Progression of Physical Frailty and the Risk of All-Cause Mortality: Is There a Point of No Return?

Abstract

OBJECTIVES:

To investigate the rate and patterns of accumulation of frailty manifestations in relationship to all-cause mortality and whether there is a point in the progression of frailty beyond which the process becomes irreversible and death becomes imminent (a.k.a. point of no return).

DESIGN:

Longitudinal observational study.

SETTING:

Community or a non-nursing home residential care setting.

PARTICIPANTS:

Two thousand five hundred and fifty seven robust older adults identified at baseline in 2011 with follow-up for all-cause mortality between 2011 and 2018.

MEASUREMENTS:

Frailty was measured by the physical frailty phenotype. Cox models were used to study the relationships of the number of frailty criteria (0–5) at each point in time and its accumulation patterns with all-cause mortality. Markov state-transition models were used to study annual transitions between health states (i.e., frailty, recovery, and death) after becoming frail among those with frailty onset (n = 373).

RESULTS:

There was a nonlinear association between greater number of frailty criteria and increasing risk of mortality, with a notable risk acceleration after having accumulated all five criteria (hazard ratio (HR) = 32.6 vs none, 95% confidence interval (CI) = 15.7–67.5). In addition, the risk of one-year mortality tripled, and the likelihood of recovery (i.e., reverting to be robust or pre-frail) halved among those with five frailty criteria compared to those with three or four criteria. A 50% increase in mortality risk was also associated with frailty onset without (vs with) a prior history of pre-frailty (HR = 1.51, 95% CI = 1.20–1.90).

CONCLUSION:

Both the number and rate of accumulation of frailty criteria were associated with mortality risk. Although there was insufficient evidence to declare a point of no return, having all five-frailty criteria signals the beginning of a transition toward a point of no return. Ongoing monitoring of frailty progression could aid clinical and personal decision-making regarding timing of intervention and eventual transition from curative to palliative care.

INTRODUCTION

Frailty is a debilitating medical condition that places a person in a state of increased systemic vulnerability to intrinsic (e.g., disease pathology, acute infection) or extrinsic (e.g.,environmental stressors) insults. It is posited that the progression of frailty consists of a series of transitions between states of dynamic equilibrium of decreasing cellular, molecular, and physiologic integrity even before reaching a critical threshold beyond which frailty becomes clinically observable.¹ Although frailty appears to be reversible,^2,3 with progressive severity, a complete remission becomes an increasingly rare event in the absence of an intervention.² If the syndrome of frailty is progressive in nature, like many other chronic diseases and conditions of aging, it is important to know whether there is a point in the progression of frailty beyond which the process becomes irreversible and death becomes imminent (a.k.a. point of no return). In addition, given that the entry point into the “cycle of frailty” tends to be heterogeneous, with studies showing either muscle weakness or exhaustion to be the most common first manifestation^4,5 it is worth investigating whether specific patterns of emergence of frailty manifestations is as relevant as the number of manifestations in determining the point of no return.

To evaluate these questions, we assessed the relationship between frailty progression and the risk of all-cause mortality among older adults aged 65 and older in the United States. Specifically, we evaluated progression in the context of the clinical syndrome of frailty.^6,7 Our primary hypothesis is that there is a threshold relationship between number of frailty manifestations and mortality risk. In addition, we hypothesize that the strength of the association depends on the rate and pattern of accumulation of frailty criteria. The hypothesized relationship has implications for clinical prognostication, potentially guiding clinical and personal decision-making regarding timing of intervention and eventual transition from curative or life-prolonging care to palliative care.

METHODS

Study Design and Participants

The National Health and Aging Trends Study (NHATS) is a prospective study of older adults designed to advance understanding of national trends in late-life functioning and the consequences of late-life disability for individuals, families, and society.⁸ The original study cohort consists of 8,245 men and women aged 65 and older who were selected from the Medicare enrollment file as of September 30, 2010, and were living in the contiguous United States at study entry. The weighted NHATS sample is nationally representative of the 65 and older Medicare population.⁹ Study participants were contacted every year starting in 2011. The analytic sample for this study is comprised of 2,557 robust older adults identified using the physical frailty phenotype⁶ who were living in the community or a non-nursing home residential care setting in 2011 (Supplemental Figure S1). The participant’s death including month and year of death was reported by informants during attempts to contact the participant for the annual interview.

