Identification of Frailty Using a Claims-Based Frailty Index in the CoreValve Studies: Findings from the EXTEND-FRAILTY Study.

Background In aortic valve disease, the relationship between claims-based frailty indices (CFIs) and validated measures of frailty constructed from in-person assessments is unclear but may be relevant for retrospective ascertainment of frailty status when otherwise unmeasured. Methods and Results We linked adults aged ≥65 years in the US CoreValve Studies (linkage rate, 67%; mean age, 82.7±6.2 years, 43.1% women), to Medicare inpatient claims, 2011 to 2015. The Johns Hopkins CFI, validated on the basis of the Fried index, was generated for each study participant, and the association between CFI tertile and trial outcomes was evaluated as part of the EXTEND-FRAILTY substudy. Among 2357 participants (64.9% frail), higher CFI tertile was associated with greater impairments in nutrition, disability, cognition, and self-rated health. The primary outcome of all-cause mortality at 1 year occurred in 19.3%, 23.1%, and 31.3% of those in tertiles 1 to 3, respectively (tertile 2 versus 1: hazard ratio, 1.22; 95% CI, 0.98-1.51; P=0.07; tertile 3 versus 1: hazard ratio, 1.73; 95% CI, 1.41-2.12; P<0.001). Secondary outcomes (bleeding, major adverse cardiovascular and cerebrovascular events, and hospitalization) were more frequent with increasing CFI tertile and persisted despite adjustment for age, sex, New York Heart Association class, and Society of Thoracic Surgeons risk score. Conclusions In linked Medicare and CoreValve study data, a CFI based on the Fried index consistently identified individuals with worse impairments in frailty, disability, cognitive dysfunction, and nutrition and a higher risk of death, hospitalization, bleeding, and major adverse cardiovascular and cerebrovascular events, independent of age and risk category. While not a surrogate for validated metrics of frailty using in-person assessments, use of this CFI to ascertain frailty status among patients with aortic valve disease may be valid and prognostically relevant information when otherwise not measured.

aortic valve replacement

continued access study

claims‐based frailty index

CoreValve high risk trial

major adverse cardiovascular and cerebrovascular event

surgical aortic valve replacement

surgical or transcatheter aortic‐valve replacement in intermediate‐risk patients trial

transcatheter aortic valve replacement

What Is New?

In this study using US CoreValve Pivotal Studies data linked to Medicare claims, a claims‐based frailty index based on the Fried index, identified individuals with worse impairments in frailty, disability, cognitive dysfunction, and nutrition and a higher risk of death, hospitalization, bleeding, and major adverse cardiovascular and cerebrovascular events, independent of age and risk category.

What Are the Clinical Implications?

While well‐validated quantitative metrics based on in‐person assessments represent the gold‐standard for frailty assessment, this claims‐based frailty index represents an alternative for retrospective ascertainment of frailty status when otherwise unmeasured.

Frailty, defined as “a state of increased vulnerability and reduced ability to maintain homeostasis after a stressful event resulting from impairment in multiple physiologic systems,” is an important and often unmeasured risk factor for adverse outcomes among individuals undergoing aortic valve replacement (AVR) for severe aortic valve disease. , While multiple scales exist to measure frailty, they may be broadly categorized into 2 main types: a deficit‐based frailty index (Rockwood index) that conceptualizes frailty as an accumulation of deficits over time and a phenotype‐based frailty index (Fried index) that conceptualizes frailty as a biologic syndrome. This latter construct conceptualizes frailty as a biologic phenotype consisting of impairments across 5 domains: shrinking (ie, weight loss), exhaustion, weakness, slowness, and low physical activity. By this latter definition, frailty is present in up to 63% of those undergoing transcatheter aortic valve replacement (TAVR) and is associated with a nearly 4‐fold increased risk of death 1 year after TAVR , , as well as functional decline at 6 months and may be incrementally predictive of adverse risk beyond age and comorbidities alone. ,

At the same time, frailty is often unmeasured in clinical trials and is not captured by traditional risk scores used to risk stratify individuals for AVR. Moreover, in contrast to well‐defined and validated frailty metrics using in‐person measurements, physicians’ subjective assessments of frailty may not significantly predict risk in TAVR. As such, retrospective ascertainment of frailty status, using claims algorithms developed on the basis of such validated metrics of frailty, has been advocated to improve risk prediction, assessment of hospital care quality, and evaluation of study generalizability when frailty assessment using these validated techniques has not been performed. While we have previously demonstrated that a claims‐based frailty index (CFI) identifies individuals undergoing TAVR at higher risk of adverse outcomes than comorbidities alone using nationwide claims data, whether it identifies individuals with a greater burden of frailty‐related health deficits and similarly identifies an increased risk of adverse outcomes in a clinical trial population remains uncertain. Although not intended to replace previously validated techniques to assess frailty using in‐person measurement, it is possible that CFIs could have a role in retrospectively ascertaining one’s frailty status in data sets where this key risk marker is not otherwise measured (Figure 1).

Figure 1

Schematic depicting relative advantages and disadvantages of methods to ascertain frailty in patients with severe aortic stenosis being evaluated for aortic valve replacement.

