease
Parkinson’s disease	17 (0.7)	0.90 (0.33, 1.46)	1.07 (0.15, 1.98)	.022
Dementia	14 (0.6)	0.59 (0.31, 0.86)	1.23 (0.22, 2.23)	.017
Stroke	60 (2.4)	0.51 (0.29, 0.73)	1.08 (0.58, 1.57)	<.001
Claudication	93 (3.8)	0.47 (0.18, 0.76)	0.32 (−0.08, 0.73)	.120
Diabetes mellitus	385 (15.7)	0.39 (0.21, 0.57)	0.35 (0.14, 0.57)	.001
Osteoporosis	250 (10.2)	0.37 (0.21, 0.54)	0.10 (−0.16, 0.37)	.446
Kidney disease	35 (1.4)	0.36 (0.03, 0.70)	0.74 (0.10, 1.38)	.023
Depression	166 (6.8)	0.36 (0.16, 0.56)	0.52 (0.21, 0.83)	.001
Congestive Heart failure	90 (3.7)	0.34 (0.12, 0.56)	0.78 (0.36, 1.20)	<.001
Arthritis	1,269 (51.7)	0.32 (0.21, 0.42)	0.28 (0.12, 0.43)	.001
Cancer	535 (21.8)	0.18 (−0.35, 0.39)	0.03 (−0.16, 0.21)	.773
Eye disease	363 (14.8)	0.04 (−0.13, 0.21)	0.14 (−0.07, 0.36)	.191
COPD	269 (11.0)	0.02 (−0.13, 0.16)	0.03 (−0.21, 0.28)	.796
Coronary heart disease	597 (24.3)	−0.10 (−0.24, 0.03)	0.09 (−0.11, 0.28)	.380
Number of medications, mean ± SD	2.9 ± 2.6	0.11 (0.08, 0.13)	0.04 (0.01, 0.07)	.024
Age, years, mean ± SD	75.1 ± 2.9	0.12 (0.10, 0.13)	0.14 (0.08, 0.20)	<.001
Age2		0.0046 (0.0025, 0.0067)	−0.00097 (−0.01293, 0.01098)	.873
Age3		−0.0003 (−0.00043, −0.00016)	−0.00012 (−0.00359, 0.00335)	.947
Male	1,235 (50.3)	0.01 (0.10, 0.12)	−0.17 (−0.34, −0.01)	.035
Black	958 (39.0)	0.32	−0.03 (−0.20, 0.14)	.703
Intercept		3.46	4.60 (4.37, 4.83)	<.001
R2		0.08	0.17

Note: CHS = Cardiovascular Health Study, COPD = Chronic obstructive pulmonary disease, HABC = Health aging and body composition, SAVE = Scale of Aging Vigor in Epidemiology.

COPD: chronic bronchitis, emphysema, and asthma (in our study COPD included chronic bronchitis, emphysema, and asthma).

There were no differences in age, sex, and race between the three resilience groups due to the inclusion of these demographic variables in the regression model (Table 3). Education level, smoking, and BMI were also similar between the three groups.

Table 3.

Baseline Characteristics by Resilience Groups (N = 2,457)

	Premature Frailers (n = 415)	Expected Agers (n = 1645)	Adapters (n = 397)	p-Value
	Mean ± SD or N (%)
Memphis (vs Pittsburgh)	173 (41.7)	812 (49.4)	230 (57.9)	.001
Age (years)	75.2 ± 2.9	75.0 ± 2.9	75.3 ± 2.8	.168
Female	210 (50.6)	812 (49.4)	200 (50.4)	.561
Black (vs White)	164 (39.5)	630 (38.3)	164 (41.3)	.587
Education (years)	13.2 ± 3.1	13.1 ± 3.2	13.1 ± 3.2	.417
Smoking				.073
Never	158 (38.1)	656 (39.9)	165 (41.6)
Former	211 (52.5)	847 (51.5)	218 (53.2)
Current	39 (9.4)	142 (8.6)	21 (5.3)
Body mass index (kg/m2)	27.4 ± 5.2	27.1 ± 4.8	27.1 ± 4.3	.484

Association Between Resilience and Health Outcomes

We observed a steep increase in YAL, YHL, and YHAL from premature frailers to adapters (Figure 2). For instance, the average YHAL was 5.1, 7.7, and 9.1 years among premature frailers, expected agers, and adapters, respectively (p for comparison <.001). Table 4 shows the rates of ADL disability, hospitalization, and mortality, and the proportion surviving to 90 years old by the three resilience groups. Compared with premature frailers and expected agers, adapters had significantly lower rates of incident ADL disability, hospitalization, and all-cause mortality. Moreover, adapters had the highest proportion surviving to 90 years (39.7%) while premature frailers had the lowest (20.4%). The association between resilience and each health outcome persisted after multivariable adjustment (Supplementary Table S4). In the multivariable model where we additionally adjusted for frailty, premature frailers had a 37% (95% CI: 9%, 57%) lower odds of surviving to 90 years than expected agers. The odds ratio of survival to 90 years for adapters compared to expected agers was 1.38 (95% CI: 0.98, 1.95).