Frailty Assessment

Frailty was assessed using the physical frailty phenotype (PFP) that was developed¹⁰ and then operationalized, validated,⁶ and cross-validated.⁷ Although many frailty assessment tools exist, we chose the PFP because of its validation and emerging consensus as a medical syndrome,^7,11 its dominance in the research literature,¹² and the biologically-based theory and specificity of hypothesized biological underpinnings embedded in the PFP that is conducive to studying mechanisms and physiological etiology.¹³ The PFP used in this study was identical to that of a previous study in the same cohort,¹⁴ and it included five binary criteria (1) exhaustion, (2) low physical activity, (3) slow gait, (4) muscle weakness, and (5) unintentional weight loss. For criteria (3) and (4), those not tested because of safety concerns, recent surgery, pain, or who attempted but were unable to complete a test, were scored as “1” following recommended practice^15,16. We classified subjects meeting three or more criteria as frail, one or two as pre-frail, and none as robust. Frailty status was treated as missing if more than two criteria were missing. Frailty status was ascertained at baseline and each subsequent annual visit with a maximum of eight visits between 2011 and 2018.

Covariates

Demographic characteristics include age, sex, race/ethnicity (white non-Hispanic, black non-Hispanic, Hispanic, other), education (less than high school, high school or equivalent, more than high school) and residence (community vs residential care). Prevalent and incident diseases and conditions were assessed at baseline and during study follow-up based on self-report of doctor’s diagnosis of arthritis, diabetes mellitus, heart disease, high blood pressure, lung disease, osteoporosis, stroke, hip fracture, and cancer.

Statistical Analysis

To test the hypothesis of point-of-no-return, we first calculated mortality rate per 100 person years and plotted age-adjusted Kaplan–Meier survival curve by the number of frailty criteria accumulated. Next, the number of frailty criteria at any point in time (termed frailty score hence-forth) between 2011 and 2017 was modeled as a time-varying covariate (via dummy variables with zero as the reference group) in a Cox model to examine its relationship with risk of mortality between 2011 and 2018. Finally, Markov state-transition model was used to examine between-visit (one-year) changes in health following frailty onset. This analysis was restricted to the subset of those who became frail during the study follow-up (n = 373). The model estimated the probability of either transitioning from a frailty score of 3, 4, or 5 at Visit j_(i) to a state of death or recovery (i.e., frailty score < 3), or maintaining the frail state (i.e., frailty score ≥ 3) at Visit j_(i) + 1, where j_(i) denote subject i's sequence of visits from time of frailty onset to recovery, death, or end of follow-up. Both time-independent (gender, race, and education) and time-varying confounders (age and number of diseases measured at Visit j_(i)) were adjusted in the model. In both the Cox model and the Markov model, we view threshold effect as a special case of nonlinearity characterized by the abruptness of change in mortality or recovery risk associated with change in frailty score. Specifically, we defined abruptness as discontinuity in the first derivative of the risk curve. Evidence that supports the hypothesis of point of no return would be a finding of a threshold count of frailty manifestations after becoming frail beyond which both the risk of death is greatly elevated and the likelihood of recovery is greatly diminished.

Next, we tested our secondary hypothesis that the strength of the frailty-mortality association depends on the rate and pattern of accumulation of frailty criteria. First, we distinguished two patterns of frailty incidence: frailty onset with vs without prior history of pre-frailty onset. To test the hypothesis that the latter signals a catastrophic decline and therefore is more strongly associated with mortality, we defined a time-varying categorical variable coded as 1 if frailty onset following pre-frailty onset, 2 if frailty onset without prior history of pre-frailty, and 0 otherwise; and its association with mortality was assessed via dummy variables in a Cox model. Next, we summarized the changes in frailty score over time between 2011 and time of death or censoring into three mutually exclusive patterns: being robust throughout (termed “stable”), both increase(s) and decrease(s) (termed “fluctuating”), and increase at least once without a decrease at a later time (termed “monotone”). To address potential length bias in the sense that a minimum of three observations are needed to characterizing a “fluctuating” pattern, we restricted the analysis to those with at least three frailty measurements between 2011 and 2018 (n = 1,805).

