Principles and Issues for Physical Frailty Measurement and Its Clinical Application

Abstract

Introduction

“Frailty” has attracted attention for its promise of identifying vulnerable older adults, hence its potential use to better tailor geriatric health care. There remains substantial controversy, however, regarding its nature and ascertainment. Recent years have seen a proliferation of frailty assessment methods. We argue that the development of frailty assessments should be grounded in “validation”—the process of substantiating that a measurement accurately and precisely measures what it intends, identify unresolved measurement issues, and highlight measurement-related considerations for clinical practice.

Methods

Principles for validating frailty measures are elucidated. We follow principles—articulated, for example, by Borsboom—in which a construct must be clearly defined and then analyses undertaken to substantiate that a measurement accurately and precisely measures what it intends. Key elements are content validity, criterion validity, and construct validity, with an emphasis on the latter.

Results

We illustrate the principles for a physical frailty phenotype construct.

Conclusions

Unresolved conceptual issues include the roles of intersecting concepts such as cognition, disease severity, and disability in frailty measurement, conceptualization of frailty as a state versus a continuum, and the potential need for dynamic measures and systems concepts in furthering understanding of frailty. Clinical considerations include needs to distinguish interventions designed to address frailty “symptoms” versus underlying physiology, improve “prefrailty” measures intended to screen individuals early in their frailty progression, address feasibility demands, and further visioning followed by rigorous efficacy research to address the landscape of potential uses of frailty assessment in clinical practice.

With a burgeoning older population, chronic diseases, disability, and adverse health events will exert an increasing footprint on individuals, families, and the public. Concurrently, geriatricians are in short supply. Accordingly, methods to identify persons most vulnerable to adverse outcomes, and to prevent and ameliorate adverse outcomes in such individuals, are urgently needed. Better yet would be to delay the onset of vulnerabilities themselves or to fortify resilience in persons who are becoming vulnerable.

“Frailty” has attracted attention in recent years for its promise for identifying vulnerable older adults, hence its potential to meet the needs just identified (1). However, there remains substantial controversy regarding its nature and ascertainment. At least four efforts to develop consensus have been pursued (1–4). The most recent succeeded in defining frailty as “a medical syndrome with multiple causes and contributors … characterized by diminished strength, endurance, and reduced physiologic function that increases an individual’s vulnerability for developing increased dependency and/or death” but did not resolve disagreements regarding operational definitions (4). Meanwhile, dozens of frailty assessment tools have been proposed (5). Such a proliferation may be deleterious because different frailty instruments identify different “frail” people (6). Aguayo and colleagues found that only 10.4% of the 595 paired comparisons among 35 frailty scores achieved better than “fair” agreement by standard Kappa statistic criteria (7).

We argue that development of frailty assessments should be grounded in “validation”—the process of substantiating that a measurement accurately and precisely measures what it intends. As such, our paper expands the scope of validation beyond verification of a measure’s usefulness for predicting adverse outcomes (8), as advocated in a 2005 commentary (9). Instead, we consider frailty as a specifiable state of health. Accordingly, our perspective emphasizes clear description of the state that is the intended target of measurement. Then, validation primarily must demonstrate that the measurement faithfully reflects the state it assesses; predictive ability is secondary (10,11).

Our article first proposes the validation paradigm using the physical frailty phenotype as an example conceptual framework (12). Other conceptual frameworks could well have been chosen: We pursued this one because our group has discussed it most deeply and rigorously. Then we discuss conceptual issues we consider unresolved and close by raising considerations for the use of frailty measures in clinical practice.

Target of Validation: The Construct to be Measured

Formally, what one aims to measure is termed as a “construct,” i.e., “a postulated attribute … that cannot be measured directly but can be assessed using indirect measures (13).” The more specific the construct definition, the better one can choose measures to assess it and then evaluate whether these behave as they should if one is measuring what one intends.