Schematic depicting the role of different data sources, namely, administrative claims and validated in‐person metrics for evaluation of the frailty phenotype. Claims‐based frailty indices are easy and inexpensive to measure and can be applied retrospectively but few have been validated against gold‐standard definitions for frailty. Conversely, validated metrics of frailty using in‐person measures may be useful prospectively to assess an individual’s frailty status but may be limited by time, expense, and availability and are challenging to apply retrospectively. AVR indicates aortic valve replacement and VA, Veteran's Affairs.

As such, in the US CoreValve studies, we evaluated the concordance between health deficits related to frailty, measured using rigorous trial assessments, and a single CFI measure, validated against the Fried definition of frailty, , to assess whether claims can validly identify individuals with a greater number of frailty‐related deficits and predict the occurrence of adverse outcomes in this setting.

Methods

Data Availability

As per prior data use agreements with Centers for Medicare and Medicaid Services and Medtronic, the data supporting the current study are not publicly available for review.

Study Population

As part of the National Heart, Lung, and Blood Institute–sponsored (1R01HL136708) EXTEND (Extending Trial‐Based Evaluations of Medical Therapies Using Novel Sources of Data) study, we previously linked Medicare inpatient claims to the US CoreValve Pivotal Trials data set. We subsequently examined the relationship of a single CFI and baseline covariates and outcomes in the US CoreValve Pivotal Trial data set as part of the EXTEND‐FRAILTY substudy. This data set consists of a set of trials comparing TAVR using the self‐expanding Medtronic CoreValve bioprosthesis with surgical AVR (SAVR). Details on this linkage have been published previously. Data from patients included in the US CoreValve HiR (High Risk trial), SURTAVI (Surgical or Transcatheter Aortic‐Valve Replacement in Intermediate‐Risk Patients) trial, and single‐arm CAS (Continued Access Study) who could be successfully linked to US Centers for Medicare and Medicaid Services Medicare Provider Analysis and Review data set with procedure dates February 2, 2011, to September 30, 2015, were included. The CoreValve HiR randomized individuals at high surgical risk with severe aortic stenosis to undergo TAVR with the Medtronic CoreValve bioprosthesis versus SAVR. The SURTAVI trial randomized individuals at intermediate surgical risk with severe aortic stenosis to undergo TAVR with the Medtronic CoreValve bioprosthesis versus SAVR. The CAS represents a single‐arm cohort study of both extreme risk and high risk US TAVR recipients included in the US CoreValve trials intended for follow‐up of outcomes and adverse events. Only high‐risk CAS patients were included in this analysis. These particular studies were chosen on the basis of the high prevalence of frail individuals and overlap with the use of International Classification of Diseases, Ninth Revision, Clinical Modification (ICD‐9‐CM) claims before October 1, 2015.

The Medicare Provider Analysis and Review database used consists of a 100% sample of inpatient discharge claims (Part A) for Medicare Fee‐for‐Service beneficiaries and has been used extensively for health services research. As direct patient identifiers were not available in the US CoreValve Pivotal Trials data set, we previously linked the Medicare Provider Analysis and Review and trial data sets using a deterministic matching strategy based on age, date of birth, sex, procedure dates, admission and discharge dates, and hospital identification. Patients aged <65 years or those undergoing AVR at a Veterans Affairs or European hospital were excluded.

Of those initially included in the US CoreValve HiR (N=750), SURTAVI (N=1660), or the high‐risk group of the CAS (N=1108), 2520 (71.6%) were able to be successfully linked including 600 (80%) individuals in the US CoreValve HiR, 1005 (60.5%) individuals in SURTAVI, and 915 (82.6%) individuals in the CAS (Figure 2). Of those in the US CoreValve HiR, 15 individuals (2.0%) were excluded because they were aged <65 years or had an index procedure at a Veterans Affairs or European hospital. Of those in SURTAVI, 355 individuals (21.4%) were excluded because they were aged <65 years or had an index procedure at a Veterans Affairs or European hospital. Of those in the High Risk CAS, 34 individuals (3.1%) were excluded because they were aged <65 or had an index procedure at a Veterans Affairs or European hospital. As Medicare Advantage health maintenance organizations represented 13% to 30% of Medicare enrollees during the time period, the majority of nonmatched individuals were likely enrolled in Medicare Advantage, for which claims data were not available. As the SURTAVI trial enrolled a younger and more international cohort, rates of linkage were lower. Subsequently, 163 individuals (9.8%) in SURTAVI were excluded because of procedure dates after October 1, 2015, the date when International Classification of Diseases, Tenth Revision was introduced in the United States. The study was approved by the Institutional Review Board at Beth Israel Deaconess Medical Center with a waiver of informed consent.

Figure 2

Study schematic displaying results of study linkage.

Linkage strategy used in the current study. Of 750 individuals in the US CoreValve High Risk Trial, 15 were excluded because they were aged <65 years or undergoing aortic valve replacement at a Veterans Affairs or European hospital. Of the 735 remaining, 135 were unable to linked to Medicare data. Of the 1108 individuals in the High Risk Continued Access Study (CAS), 34 were excluded because they were aged <65 years old or undergoing aortic valve replacement at a Veterans Affairs or European hospital. Of the 1074 remaining, 159 were unable to linked to Medicare data. Of the 1660 individuals in the SURTAVI trial, 355 were excluded because they were aged <65 years old or undergoing aortic valve replacement at a Veterans Affairs or European hospital. Of the 1305 remaining, 200 were unable to be linked to Medicare data. Subsequently, 163 were excluded because of procedure dates after October 1, 2015. CMS indicates Centers for Medicare and Medicaid Services; VA, Veterans Affairs; and SURTAVI, Surgical or Transcatheter Aortic‐Valve Replacement in Intermediate‐Risk Patients Trial.