In addition, we also examined the associations of mortality risk with patterns of first manifestations that include having: (1) weakness, slowness, and/or low activity first, (2) exhaustion and/or weight loss first, or (3) co-occurrence of one or more criteria in (1) with one or both criteria in (2). We constructed these patterns to test a priori hypothesis that early development of weight loss or exhaustion (i.e., patterns ii and iii) signals more rapid progression to frailty and therefore a strong predictor of mortality.⁴ We restricted this analysis to those who were robust at baseline and developed pre-frailty or frailty during the study (n = 1,556). Cox models were used with time origin being the time of the first onset of frailty manifestations. All Cox models adjusted for age, race, gender, education, number of prevalent diseases at baseline, and number of incident diseases during the study follow-up. Supplementary Table S1 provides details of the statistical models used to test the study hypotheses. To account for oversampling of the oldest age groups, Black non-Hispanic persons, and nonresponse in NHATS, sampling weights were incorporated in all analyses. STATA version 15 was used for all analyses.

RESULTS

Of the 2,557 study participants who were robust at baseline, 97% were living in the community, 85% were non-Hispanic white, and 6.6% were non-Hispanic black, 14% and 26% had less than high school and high school or equivalent education, respectively. The sample, on average, were 73 years old with two chronic diseases (Table 1). The median follow-up time was 6.2 years; and 1,511 (12.2 per 100 person-years) developed pre-frailty, 373 (3.0 per 100 person-years) developed frailty, 325 (2.6 per 100 person-years) died between 2011 and 2018 (Supplementary Table S2).

Table 1.

Distribution of Baseline Characteristics of the Robust NHATS Participants in 2011 (N = 2,557)^a

Age, years, Mean ± SD	73.2 ± 6.9
Race, no (%)
Non-Hispanic white	1897 (84.7)
Non-Hispanic black	463 (6.6)
Hispanic	113 (4.9)
Other	84 (3.8)
Type of residence, n (%)
Community	2,469 (97.1)
Residential care	88 (2.9)
Education, n (%)
Less than high school	440 (13.6)
High school or equivalent	675 (25.7)
More than high school	1,436 (60.7)
Diseases
Heart disease, n (%)	290 (11.2)
Hypertension, n (%)	1,519 (56.3)
Arthritis, n (%)	1,065 (40.9)
Osteoporosis, n (%)	389 (15.6)
Diabetes mellitus, n (%)	482 (17.3)
Lung disease, n (%)	234 (9.2)
Stroke, n (%)	141 (4.8)
Cancer, n (%)	613 (23.9)
Hip fracture, n (%)	63 (2.1)
Number of diseases, Mean ± SD	1.88 ± 1.28
0	361 (15.3)
1 or 2	1,452 (57.5)
3 or 4	666 (24.4)
5 or more	78 (2.8)

^aWeighted mean (standard deviation) or weighted prevalence are presented.

Frailty Score and Mortality

The crude mortality rate increased with increasing number of frailty criteria with the mortality rate for frailty score of 5 being more than three times higher than that of frailty scores of 3 and 4 (Figure 1a). The inflection point (i.e., discontinuity) at frailty score = 4 is suggestive of a threshold effect. The age-adjusted Kaplan–Meier curves shows a similar trend (Supplementary Figure S2). On a relative risk scale, the Cox model found a nonlinear growth of relative hazard of mortality with increasing number of frailty criteria (Table 2, Model 1). Specifically, the mortality risk was 10, 7, and 33-fold higher among those who had accumulated three, four, and all five frailty criteria at any point in time, respectively, than that of those who remained robust after covariate adjustment. However, no evidence of a threshold effect was detected on the log-hazard ratio scale. Among the 21 subjects who reached a frailty score of 5 during the follow-up, 10 (48%) died within 6 months on average after having accumulated five frailty criteria (median 5 months), compared to 19–20 months on average for frailty scores 3 and 4 (Supplemental Table S3).

Figure 1.

(a) Crude mortality rate per 100 person years (95% confidence interval) after reaching frailty score of 1, 2,…, 5 for the first time during study follow-up; and (b) estimated probability and 95% confidence interval (on the y-axis) from the Markov model of between-visit transition over a one-year interval from a frailty state with 3, 4, or 5 criteria (on the x-axis) to a state of death, or recovery (i.e., number of frailty criteria <3) or maintaining the frail state (i.e., number of frailty criteria ≥3).

Table 2.