Example construct - Frailty as a “biologic syndrome of decreased reserve and [reduced] resistance to stressors (12).” This is the construct targeted by the physical frailty phenotype (PFP) as articulated by Fried and colleagues, who went on to hypothesize frailty as “…resulting from cumulative declines across multiple physiologic systems and causing vulnerability to adverse outcomes (12).” Here, we attempt to synthesize and crystallize the PFP construct elements:

Exemplar Conceptual Framework

1. Frailty is a state of health resulting from dysregulation in physiological mechanisms. It is not just a condition secondary to chronic disease(s) or their symptoms.

2. It is distinct from chronological age and does not necessarily precede disability.

3. It can be successfully intervened upon if addressed sufficiently early in its progression.

4. Its hallmarks are diminished fitness in physiologic systems governing stress response, dysregulated energetics, and decreased muscle function. These manifest syndromically.

5. Frailty increases vulnerability to negative health events upon exposure to predisposing stressors, as well as diminished capacity to rebound from such events when not fatal.

Validation Process

The paradigm of Cronbach and Meehl (13) recognizes three overarching aspects of a measure’s validity: content, criterion and construct. Some also identify “face validity” as a useful criterion, which in our context is to say that a frailty measure should identify individuals whom researchers and clinicians can credibly consider as vulnerable.

Definitions of validity aspects and their illustration in the context of the exemplar framework are provided in Table 1. Further elaboration follows:

Table 1.

Validity Definitions and Illustrative Examples

Validation Aspect	Definition	Illustrative Example of a Measure Meeting Validity for the Exemplar Construct
1. Content Validity	Whether the measure’s component criteria faithfully represent the construct targeted by measurement	Criteria chosen to reflect impaired energy availability, diminished muscle tissue quantity/quality and stress response
2. Criterion Validity	Whether the measure associates with variables with which it is designed to associate/not with variables with which it is not designed to associate
a. Concurrent	Whether the measure cross-sectionally associates with other measures of the same construct or related constructs	Persons identified as “frail” by the measure, and persons qualitatively considered “frail” by a knowledgeable geriatrician panel, are highly concordant
b. Predictive	Whether the measure forecasts later outcomes one aims for them to predict	Person designated “frail” by the measure have higher risk for subsequent adverse events and declines in health/functioning than those designated “not frail”
3. Construct Validity	Whether measures behave the way the theory defining what one seeks to characterize mandates
a. Internal	Whether criteria measuring the construct co-occur in patterns one’s theory predicts they should	Criteria are manifested in a “syndromic” fashion.
b. External	Whether measures of one’s construct empirically relate with measures of hypothesized antecedents or consequences (convergent validity), and are unrelated with measures of unrelated constructs (divergent validity)	Persons designated “frail” by the measure more frequently experience adverse outcomes following a stressor than those designated “nonfrail” following an equivalent stressor

Content Validity

To assure this, representation of all central aspects of the construct in the measured indicators, and absence of superfluous ones, both are important. The indicators should be linked causally to the underlying etiology.

Example: Implicit in the Fried (12) PFP operationalization of the conceptual framework given above is that a few sentinel signs and symptoms reflecting systems central to physiological robustness provide sufficient content to recognize frailty. Criteria of low grip strength, exhaustion, and weight loss reflect diminished muscle mass or function and an energy deficit, and gait speed is a measure integrating these aspects with neurological control. Diminished physical activity also measures the integrated aspects just named but could also be seen as an exogenous cause: A measure of physical activity capacity (V0₂ max) or endurance (6-minute walk) might provide better content. For our example (but not necessarily other frailty conceptualizations), age, disease diagnoses, and disability indicators should not be included as indicators if content validity is to be maintained.

Criterion Validity

This pertains to achieving desirable associations of one’s measure with others. Hence, it includes the “usefulness” aspect emphasized by some validation paradigms (9). Predictive validity is a nearly universal expectation and, by far, the aspect most frequently assessed and confirmed for frailty measures.