Covariates

Clinical and comorbidity data were determined for individuals undergoing AVR using baseline variables as defined and recorded in the trial data sets, broadly categorized into demographics, comorbidities, risk scores, and procedural variables (Table S1).

Frailty‐Related Health Deficits

Well‐understood health deficits, related to frailty or disability status and measured during trial baseline assessment, were broadly categorized into domains of functional status assessment, severity of lung disease, nutrition, weakness/slowness, cognitive dysfunction, and disability (Table S2).

Functional status was defined using baseline symptom questionnaires. Individuals in the CoreValve trials underwent baseline assessment of the Kansas City Cardiomyopathy Questionnaire, the 5‐level EQ‐5D questionnaire, the 12‐item Short Form Survey, and the 36‐item Short Form Survey (SF‐36; SURTAVI only). , , , The Kansas City Cardiomyopathy Questionnaire is a 23‐item questionnaire, graded from 0 to 100 with higher scores reflecting better health status, with 2 summary scores for overall functioning (overall summary score) and symptom‐specific functioning (clinical summary score). The EQ‐5D is a 5‐item questionnaire assessing generic health status on 5 dimensions, those being mobility, self‐care, ability to perform usual activities, pain/discomfort, and anxiety/depression. The 12‐item Short Form Survey represents a 12‐item questionnaire, graded from 0 to 100 with higher scores reflecting better health status, with 2 composite scores for physical and mental functioning. Similarly, the SF‐36 represents a 36‐item questionnaire, graded from 0 to 100 with higher scores reflecting better health status, with 2 composite scores for physical and mental functioning.

Outcomes

The primary outcome for this analysis was 1‐year all‐cause mortality. All outcomes were defined as per the original trial protocols (Table S3). Secondary outcomes included acute kidney injury, bleeding, all stroke or transient ischemic attack, aortic reintervention, hospitalization, myocardial infarction, and major adverse cardiovascular and cerebrovascular event (MACCE), defined as a composite of all‐cause death, myocardial infarction, stroke, or aortic reintervention.

Ascertainment of Frailty Status

Frailty status was ascertained using claims via the Johns Hopkins CFI published by Segal et al. This CFI was developed using International Classification of Diseases, Ninth Revision, Clinical Modification Medicare claims linked to data from the Cardiovascular Health Study and externally validated in the National Health and Aging Trend Study. , This CFI was developed and validated using claims chosen on the basis of their correlation with the Fried index as the reference standard, predicts outcomes similarly to the Fried index, and has been used previously in studies of frailty. While other CFIs have been developed previously, all other scales are based on the accumulation of deficits definition of frailty. A CFI based on the Fried conception was chosen so as to be more closely aligned with the conceptualization of frailty as a physical syndrome that can improve or worsen over time. Claims (Table S4) for hospitalizations in the 6 months preceding the baseline visit were used to construct the frailty index. In addition to treating the index as a continuous variable, a previously proposed cutoff of >0.20 was used to define an individual’s frailty status.

Statistical Analysis

The distribution of the CFI in the sample was described using means and SDs and displayed graphically using a histogram. The number and proportion of the sample with frailty using a dichotomous cutoff of >0.20 was determined. We first divided the CFI into tertiles and compared baseline trial variables (described using means±SDs for continuous variables and frequencies and percentages for categorical variables) across tertiles using analysis of variance for continuous variables and chi‐squared tests for categorical variables. Overall rather than study‐specific tertiles were used throughout the analysis.

Subsequently, to evaluate the construct validity of the CFI, we compared frailty‐related variables, collected during baseline trial assessments, across CFI tertiles using analysis of variance for continuous variables and chi‐squared tests for categorical variables.

Next, rates of adverse outcomes at 1 year by CFI tertile were calculated using Kaplan‐Meier estimates in both the combined data set and stratified by study group (eg, HiR, CAS, or SURTAVI). Kaplan‐Meier curves and the log‐rank test were used to compare the primary outcome of death across CFI tertiles. Cox proportional hazards models were used to model time to all nondeath outcomes by CFI tertile, accounting for the competing risk of death using Fine‐Gray competing risk estimates. Individuals without events were censored at 1 year after their procedure. Analyses were subsequently adjusted for age, sex, New York Heart Association class, and the Society of Thoracic Surgeons risk score to assess if results changed. All analyses were performed using SAS version 9.4 or JMP version 15.0 (SAS Institute, Cary, NC) using a 2‐tailed P value <0.05 to define significance.

Results

Overall Results

A total of 2357 (67.0%) of individuals in the CoreValve HiR, SURTAVI, and High Risk group of the CAS were able to be successfully linked to Medicare data and were included in the analysis (Figure 2). Individuals whose records could not be linked were overall similar to those who could be linked (Table S5), although the age, Society of Thoracic Surgeons risk score, and the proportion with heart failure were higher, and the Logistic EuroSCORE was lower in the linked group.