Crude Mortality Rate and Hazard Ratio of All-Cause Mortality by Type of Frailty Risk Exposure Among Those Who Were Robust at Baseline

Type of Exposure		Crude HR (95% CI)	Adjusteda HR (95% CI)	Crude mortality rate per 100 person-years (95% CI)b
Model 1. Number of frailty criteria at any point in time (n = 2,557)	0	Ref	Ref	2.31 (1.78, 2.84)
	1	2.80 (2.03, 3.85)	2.18 (1.58, 2.99)	3.78 (3.22, 4.23)
	2	4.59 (3.21, 6.57)	3.38 (2.35, 4.87)	6.79 (5.62, 7.96)
	3	16.00 (11.07, 21.98)	10.28 (7.15, 14.77)	13.46 (10.71, 16.21)
	4	12.86 (7.47, 22.14)	6.94 (3.97, 12.13)	14.34 (9.29, 19.39)
	5	67.79 (31.56, 145.59)	32.57 (15.73, 67.45)	50.65 (19.29, 82.06)
Model 2. Rate of frailty onset (n = 2,557)	No Frailty onset (n = 2,184)	Ref	Ref	2.12 (1.84, 2.41)
	Pre-frailty onset before frailty onset (n = 274)	2.73 (2.09, 3.58)	1.97 (1.47, 2.64)	3.96 (3.03, 4.89)
	Frailty onset without prior pre-frailty onset (n = 99)	4.32 (3.43, 5.44)	2.97 (2.29, 3.85)	20.12 (16.16, 24.08)
Model 3: Pattern of change in frailty score over timec (n = 1,805)d	Stable (n = 390)	Ref	Ref	1.40 (0.91, 1.89)
	Fluctuating (n = 1,061)	1.13 (0.72, 1.78)	0.83 (0.52, 1.31)	1.67 (1.36, 1.97)
	Monotone (n = 354)	4.59 (2.86, 7.38)	2.80 (1.68, 4.66)	5.30 (4.25, 6.35)

Abbreviations: CI, confidence interval; HR, hazard ratio.

^aAdjusted by age, race, gender, education, number of prevalent diseases at baseline, and number of incident diseases during the study follow-up.

^bIn Model 1, crude mortality rate was calculated with time origin being baseline for those who remained robust before death or censoring, and for those with frailty progression, the time of reaching frailty score of 1, 2,..., 5 for the first time during study follow-up.

^c“Stable”: being robust throughout, “Fluctuating”: both increase(s) and decrease(s) in frailty score, “Monotone”: increase in frailty score at least once without a decrease at a later time during study follow-up.

^dRestricted to the subset with at least three measurements of frailty during the study follow-up.

Regarding the likelihood of recovery among those who became frail during the study follow-up (n = 373), the Markov state-transition model estimated that the probability of death tripled and the probability of recovery in the meantime halved within a year if one’s current frailty score was five compared to a frailty score of three or four after adjusting for covariates (Table 3, Figure 1b). Within the category of recovery, the likelihood of frailty score reverting to 0 (i.e., robust) after becoming frail was low, particularly among those with 4 or 5 criteria (Supplementary Table S4).

Table 3.

Probability of Either Transitioning From a Frail State to a Recovery (Non-Frail State), Maintaining the Frail State, or Evolving to Death in Those who Reached the Frail State (n = 373)^a

Previous Visit Status Number of Frailty Criteria	Current Visit Status
	Death	(95% CI)	Frailty	(95% CI)	Recovery	(95% CI)
3	0.14	(0.09, 0.18)	0.33	(0.26, 0.39)	0.54	(0.47, 0.61)
4	0.11	(0.04, 0.19)	0.35	(0.25, 0.45)	0.54	(0.43,0.65)
5	0.38	(0.10, 0.66)	0.35	(0.05, 0.64)	0.28	(0.00,0.55)

^aBased on the Markov state transition model of the probability of either transitioning from a frailty state with 3, 4, or 5 criteria at Visit j_(i) (termed “previous visit”) to a state of death, or recovery (i.e., number of frailty criteria <3) or maintaining the frail state (i.e., number of frailty criteria ≥3) at Visit j_(i) + 1 (termed “current visit”), where j_(i) denote subject i’s sequence of visits beginning at time of frailty onset to end of follow-up. Both recovery and death marked the end of follow-up (i.e., absorbing states) in this analysis. The covariates adjusted in the model included age, gender, race, education, number of diseases; age and number of diseases were treated as time-varying, i.e., measured at the previous visit.