Examples: The tabled example illustrates the difficulties of concurrent validity in the absence of a gold standard: Discordance may not invalidate either measure under comparison, but only disallow that both are valid. A measure would fail predictive validity if one intended it to identify older adults at risk for specific adverse outcomes, but in fact it failed to do so.

Construct Validity Assesses whether one’s measures co-occur (“internal” validity) or relate to causes or consequences (“external validity”) as the theory defining the target of measurement mandates. Only a few publications have examined internal construct validity for frailty measures (14–17); more have compared risk or severity of adverse outcomes between frail and nonfrail persons following clinical stressors (eg, ref. (18–21)) as a form of external validity.

Example: The form of external construct validity just noted relates to concept element #5. Other aspects of the exemplar theory are less amenable to construct validation. To cohere with elements 1–2, measured “frailty” initiation or worsening must sometimes occur in the absence of worsening disease or disability, or else disease or disability must sometimes occur or worsen without “frailty.” A measure failing this would exhibit poor construct validity. How to define “sometimes” so as to produce compelling evidence is a crucial question. Evidencing assertions 3–4 easily becomes circular, moreover, because one might (a) design interventions to which the measures themselves are susceptible independently of the underlying biology or (b) include such direct measures of stress response physiology in one’s content that association with eroding stress response physiology must be observed.

Unresolved Conceptual Issues

Our exemplar construct highlights the physiological underpinnings of frailty and physical aspects of its manifestation. The role of higher cognitive abilities, mood, and psychosocial milieu in frailty measurement remains under debate. Roughly half of the frailty instruments in most frequent usage include cognitive criteria (5). Our view, supported by findings of distinction in population-based studies (22), is that cognitive, emotional, and psychosocial vulnerability will prove to be important, but unique, constructs most fruitfully measured separately from each other and from physical frailty. If measured separately, one could assess synergistic risks or harms from concurrent vulnerabilities in different domains or, conversely, synergies in different domains of reserve for compensating against diseases, impairments and life stressors. Alternatively, the construct of “cognitive frailty” is a new expansion upon frailty, aiming to detect a premorbid cognitive state caused by physical frailty rather than neurodegenerative pathologies (23). The stated causal relationship underlying this new clinical entity, however, has not been validated.

Analogously, the intersection among frailty, disability, and disease severity poses conceptual and measurement challenges. While conceptually distinct, these constructs are empirically inter-related with a tendency to manifest concurrently; this was illustrated by a Venn diagram in the original PFP paper ((12); recapitulated in Figure 1). We consider it undeniable that diseases may cause signs and symptoms indistinguishable from those defining a given frailty measure—a phenomenon labeled as “secondary frailty” (24). But we hypothesize there also exists “primary frailty”—that arises independently of disease or disablement, during a period of stability in these other conditions, due to eroding effectiveness of coregulation among key physiological systems independent of disease-specific pathologies. This proposition is consistent with Figure 1—but its correctness, and whether this can be judged using data that can feasibly be collected, remain under debate. Matters complicate further when one considers that primary and secondary frailty might temporally coexist. While there are relatively few older adults who are frail but have no disability or comorbidity, this special population merits further study.

Figure 1.

Intersection among frailty, disability, and comorbidity. Venn diagram recapitulated from reference (12): region areas are proportional to numbers in category.

Additional complications arise when one juxtaposes frailty with related geriatric syndromes and processes including physical resilience, sarcopenia, and biological aging, whose definitions and measurement remain under active debate. A recent pair of editorials on physical resilience illustrates the heterogeneity of opinion regarding this juxtaposition: One opined that “one’s likelihood to decline physically following a stressor (frailty) and one’s likelihood to … recover from functional loss during and after stressors (physical resilience) may depend on different mechanisms,” (25) while the other, whose authors contributed to the current paper, hypothesized that “physical … frailty is a disordered state in the dynamical systems that produce resiliency when these same physiological systems are functioning in good order.” (26) Debate regarding biological aging is more nuanced: the frailty index was initially introduced as a proxy measure of aging (27), whereas the frailty phenotype is theorized to emerge when progressive homeostatic dysregulation associated with aging crosses a threshold (28). Whether primary frailty exists, how it relates to other geriatric syndromes and constructs, and whether coregulation among key physiological systems is an important driver of frailty independently of disease or disability are questions that matter because strategies by which to intervene on frailty may fundamentally differ depending on their answers.