Figure 2.

Years of able, years of healthy life, and years of healthy and able life according to resilience group. p Value for each of three comparisons across resilience group was <.001.

Table 4.

Unadjusted Association Between Resilience and Outcomes

	Premature Frailers	Expected Agers	Adapters	p-Value
Incident ADL disabilitya, per 1,000 PYs (95% CI)	142.3 (125.4, 161.6)	103.4 (97.3, 109.9)	85.9 (75.9, 97.1)	<.001
Incident hospitalizationa, per 1,000 PYs (95% CI)	215.4 (194.8, 238.2)	158.6 (150.5, 167)	145.9 (131.1, 162.3)	<.001
Mortalitya, per 1,000 PYs (95% CI)	72.3 (64.6, 80.9)	53.7 (50.5, 57.1)	46.1 (40.4, 52.6)	<.001
Survival to 90 yearsb, n (%)	71 (20.4)	383 (30.6)	119 (39.7)	<.001

Note: ADL = Activities of daily living, PY = Person-year.

^aUnadjusted Cox model.

^bOnly persons who died or were at least 90 years of age at the end of the study period were included in the analysis for survival to 90. Chi-square test was used for univariate analysis.

Association Between Physiological Measures and Resilience

In the univariate analyses, IL-6, cystatin C, and AAI were associated with resilience at a significance level of p < .10 (Supplementary Table S5). Mutually adjusting for these three biomarkers in the multinomial regression model showed that the lowest-risk group of cystatin C was associated with a higher likelihood of being in the “Adapter” group than in the “Expected Ager” group (Table 5). The lowest-risk group of cystatin C and AAI was associated with a lower likelihood of being in the “Premature Frailer” group than in the “Expected Ager” group.

Table 5.

Associations Between Physiological Measures and Three Resilience Groups

Physiological Measure	“Premature Frailers” Relative to “Expected Agers”	“Adapter” Relative to “Expected Agers”
	Relative Risk (95% confidence interval)
Interlukin-6 (pg/mL)
≥1.47	Ref.	Ref.
<1.47	0.82 (0.59, 1.09)	1.08 (0.82, 1.47)
Cystatin C (mg/L)
≥0.86	Ref.	Ref.
<0.86	0.74 (0.54, 1.00)	1.43 (1.09, 1.87)
Ankle-arm index
<0.9	Ref.	Ref.
0.9–1.0, or >1.4	0.86 (0.55, 1.36)	1.23 (0.72, 2.11)
1.01–1.4	0.69 (0.49, 0.97)	1.43 (0.95, 2.16)

Note: All three risk factors were modeled simultaneously. Each risk factor was associated with resilience in the univariate analysis at a significance level of 0.1.

Discussion

In this study, we validated a novel approach for quantifying and classifying physical resilience, modified to data available in a high functioning white and black older cohort of the Health ABC Study. We found that adapters—persons whose frailty level is lower than expected from their biological features and disease burden—had longer healthy life spans and lower rates of disability, hospitalization, and mortality than expected agers. In contrast, premature frailers—persons with worse-than-expected frailty—had higher rates of adverse outcomes and spent more of their remaining life feeling less healthy and disabled.