Pattern of Frailty Emergence with Mortality

Of the 373 subjects who became frail, 274 (73%) had a prior history of pre-frailty onset before frail onset; whereas 99 (27%) transitioned directly from being robust to frail. The Cox model found a 50% increase in mortality risk associated with frailty onset in the absence (vs presence) of a prior history of pre-frailty (hazard ratio (HR) = 1.51, 95% confidence interval (CI) = 1.20–1.90, see also Table 2, Model 2).

In addition to rate of frailty onset having an effect on mortality risk, we found that those exhibiting “Monotone” increase in frailty score over 7 years (n = 354) were almost three times more likely to die than those who remained robust (n = 390) after covariate adjustment (Table 2 Model 3: HR = 2.80, 95% CI = 1.68, 4.66). There was no significant difference between the “Stable” and the “Fluctuating” groups after covariate adjustment (HR = 0.83, 95% CI = 0.52, 1.31).

Next, we examined whether a priori chosen order of initial manifestations may differentiate the risk of mortality. The risk was comparable between having slowness, weakness, and/or low activity first and having weight loss and/or exhaustion first (adjusted HR = 1.09, 95% CI = 0.78–1.51). Mortality risk however increased by 74% when exhaustion and/or weight loss co-occurred with slowness, weakness, and/or low activity at the nascence of frailty manifestation, compared with having slowness, weakness, and/or low activity first (Table 4, Model A: adjusted HR = 1.74, 95% CI = 1.23–2.46). To account for the fact that the frailty score may vary within the former pattern of co-occurrence, we reran the model by subdividing the subset into having two vs three vs four or five incident frailty criteria. Interestingly, consistent with the results of Model 2 in Table 2, it was the number rather than the pattern of initial manifestations that was associated with increased risk of mortality. Specifically, those with three or more criteria co-occurring (i.e., frailty onset with no prior history of pre-frailty), had about three-fold increase in mortality risk (Table 4, Model B), compared with having slowness, weakness and/or low activity first.

Table 4.

Crude Mortality Rate and Hazard Ratio of All-Cause Mortality Comparing Patterns of First Incident Frailty Manifestations Among Those who were Robust at Baseline and Developed Pre-Frailty and or Frailty During Study Follow-Up Between 2011 and 2018 (n = 1,556)

Pattern of Initial Manifestation	Sample Size	Crude HR (95% CI)	Adjusteda HR (95% CI)	Crude Mortality Rate Per 100 Person-Years (95% CI)
SL and/or WK and/or LA first (1 criterion: n = 654, 88.0%; 2 criteria: n = 83, 11.2%; 3 criteria: n = 6, 0.8%)	743	Ref	Ref	2.49 (2.02, 2.95)
EX and/or WL first (1 criterion: n = 533, 92.5%; 2 criteria: n = 43, 7.5%)	576	0.99 (0.71, 1.38)	1.09 (0.78, 1.51)	2.26 (1.76, 2.75)
Model Component A: SL and/or WK and/or LA same time as EX and/or WL	237	2.00 (1.39, 2.87)	1.74 (1.23, 2.46)	4.92 (3.71, 6.12)
Model Component Bb: SL and/or WK and/or LA same time as EX and/or WL	237	1.21 (0.75, 1.94)	1.10 (0.70, 1.73)	3.22 (2.01–4.43)
2 incident criteria (n = 144, 60.8%) 3 incident criteria (n = 66, 27.9%)		3.23 (1.88, 5.57)	2.80 (1.70, 4.63)	7.92 (4.87, 10.96)
4 or 5 incident criteria (4 criteria: n = 23, 9.7%; 5 criteria: n = 4, 1.7%)		4.46 (2.45,8.12)	3.00 (1.50, 5.98)	8.15 (3.33, 12.98)

Abbreviations: EX, exhaustion; HR, hazard ratio; LA, low activity; SL, slowness; WK, weakness; WL, weight loss.

^aAdjusted by age, race, gender, education, number of prevalent diseases at baseline, and number of incident diseases before or at the time of initial frailty manifestation.

^bReplace Model component A with Model Component B by subdividing the group into those having two, three, and four or more incident frailty criteria.