We consider the development of dynamic stress-response measures of frailty to be crucial for advancing both measurement quality and theoretical understanding. Generalized vulnerability to stressors is a hallmark that uniquely distinguishes the frailty concept: It may be that only with a multifaceted battery of stress responses can frailty be fully elucidated. The Women’s Health and Aging Study II explored the potential of such measures among women aged 84 years and older: Participants were subjected to an oral glucose tolerance test followed by repeated blood draws by which to assess biomarker response, among several other exposures. Differences in dynamic response were observed for frail versus other women (29). Beyond research value, such testing also stands to have high utility for identifying frail older adults clinically and intervening to their benefit.

To separate the measurement of disease, disability and the various domains of frailty is not to say that these features of health are unrelated entities. Each may impact the course of the other. If resilience and frailty reflect the cooperation of multiple physiologic systems working together (or not) to maintain or regain homeostasis, then the functioning of the individual systems and organs—hence diseases—stands to influence the health of the integrated system. So too does consequent disability, which may feed-back to further impair ability to maintain homeostasis. How to measure frailty in a way that respects these various influences, while maintaining the ability to distinguish etiology and outcomes, is challenging, but worthy.

The conceptual framework exemplified above hypothesizes frailty as a state; others hypothesize frailty as existing along a continuum ranging from highly resilient to extremely vulnerable (27,30). Methods that yield a continuous measurement have been praised for their enhanced ability to detect change, particularly in the context of interventions. The frailty-as-a-state hypothesis arises, in part, from viewing frailty as an outcome of dysregulation in dynamical systems involved in stress-response physiology: To the extent that the system dynamics are nonlinear, one expects the “emergence” of discrete alterations once a threshold of dysregulation is reached. This remains a theory, however. One can easily envision a continuum of resiliency up to a point, followed by the emergence of a disordered frailty state; or a series of increasingly disordered states (31); or a continuum after all. There are important implications for reversibility of declines: Under the state hypothesis, there may well exist a threshold of no return whose detection becomes critical if frailty is to be ameliorated. Ultimately, the appropriate scaling for gold standard intervention response will reflect the nature of the construct itself.

Clinical Considerations

The field is beginning to see clinical usefulness in frailty assessment, notably to risk-stratify older adults facing potential stressors and refer older adults for comprehensive geriatric assessment (32). We foresee at least four prospects for the integration of frailty measurement in clinical practice. Three have to do with geriatric risk screening: (a) to target potentially stressful interventions; (b) to monitor and address changes in frailty status/level over time; and (c) as a public health means to forestall frailty. One also can envision (d) intervention to ameliorate frailty itself.

Type (a) geriatric risk screening for frailty has been envisioned for surgeries (eg, refs. (18,20)), and evidence supports the potential utility of type (b) screening, for example, following hospitalization (19) and stroke (21). Frailty assessment, importantly, provides a means by which to include robust older adults for procedures for which they might otherwise be screened out based solely on their age, as well, potentially, as to “prehabilitate” prefrail older adults before being treated (33). Monitoring during prehabilitation could identify persons who improve and pursue an elective surgery, or alternatively whose frail status deteriorates nonetheless, and for whom palliative care might be most appropriate. Based on these examples and on reported experience with comprehensive geriatric assessment (32), frailty assessment promises usefulness in the course of usual geriatric care. In order for type (c) screening to be useful, interventions to forestall frailty itself (prospect d) and effective prefrailty measures will be needed.