Of those included, the mean CFI score was 0.27±0.13 and the median (interquartile range) CFI was 0.25 (0.17–0.35) (Figure S1). The mean age was 82.7±6.2 years and 1015 (43.1%) were women. Overall, 1656 (70.3%) received TAVR and 701 (29.7%) received SAVR with femoral access in 1445 of those receiving TAVR (87.3%). Using a threshold cutoff of 0.20 to define frailty, 1529 (64.9%) of the sample was considered frail. A total of 787 (33.3%) were in tertile 1 (CFI ≤0.20), 788 (33.4%) were in tertile 2 (CFI, 0.21–0.31) and 782 (33.2%) were in tertile 3 (CFI ≥0.32).

Comparison of Baseline Trial Characteristics Across CFI Tertiles

Use of TAVR was higher in the higher CFI tertiles (Table 1; tertile 3 versus 1: 74.7% versus 63.2%; P<0.001 across tertiles). Those in higher CFI tertiles were older (tertile 3 versus 1: mean age, 87.2±4.0 versus 77.4±5.6; P<0.001) and more frequently women (tertile 3 versus 1: 54.0% versus 32.8%; P<0.001). The number of comorbidities was similar across frailty tertiles (tertile 3 versus 1: mean Charlson comorbidity index, 5.0±2.3 versus 5.1±2.1; P=0.77) with nearly all individuals having a history of hypertension (tertile 3 versus 1: 92.7% versus 95.7%; P=0.03). Those in higher frailty tertiles more frequently had a history of heart failure (tertile 3 versus 1: 92.2% vs. 53.5%, P<0.001) but less frequently had insulin‐dependent diabetes mellitus (tertile 3 versus 1: 8.7% versus 17.5%; P<0.001). Despite similar rates of prior percutaneous coronary intervention across tertiles (P=0.93), rates of coronary artery bypass grafting were lower among those in higher frailty tertiles (tertile 3 versus 1: 17.1% versus 36.7%; P<0.001). The Society of Thoracic Surgeons risk score was higher in those with a higher CFI (tertile 3 versus 1: 8.0±3.4 versus 5.4±2.5; P<0.001).

Table 1

Baseline Demographic, Procedural, Risk Score, and Comorbidity Characteristics of CoreValve Study Participants Across CFI Tertiles

Characteristic	Observations, N	Tertile 1 (n=787)	Tertile 2 (n=788)	Tertile 3 (n=782)	P value
Demographics
Age, y	2357	77.4±5.6	83.7±4.4	87.2±4.0	<0.001
Female sex, n (%)	2357	258 (32.8)	335 (42.5)	422 (54.0)	<0.001
Risk scores
Society of Thoracic Surgeons risk score	2357	5.4±2.5	6.5±2.8	8.0±3.4	<0.001
Logistic EuroSCORE	2355	14.2±11.2	17.7±12.4	19.3±12.3	<0.001
Charlson comorbidity index	1511	5.1±2.1	5.1±2.2	5.0±2.3	0.77
Comorbidities
Diabetes mellitus, n (%)
Total	1515	172 (21.9)	192 (24.4)	221 (28.3)	<0.001
Controlled by insulin	2357	138 (17.5)	78 (9.9)	68 (8.7)	<0.001
History of hypertension	2357	753 (95.7)	749 (95.1)	725 (92.7)	0.03
Peripheral vascular disease	2351	339 (43.1)	314 (39.8)	311 (39.8)	0.33
Prior stroke	2356	80 (10.2)	79 (10.0)	90 (11.5)	0.57
Connective tissue diseases	1513	<11	<11	14 (2.2)	0.53
Immunosuppressive therapy	2355	97 (12.3)	70 (8.9)	75 (9.6)	0.06
Prior transient ischemic attack	2356	71 (9.0)	89 (11.3)	78 (10.0)	0.32
Cirrhosis	2355	14 (1.8)	<11	<11	0.02
Cardiac risk factors, n (%)
Coronary artery disease	2357	598 (76.0)	588 (74.6)	527 (67.4)	<0.001
Prior CABG	2357	289 (36.7)	231 (29.3)	134 (17.1)	<0.001
Prior PCI	2357	255 (32.4)	260 (33.0)	251 (32.1)	0.93
Pacemaker or implantable defibrillator	2357	116 (14.7)	139 (17.6)	151 (19.3)	0.051
Prior myocardial infarction	2357	193 (24.5)	174 (22.1)	158 (20.2)	0.12
Congestive heart failure	2357	421 (53.5)	601 (76.3)	721 (92.2)	<0.001
Atrial flutter or fibrillation	2357	270 (34.4)	300 (38.1)	322 (41.3)	0.02
Procedural variables
Treatment assignment, n (%)
TAVR	2357	497 (63.2)	575 (73.0)	584 (74.7)	<0.001
SAVR		290 (36.9)	213 (27.0)	198 (25.3)
Presence of a calcified aorta, n (%)
No calcification	2353	128 (17.5)	102 (12.9)	102 (13.0)	0.15
Mild calcification		392 (49.8)	405 (51.4)	371 (47.4)
Moderate calcification		207 (26.3)	201 (25.5)	227 (29.0)
Severe calcification		60 (7.6)	77 (9.8)	79 (10.1)
Chest wall deformity, n (%)	2357	<11	<11	<11	0.95
Hostile mediastinum, n (%)	1511	22 (6.2)	<11	<11	<0.001
Arterial access site, n (%)
Femoral	1655	433 (55.0)	500 (63.5)	512 (65.5)	0.64
Subclavian		21 (2.7)	17 (2.2)	19 (2.4)
Aortic		39 (5.0)	50 (6.3)	46 (5.9)
Other		<11	<11	<11
Number of valves implanted	1656	1.1±0.2	1.1±0.2	1.0±0.2	0.84

Values are listed as means±standard deviations unless otherwise specified. Individuals in tertile 1 had a CFI ≤0.20, those in tertile 2 had a CFI of 0.21 to 0.31, and those in tertile 3 had a CFI ≥0.32. CABG indicates coronary artery bypass grafting; CFI, claims‐based frailty index; PCI, percutaneous coronary intervention; SAVR, surgical aortic valve replacement; and TAVR, transcatheter aortic valve replacement.