DISCUSSION

In this large national population-based study, we found a nonlinear association between greater number of incident frailty criteria and increasing risk of mortality, with a notable risk acceleration in mortality rate upon accumulating five criteria. In addition, the risk of short-term mortality tripled, with one in three expected to die within 1 year, and the likelihood of recovery (i.e., reverting back to be robust or pre-frail) halved among older adults with five frailty criteria compared to those with three or four criteria. Taken together, these findings provide preliminary evidence that having all five frailty criteria represents what Scheffer termed a state of health “signaling a system’s approaching to a ’tipping point’ beyond which a minor trigger can invoke a self-propagating shift to a contrasting state,”¹⁷ where the “tipping point” and “contrasting state” represent point-of-no-return and death respectively in this study. Notwithstanding the qualitative evidence, given that only a small group of older adults reached high degree of frailty severity in the study and the rate of short-term recovery was non-negligible, there was insufficient evidence to declare a point of no return.

The association between frailty and mortality has been firmly established. However, the understanding of the accumulation of frailty manifestations as a dynamic process and its relationship with mortality is limited. To our knowledge, this is the first longitudinal study of such relationship in a representative sample of older adults in the United States. Given the mounting interest in translation of frailty research into clinical practice, we need to have a better understanding of the prognostic value of frailty assessment for older adults, particularly those with multiple comorbidities for whom disease-specific guidelines and recommendations could be inadequate and potentially harmful.¹⁸ As a measure of integrative function, frailty has been shown to confer added prognostic value beyond traditional clinical risk factors for adverse health outcomes in diverse patient populations.^19–21 Our finding of an accelerated growth of mortality risk suggests that as frailty progresses, at some critical tipping point, the physiological systems may become overwhelmed and can no longer harness the resources needed to maintain integrity, leading to an irreversible decline that ends in death. The observation of a sharp demarcation of mortality risk when all five frailty criteria are met therefore may signal the beginning of a critical descent toward a point of no return. Such finding provides medical providers, patients and their caregivers the objective data needed to start the conversation and planning for a transition from curative treatment to palliative care by focusing on relief of symptoms instead of treating underlying conditions. This is also consistent with older adults’ preference for dignity, independence, and quality of life over longevity with respect to end of life care.²² It could be argued that palliative care should be incorporated into active curative treatment as soon as one crosses the clinical threshold of being frail (i.e., having three criteria). That way, a phased transition can take place to ensure continuity of care that would be difficult to achieve otherwise if a referral to specialist palliative care services only happens at severe stage of frailty, causing abrupt change in care provider, location and goals.²³ Evidence suggests that continuity of care is associated with greater patient and family satisfaction.²³

Regarding the patterns of emergence of frailty manifestations and their relationships with mortality, contrary to our hypothesis prompted by findings from an early study,⁴ we did not find evidence of specific frailty criteria occurring alone or co-occurring that had differential impact on mortality risk. Instead, it was the co-occurring of three or more criteria at the outset of frailty manifestation (i.e., skipping pre-frailty phase) that was associated with elevated risk of mortality. This suggests that there are both progressive and catastrophic pathways to frailty that may be differentially associated with mortality risk. In this study, the prevalence of hip fracture, cancer, stroke, and heart disease was the highest among those on the “catastrophic” pathway (Supplementary Table S5). In addition, incident diseases such as stroke and heart disease were more likely to coincide with the catastrophic onset of frailty than the progressive type despite lower overall incidence in the former (results not shown). Therefore, the catastrophic onset of frailty might have been driven primarily by severe health events and diseases (aka. Secondary frailty²⁴). Distinguishing the catastrophic from progressive pathway has important clinical implications with regard to care planning and prioritization. In the case of catastrophic pathway, the focus of care arguably should remain curative and palliative care is likely to come later if the underlying disease and associated frailty continue to deteriorate.

When examining changes in frailty score over time, we found that almost 60% had both an increase and a decrease in frailty score, which is consistent with observations from other studies that frailty is a dynamic process and it may involve both progressive and regressive transitions.^2,3 Our finding that the “fluctuating” group did not experience increased mortality relative to the “stable” group at first seemed counterintuitive. One possible explanation is imperfect measurement (i.e., measurement error), and therefore bears no clinical significance relative to stable trajectory. Alternatively, we may view stability as an indicator of robustness (i.e., resistance to stressors) and fluctuation as an indicator of physical resilience (i.e., ability to recover in the face of adversity in life). Without information on the two key determinants of robustness and resilience, i.e., magnitude of a stressor and physiological reserve (i.e., “the potential capacity of a cell, tissue, or organ system to function beyond its basal level in response to alterations in physiologic demands²⁵”), the relative magnitude of their associations with mortality however could not be determined. For example, the ability to recover is expected to depend on the stress level for a given level of reserve. In fact, a closer examination of the “fluctuating” group revealed that the majority (74%) had at most 1–2 frailty criteria at any point in time; 17%, 8%, and 1% had three, four, and five frailty criteria, respectively, suggesting that recovery became increasingly difficult after becoming frail. Consistent with this observation, the Markov model showed that greater severity of frailty was associated with greater short-term mortality and lower likelihood of recovery. Taken together, our findings provide additional evidence that frailty represents a state where one’s ability to withstand, compensate and overcome stressors is greatly compromised.