Physical resilience is theoretically defined as one’s ability to recover and restore optimal function following stressors (3,10). In the absence of a gold standard (e.g. specific biomarkers), physical resilience has been operationally defined in different ways (3,11–14). Colon-Emeric and colleagues proposed two approaches for measuring physical resilience based on two different conceptual frameworks. The “recovery phenotype” approach describes how quickly and completely a person recover from a physical stressor, while the “expected recovery differential” approach identifies persons who recover faster than expected based on their demographics and health status (under review). Alternatively, Arbeev and colleagues conceptualized resilience as physiological dysregulation—deviation from a normal physiological state—and found that deviations of physiological markers from their normal states were associated with higher mortality and shorter survival after the onset of cancer, cardiovascular disease, and diabetes (12). Our approach for quantifying and classifying resilience is derived from the theory that persons who are resilient have the intrinsic ability to adapt to and mitigate the consequences of cumulative damage in organ systems. In this sense, the degree of mismatch between persons’ frailty level and disease burden is suitable to quantify their physical resilience. Our approach for quantifying and classifying resilience can be used to build clinical prediction tools to predict recovery, which is critical for setting patient-centered recovery goals and the effective use of medical resources. The resilience phenotype we created may also assist in the identification of older adults needing the greatest amount of care in the hospital and skilled nursing facility. Similar to the physiological dysregulation approach, the measurement of physical resilience in our study does not rely on health or functional trajectories following a stressor. This is an intriguing feature because we could identify risk and protective factors to design targeted interventions for promoting resilience prior to the occurrence of a stressor.

We revealed several biomarker correlates of physical resilience. The lowest-risk group of cystatin C was associated with higher chance of being an adapter than an expected ager; the lowest-risk group of cystatin C and AAI was associated with lower likelihood of being a premature frailer than an expected ager. These results were echoed by our prior work in the CHS, in which adapters had better markers of kidney function and subclinical disease than expected agers and premature frailers (4). However, we need to be cautious in interpreting these results because the main goal of our study was to develop and validate a new measure of physical resilience and these risk factor analyses were conducted in an exploratory fashion. Future research is needed to elucidate whether these markers may serve as targets of interventions for promoting physical resilience.

We acknowledge several limitations. First, physical resilience, which is likely to be a dynamic phenomenon (15), was only measured once in our study. This may potentially lead to measurement error and misclassification. The availability of repeated measures of frailty and diseases in longitudinal cohorts, including the Health ABC Study, provides an opportunity to characterize the trajectories of physical resilience over time. Second, we used YHL, longevity, and incidence of adverse events to evaluate the predictive validity of our physical resilience measure. This validation approach may preclude us from differentiating resilience from robustness, a connected but conceptually distinct entity that refers to the ability to resist deviation from normal state and the onset of adverse health events (16). As suggested in a review article on physical resilience, the gold standard for the validation of a resilience measure is to examine the trajectories of function following a well-defined stressor (10). The Health ABC Study is an ideal setting to pursue this endeavor in the future due to the availability of adjudicated events, repeated measures of physical and cognitive functions, and a long follow-up period. Third, a core of measurements on body composition, a wide range of behavioral and physiological measures, and omics data (eg, genomics and metabolomics) are available in the Health ABC Study; however, we only examined a few biomarkers previously found to be associated with measures of healthy aging and frailty (17–20). Future studies of other physiological, behavioral, and social factors may help further advance our understanding of the underlying mechanisms of physical resilience among older adults. Fourth, residual confounding is inevitable because we did not use all possible diseases to model frailty in the linear regression analysis. Fifth, we used a statistical approach to define resilience based on an individual’s degree of mismatch between frailty level and disease burden, which may seem difficult to apply clinically. However, constructing this resilience measure is only slightly more complex than quantifying the frailty level, which is becoming increasingly available in geriatric practice settings. It is also noteworthy that the level of agreement between two classifications of resilience using regression coefficients derived from two independent studies was high, demonstrating model robustness across studies. Lastly, we assessed resilience based on the degree of mismatch between frailty level and disease burden, both of which may change over time but were only measured once in our study. Resilience has been viewed as a dynamic rather than static process and may involve two components—speed and completeness of recovery after a stressor (16,21). Using data from a cohort study with annual follow-ups, we were only able to quantify the level of resilience at a given time point. We need to collect more data in a clinical setting, where it is feasible to have much more frequent measurements, to monitor the dynamic change in resilience over a short time period.

In summary, we validated a novel approach for quantifying and classifying physical resilience in a cohort of well-functioning white and black older adults. Persons with high physical resilience level—frailty level is lower than expected from biological features and disease burden—had longer healthy life span and lower rates of multiple adverse outcomes. In addition, older adults can maintain relatively high function in the presence of organ impairments had better markers of different body systems. Future research is needed to provide a comprehensive physiological, behavioral, and psychosocial profile of these adapters. Such an endeavor may help design interventions that aim to promote resilience provide us a better understanding of pathways leading to successful aging.

Funding

This work was supported by a Suzhou Municipal Science and Technology Bureau (SS2019069).

Conflict of Interest

C.W. provides paid consultant services to HealthKeeperS, a health data analytics company in China.