Comparison of Frailty‐Related Health Deficits Across CFI Tertiles

Those in the higher CFI tertiles had a higher proportion of health deficits related to frailty and disability, across all domains except for severity of lung disease (Table 2). Specifically, those in the higher CFI tertiles had a greater proportion of individuals with impairments in nutrition including low serum albumin, unplanned weight loss (tertile 3 versus 1:, 9.7% versus 3.2%; P=0.04) and a lower body mass index (tertile 3 versus 1: 27.3±5.4 versus 30.2±6.4; P<0.01), despite no differences in anemia requiring transfusion (P=0.58). Those in higher CFI tertiles had greater impairments in weakness/slowness, with a greater proportion having falls within 6 months, 5‐meter walk time <0.5 m/s, and a grip strength below threshold. Additionally, cognitive function as assessed by the Mini‐Mental Status Exam was worse in higher CFI tertiles (tertile 3 versus 1: 26.3±2.8 versus 27.3±2.5; P<0.001). Those in higher CFI tertiles had increased disability with dependence in living, bathing, dressing, toileting, and transferring (all P<0.05), though no differences in urinary incontinence (P=0.11). Additionally, self‐reported quality of life was worse in those in the high CFI tertiles with worsened Kansas City Cardiomyopathy Questionnaire overall and summary scores and EQ‐5D (P<0.05 for all), despite no significant differences across tertiles in the 12‐item Short Form Survey and SF‐36 (P>0.05 for both).

Table 2

Frailty‐Related Characteristics of CoreValve Study Participants Across CFI Tertiles

Characteristic	Observations, N	Tertile 1 (N=787)	Tertile 2 (N=788)	Tertile 3 (N=782)	P value
Nutrition
Body mass index, kg/m2	2357	30.2±6.4	28.2±5.4	27.3±5.4	<0.001
Anemia requiring transfusion, n (%)	1447	55 (16.1)	75 (15.5)	110 (17.7)	0.58
Albumin <3.3 g/dL, n (%)	1492	48 (6.1)	54 (6.9)	109 (13.9)	0.009
Unplanned weight loss, n (%)	1515	25 (3.2)	47 (6.0)	76 (9.7)	0.04
Weakness/Slowness
Falls in the past 6 months, n (%)	2357	93 (11.8)	136 (17.3)	186 (23.8)	<0.001
5‐meter gait speed, s	2163	7.5±3.6	8.6±4.8	10.0±7.7	<0.001
5‐meter gait speed <0.5 m/s, n (%)	2163	106 (13.5)	182 (23.1)	245 (31.3)	<0.001
Grip strength below threshold, n (%)	2320	528 (67.1)	497 (63.1)	472 (60.4)	0.008
Cognitive dysfunction
Mini‐Mental Status Exam score	2287	27.3±2.5	27.1±2.4	26.3±2.8	<0.001
Disability
Does not live independently, n (%)	2357	23 (2.9)	33 (4.2)	71 (9.1)	<0.001
Does not bathe independently, n (%)	2357	23 (2.9)	28 (3.6)	65 (8.3)	<0.001
Does not dress independently, n (%)	2357	18 (2.3)	20 (2.5)	41 (5.2)	0.002
Does not toilet independently, n (%)	2357	<11	<11	24 (3.1)	<0.001
Does not transfer independently, no (%)	2357	13 (1.7)	17 (2.2)	40 (5.1)	<0.001
Does not feed independently, n (%)	2357	<11	<11	<11	0.79
Urinary incontinence, n (%)	2357	14 (1.8)	21 (2.7)	27 (3.5)	0.11
Functional status assessment
New York Heart Association class, n (%)
Class II	2357	214 (27.2)	188 (23.9)	160 (20.5)	0.03
Class III		496 (63.0)	526 (66.8)	540 (69.1)
Class IV		69 (8.8)	69 (8.8)	78 (10.0)
KCCQ overall summary score	2177	66.0±27.0	58.1±25.5	52.1±24.1	<0.001
KCCQ clinical summary score	2177	67.4±25.2	60.4±24.1	55.2±23.5	<0.001
EQ‐5D index score	2162	0.79±0.18	0.76±0.18	0.74±0.19	<0.001
SF‐12 physical component summary score	1445	30.6±8.6	31.5±9.2	31.0±8.4	0.41
SF‐12 mental component summary score	1445	47.5±11.5	48.1±12.4	48.7±11.6	0.29
SF‐36 physical component summary score	814	36.9±9.9	35.7±9.5	35.2±9.2	0.10
SF‐36 mental component summary score	814	50.6±11.5	49.4±11.7	49.2±11.6	0.29
Severity of lung disease
Society of Thoracic Surgeons chronic lung severity, n (%)
None	2356	401 (51.0)	479 (60.8)	486 (62.1)	<0.001
Mild		176 (22.4)	171 (21.7)	169 (21.6)
Moderate		114 (14.5)	83 (10.5)	71 (9.1)
Severe		96 (12.2)	54 (6.9)	56 (7.2)
Requirement for home oxygen, n (%)	2356	93 (11.8)	59 (7.5)	62 (7.9)	0.006
Forced expiratory volume in 1 s (mL)	1058	1850.6±806.0	1741.5±823.0	1580.1±587.0	<0.001
Diffusion capacity for carbon monoxide (%)	341	60.0±21.5	68.6±22.7	64.5±21.9	0.02