There have been interventions seeking to ameliorate frailty itself, some of which have reported success (34). We are concerned that these primarily have targeted individual criteria, rather than physiological underpinnings, of frailty. At least two trials have implemented multifactorial interventions that specifically targeted the five criteria of the PFP (35,36). Each found associated reduction in frailty: In both cases, however, intervention-driven changes in physical activity likely sufficed to reverse the phenotype’s physical activity criterion. We therefore question whether such interventions ameliorate frailty or mask its symptoms. Work is needed to develop and test interventions addressing the underlying physiology and identify ranges of decline within which frailty amelioration remains possible.

To intervene while improvement remains possible, and to successfully predict risk of earlier-stage adverse outcomes, measures are needed to identify older adults who are preclinically (“pre-”) frail. Prefrailty measures exist—for example, having 1–2 PFP criteria: However, such a characterization may easily arise for reasons other than an early stage frailty process (eg, isolated grip strength deficit due to arthritis). Dynamic measures of vulnerability may hold particular promise for prefrailty measurement. Whether for the PFP or other measures, we are not aware of work to identify thresholds suggestive of impending decline or to optimize prefrailty ranges for prediction of adverse outcomes.

If frailty measurement is to be incorporated in clinical settings, assessment must be sufficiently simple, quick, and inexpensive to be widely practical. Some patients, additionally, will have limitations on the measurements they can perform, for example, grip strength for arthritis patients. In some settings, then, it may be necessary to use abbreviated assessments. Modifications, such as substituting self-report of difficulty walking for assessment of gait speed performance, already proliferate (37). Validation then becomes particularly important to assure that the measure achieves a minimum standard of predictive accuracy for key adverse outcomes and continues to assess what it was originally designed to measure (38). It may also be clinically important to develop two-step approaches that use brief, highly sensitive screening tools administered to all older adults, followed by longer, still-sensitive but more specific evaluations for those who screen “positive” on the brief test.

We—and others (39)—consider clinical and translational research regarding frailty still to be early. Visioning is needed to better define potential clinical uses. Research is then needed to evaluate specific uses for efficacy, effectiveness, and implementation. Guidelines should then be developed and disseminated.

Closing Remarks

There may prove to be usefulness in different conceptual frameworks regarding frailty. We expect, for example, that a physiologically based state of general vulnerability is only one factor determining older adults’ risk for adverse outcomes: Even a precise measure of this state may not forecast risk nearly optimally compared to other risk predictors one might construct. Thus, one imperative is to tailor one’s measure of “frailty” to the intended use. Limited guidance on the choice of measures for given purposes and settings has been published (5); more is needed. Ultimately, the field may need to “more strongly [distinguish] disparate concepts which currently share the single label of ‘frailty.’” (40).

After years of delineating the constituents of frailty measures, we believe the time has come to more deeply delineate the frailty construct itself. Whatever construct one conceptualizes, we consider its specification, and then validation of measures proposed to assess it, as necessary in the development of a worthy frailty assessment. The concept frames the choice of appropriate measures. Predictive validity is important, but assessment of content and construct validity is equally necessary—even if one intends only for frailty to predict risk: In this case, the “content” is a span of relevant predictors that together achieve high predictive accuracy and arguably should be identified through analysis to optimize prediction. Conversely, to emphasize predictive ability above other aspects of validity may produce a measure without a clear conceptual underpinning, hence no rational basis for clinical intervention. In clinical settings, frailty measurement must additionally have value beyond assessments already in place, be easy, and have a treatment plan. Accomplishing all the necessities is a tall order, but, we believe, well worthwhile in order to accelerate the field’s discovery of frailty etiology and efforts to manage threats to older adults’ health and well-being.

Funding

This work was supported by the National Institute on Aging at the National Institutes of Health (P30 AG021334, as well as R01 AG055781 to M.M.-D. and K01 AG050699 to A.L.G.).

Conflicts of Interest

None reported.