The strengths of this study include a nationally representative sample of older adults, longitudinal data with up to eight observations of frailty status per person, a validated physical frailty phenotype that was used previously to derive a nationally representative profile of frailty in the United States,¹⁴ and a comprehensive analysis of both the number and pattern of accumulation of frailty criteria to facilitate nuanced discovery. This study also has limitations. First, it is possible that dropouts might be due to health reasons such that the last observation of frailty status before dropout might not reflect the frailty status before death. In this study, 36 (11%) of the 325 subjects who died between 2011 and 2018 had a gap time greater than 1 year between the time of death and last available frailty observation. Given that six of the 36 (16.7%) missed frailty assessment due to nursing home admission, compared to 0.5% among the survivors with incomplete frailty follow-up through 2018, our overall inference might be conservative because the frailty score post nursing-home entry would have been higher than the last available observation before nursing home admission. Second, if frailty is a unique clinical entity, it is important to distinguish frailty manifestations from the signs and symptoms of dying. The length of this terminal decline however is heterogeneous depending on the cause of death,²⁶ which makes it challenging to set a cutoff a priori for the time gap between frailty assessment and death. Third, the small set of discrete frailty states limited our ability to formally test threshold effect as defined in this study (i.e., discontinuity in the rate of change). Therefore, although our inference was based on collective evidence rather than individual results that included both risks of death and recovery, an awareness of subjectivity in the interpretation of our findings is warranted. Finally, given the limited duration of follow-up and that a small number of study subjects reached the highest severity of frailty, we were unable to reliably estimate mortality rate in this group; validation of our findings therefore is needed.

In summary, we have offered evidence that having all five physical frailty criteria is an important phase in the progression of frailty associated with both substantial increase in risk of all-cause mortality and decrease in likelihood of recovery. It may therefore signal a critical transition toward a point of no return beyond which the process becomes irreversible and death becomes imminent. In addition, we found further evidence of both catastrophic and progressive pathways to frailty with the former most likely driven by disease-specific cause. Therefore, both the number and rate of accumulation of frailty manifestations may be important in determining optimal care upon transition to frailty.

Supplementary Material

Supporting Information

Supplementary Figure S1: Flowchart detailing sample exclusion and inclusion criteria.

Supplementary Figure S2: Age-adjusted* Kaplan–Meier survival probability curve with time origin being baseline for those who remained robust before death or censoring or the time of reaching frailty score of 1, 2, …, 5 for the first time during study follow-up for those with frailty progression.

Supplementary Table S1: Summary of statistical models used to test the primary and secondary study hypotheses.

Supplementary Table S2: Incidence of pre-frailty, frailty, and death at each follow-up visit (year 2012–2018): the ratio of the number of incident events at visit j to the total number at risk at visit j – 1 (% event), j =2,3,…,8, are presented.

Supplementary Table S3: Number of study subjects at risk, died, and among those who died the observed mean and median time to death after reaching frailty score of 1, 2,…, 5 for the 1st time compared to those who remained robust during the study follow-up between 2011 and 2018.

Supplementary Table S4: Probability of transitioning from a frailty state with 3, 4, or 5 criteria at Visit j – 1 (termed “previous visit”) to a state of death, or frailty score of 5, 4, 3, 2, 1, or 0 at Visit j (termed “Current visit”), j = 1, 2, …, 8. Transition probabilities were estimated from the Markov transition model fit to the subset of study subjects with frailty onset (n = 373).

Supplementary Table S5: Summary of baseline demographic characteristics and prevalent and incident diseases in the study sample by pattern of frailty onset (n = 2,557)

Key Points.

• Meeting all five physical frailty criteria signals the beginning of a transition toward a point of no return beyond which the process becomes irreversible and death becomes imminent.

Why Does This Paper Matter?

This study strengthens evidence base for clinical and personal decision-making regarding optimal care upon transition to frailty.