Values are listed as means±standard deviations unless otherwise specified. Individuals in tertile 1 had a CFI ≤0.20, those in tertile 2 had a CFI of 0.21 to 0.31, and those in tertile 3 had a CFI ≥0.32. CFI indicates claims‐based frailty index; KCCQ, Kansas City Cardiomyopathy Questionnaire; SF‐12, 12‐item short form questionnaire; and SF‐36, 36‐item short form questionnaire.

Despite a lower forced expiratory volume in 1 second (P<0.001) in those with higher CFI tertiles, the proportion with severe lung disease (tertile 3 versus 1: 7.2% versus 12.2%; P<0.001) and requirement for home oxygen (tertile 3 versus 1: 7.9% versus 11.8%; P=0.006) was lower, and the diffusion capacity for carbon monoxide was higher (P=0.02).

Primary Outcome

At 1 year in the overall sample, 19.3% of those in tertile 1, 23.1% in tertile 2, and 31.3% of those in tertile 3 of the CFI had died (tertile 2 versus 1: hazard ratio [HR], 1.22; 95% CI, 0.98–1.51; P=0.07; tertile 3 versus 1: HR, 1.73; 95% CI, 1.41–2.12; P<0.001) (Figure 3). In the HiR trial, 33.1% of those in tertile 1, 31.1% in tertile 2, and 47.7% in tertile 3 died at 1 year (tertile 2 versus 1: HR, 0.91; 95% CI, 0.65–1.27; P=0.91; tertile 3 versus 1: HR, 1.57; 95% CI, 1.16–2.13; P=0.004) (Table S6). In the CAS study, 22.2% of those in tertile 1, 25.7% of those in tertile 2, and 26.2% of those in tertile 3 died at 1 year (tertile 2 versus 1: HR 1.16, 95% CI 0.81–1.67, P=0.41; tertile 3 versus 1: HR, 1.18; 95% CI, 0.84–1.66; P=0.35) (Table S7). In the SURTAVI study, 13.1% of those in tertile 1, 14.4% of those in tertile 2, and 17.0% of those in tertile 3 died at 1 year (tertile 2 versus 1: HR, 1.11; 95% CI, 0.75–1.65; P=0.59; tertile 3 versus 1: HR, 1.35; 95% CI, 0.84–2.16; P=0.22) (Table S8).

Figure 3

Kaplan‐Meier curve demonstrating time to all‐cause mortality by claims‐based frailty index tertile.

Kaplan‐Meier survival curve for all‐cause mortality in the HiR and CAS studies according to time since aortic valve replacement. The red line indicates those in the CFI tertile 1 (CFI ≤ 0.20), the green line indicates those in CFI tertile 2 (CFI, 0.21–0.31), and the blue line indicates those in CFI tertile 3 (CFI ≥ 0.32). Numbers in the risk set at each time point are indicated below. CAS indicates US CoreValve Continued Access Study; CFI, claims‐based frailty index and HiR, US CoreValve High Risk Study. Log‐rank P value for the overall comparison <0.001.

Secondary Outcomes

At 1 year in the overall sample (Table 3), 27.8% of those in tertile 1, 32.9% in tertile 2, and 36.7% of tertile 3 had experienced a MACCE event (tertile 2 versus 1: HR, 1.22; 95% CI, 1.03−1.45; P=0.03; tertile 3 versus 1: HR, 1.39; 95% CI, 1.18–1.65; P<0.001). Compared with those in tertile 1, those in tertile 2 (34.1% versus 27.3%; HR, 1.30; 95% CI, 1.10–1.55; P=0.002) and tertile 3 (42.2% versus 27.3%; HR, 1.68; 95% CI, 1.43–1.98; P<0.001) more frequently had bleeding. Compared with tertile 1, tertile 2 more frequently had hospitalizations (19.9% versus 15.8%; HR, 1.30; 95% CI, 1.03–1.63; P=0.03) but not tertile 3 (19.2% versus 15.8%; HR, 1.26; 95% CI, 1.00–1.59; P=0.052). Rates of acute kidney injury, stroke or transient ischemic attack, myocardial infarction, aortic reintervention, or other events were similar (P>0.05 for all). Considering CFI as a continuous predictor, results were similar (Table S9). These results were overall consistent across studies (Tables S6–S8). Among the subgroups who received TAVR (N=1656; Table S10) or SAVR (N=701), results were substantially unchanged (Table S11).

Table 3

Comparison of Outcomes by CFI Tertile in the Overall Cohort (N=2357)

Outcomes	Tertile 1 (N=787)	Tertile 2 (N=788)	Tertile 3 (N=782)	HR (95% CI) for Tertile 2 vs. 1	P value*	HR (95% CI) for Tertile 3 vs. 1	P value†
Death (N=579), n (%)	152 (19.3)	182 (23.1)	245 (31.3)	1.22 (0.98–1.51)	0.07	1.73 (1.41–2.12)	<0.001
MACCE (N=765), n (%)	219 (27.8)	259 (32.9)	287 (36.7)	1.22 (1.03–1.45)	0.03	1.39 (1.18–1.65)	<0.001
Acute kidney injury (N=277), n (%)	102 (13.0)	87 (11.0)	88 (11.3)	0.84 (0.63–1.11)	0.22	0.86 (0.65–1.14)	0.28
Bleeding (N=814), n (%)	215 (27.3)	269 (34.1)	330 (42.2)	1.30 (1.10–1.55)	0.002	1.68 (1.43–1.98)	<0.001
Stroke or transient ischemic attack (N=317), n (%)	100 (12.7)	108 (13.7)	109 (13.9)	1.09 (0.83–1.42)	0.54	1.11 (0.85–1.45)	0.45
Myocardial infarction (N=56), n (%)	18 (2.3)	22 (2.8)	16 (2.1)	1.23 (0.66–2.28)	0.52	0.90 (0.46–1.75)	0.75
Aortic reintervention (N=33), n (%)	15 (1.9)	12 (1.5)	<11	0.80 (0.37–1.70)	0.56	0.40 (0.16–1.04)	0.06
Hospitalization (N=431), n (%)	124 (15.8)	157 (19.9)	150 (19.2)	1.30 (1.03–1.63)	0.03	1.26 (1.00–1.59)	0.052
Other (N=74), n (%)	26 (3.3)	27 (3.4)	21 (2.7)	1.04 (0.61–1.77)	0.89	0.81 (0.46–1.44)	0.46

Represents the P value for the comparison of CFI tertile 2 vs. tertile 1.

Represents the P value for the comparison of CFI tertile 3 vs. tertile 1. Listed is the number and percentage of outcomes in each category occurring within 1 year from procedure by CFI tertile in the overall cohort. Percentages are determined using Kaplan‐Meier estimates. Additionally, the hazard ratios and 95% CIs for the comparison of tertile 2 vs. 1 and tertile 3 vs. 1 are listed with the log‐rank P values for these comparisons. For nondeath outcomes, estimates are adjusted for the competing risk of death using Fine‐Gray subdistribution hazard models. Cell numbers <11 are suppressed from publication per Centers for Medicare and Medicaid Services policy. CFI indicates claims‐based frailty index; HR, hazard ratio; and MACCE, major adverse cardiovascular and cerebrovascular events.

Multivariable Analysis

After adjustment for age, sex, New York Heart Association class, and Society of Thoracic Surgeons score, those in CFI tertile 3 but not 2 continued to have an increased risk of death (tertile 3 versus 1; adjusted HR, 1.48; 95% CI, 1.12–1.96; P=0.006; tertile 2 versus 1: adjusted HR, 1.13; 95% CI, 0.88–1.45; P=0.33) (Table S12). Analyses of secondary end points also had similar results after adjustment for trial baseline variables (Table S12).

Discussion

In this study of CoreValve HiR, SURTAVI, and CAS participants linked to Medicare claims, a CFI anchored to the Fried index demonstrated good construct validity in identifying individuals with health deficits related to frailty. Frailty was present in 64.9% of individuals. Nevertheless, higher CFI scores in this cohort identified individuals at higher risk of death, MACCE, hospitalization, and bleeding, despite adjustment for age, sex, and risk category. As a whole, these results suggest that, while well‐validated techniques to assess frailty, using in‐person measurements, remain the gold standard for assessing one’s frailty status, use of the Johns Hopkins CFI has utility for retrospective ascertainment of one’s frailty status in studies of aortic valve disease in circumstances where this important risk factor is unmeasured, and may capture relevant prognostic information over and above traditional risk scores.

Multiple CFIs exist for the evaluation of an individual’s frailty status, , , , but only one, the Johns Hopkins CFI used in this study, is anchored to the Fried index. This construct conceptualizes frailty as a biologic phenotype, associated but not synonymous with aging, consisting of impairments in 5 domains: shrinking (ie, weight loss), exhaustion, weakness, slowness, and low physical activity. By contrast, other CFIs derive from the Rockwood cumulative‐deficit conception of frailty that treats increasing frailty as an accumulation of deficits across multiple health domains. The Johns Hopkins CFI was chosen for this study to be more closely aligned with the conceptualization of physical frailty as a syndrome that can improve or worsen over time. Moreover, though 1 prior CFI has been derived under the International Classification of Diseases, Tenth Revision framework based on clusters of resource usage, this CFI has only a modest correlation with tradition metrics of frailty including the Fried and Rockwood indices. As such, it is unclear that this represents a true metric of frailty in the traditional conception, and further development of International Classification of Diseases, Tenth Revision–based frailty scales is needed.

The CFI used in the current study, which has been validated against the Fried definition of frailty as measured in the Cardiovascular Health Study, identified individuals at higher risk of death, MACCE, hospitalization, and bleeding in the CoreValve studies, despite adjustment for age, sex, New York Heart Association class, and Society of Thoracic Surgeons risk score, suggesting it may improve upon risk stratification using conventional risk metrics. Importantly, this increased risk of adverse outcomes was primarily (though not exclusively) observed among those in the highest CFI tertile, suggesting this increase in adverse risk may be most prominent in the frailest individuals.

In the current study, the Johns Hopkins CFI displayed good construct validity for identifying individuals with a higher burden of frailty‐related health deficits in addition to worse outcomes. Importantly, the Johns Hopkins CFI identified individuals with greater impairments in disability (ie, impairment in activities of daily living/instrumental activities of daily living), cachexia/malnutrition, gait speed, and self‐reported quality of life. Though the Fried index, which associates with this CFI, has been criticized for its lack of accounting for cognitive frailty, the Johns Hopkins CFI nevertheless identified individuals with more cognitive dysfunction, suggesting that cognitive frailty may have significant overlap with other domains of frailty. Interestingly, the CFI was inversely related with severity of lung disease and not related to commonly used markers of frailty such as anemia or urinary incontinence. This finding may suggest that these markers represent poor surrogates for frailty in the older aged aortic stenosis population or that the Johns Hopkins CFI does not readily identify them. Nevertheless, despite this, the Johns Hopkins CFI was associated with significant prognostic utility. As frailty likely represents a closer approximation of biologic age than chronologic age, this may in part explain the strength of using frailty for risk prediction.

It is important to note that, despite its strengths, use of this CFI is not intended to replace well‐validated metrics for assessment of frailty status using in‐person measurements. While health deficits related to one’s frailty status were measured in the CoreValve trials, there were insufficient baseline data collected to generate a Fried index, and thus there was not a clear in‐person reference standard to compare the CFI against. Furthermore, if CFIs are to be used in the clinical care of individual patients versus making broader population‐based policy decisions, it would be necessary and important to evaluate the degree of misclassification of one’s frailty status observed due to ascertainment of frailty via claims. As the Johns Hopkins CFI is not a perfect surrogate of the Fried index (area under the curve=0.75), well‐validated frailty metrics based on in‐person measurements should be considered the gold standard for frailty assessment at the current time. Though over 20 different frailty scales have been developed, regardless of the scale evaluated, frailty as a construct has been consistently associated with a 1.4‐ to 4.5‐fold increased risk of adverse outcomes following TAVR, including short‐ and long‐term mortality, prolonged hospital stay, and poor quality of life. , , , , , In the FRAILTY‐AVR (Frailty in Older Adults Undergoing Aortic Valve Replacement) study, a prospective cohort of older adults undergoing AVR at 14 centers across 3 countries that compared 7 different frailty scales, the Essential Frailty Toolset was the strongest predictor of adverse outcomes, predicting a 3.7‐fold increased risk of death and 2.1‐fold risk of disability at 1 year and should be considered the first‐line technique for evaluation of frailty in adults undergoing AVR. While this brief 4‐item scale, using chair rises, assessment of cognitive impairment, hemoglobin levels, and serum albumin to determine one’s frailty status, is straightforward to implement, it may be challenging to collect the requisite data to calculate it retrospectively. Similarly, while many other frailty metrics (eg, Fried index, Short Physical Performance Battery, etc) represent validated techniques for assessing frailty in adults undergoing AVR and may be considered alternatives to the Essential Frailty Toolset, these may also be challenging to apply retrospectively.

While not a surrogate for such validated assessments of frailty, there are multiple circumstances in which frailty is unmeasured in which retrospective ascertainment of one’s frailty status could be useful to improve risk stratification. As acquisition of in‐person measures of frailty may be limited by time, expense, and availability, missing frailty information may be an important source of unmeasured risk. In these settings, CFIs may represent a valid alternative for identifying one’s frailty status. As the current study indicates, in patients with severe aortic valve disease undergoing AVR, CFIs may capture relevant frailty information and thus allow retrospective classification of one’s frailty status to aid in evaluation of this important subgroup. This could be used at a hospital or health‐system level to identify populations of patients undergoing AVR for early mobilization, nutritional support, and intensive rehabilitation with the goal of possibly improving periprocedural outcomes or could be used for research to identify if observed effects are consistent across frailty groupings. , ,

As frailty is present in upwards of 60% of TAVR recipients (consistent with the findings from this study that 64.9% of individuals had frailty), it may represent an important unmeasured confounder in the valvular heart disease population and strongly influence treatment choice and outcomes. , In fact, consistent with prior observations that frailty may be a greater effect modifier of high‐risk interventions, the Johns Hopkins CFI was associated with a greater magnitude of risk in the SAVR than TAVR population in the current study. Despite being associated with greater risk in SAVR versus TAVR, it is unknown if CFIs could validly be used to identify differential benefit from one procedure versus the other and, as such, propose these as future topics of investigation. While disability and weight loss have been shown to be the most important individual components of frailty to discriminate among those likely to have poor outcomes after TAVR, it is unknown if these or other frailty‐related factors also identify individuals with the most benefit from TAVR versus SAVR.

Our study has several limitations. First, we only evaluate a single CFI, and it is possible that other CFIs may perform differently. As other published CFIs require outpatient and durable medical equipment files that are not as commonly used or widely available, we chose to focus the current analysis on the CFI by Segal et al, which was validated against the Fried index, often considered the gold standard for physical frailty. Nevertheless, these findings may not generalize to other CFIs. Second, there were insufficient data to construct the Fried index from available variables, and thus the CFI cannot be compared with this index to demonstrate construct validity and investigate the degree of misclassification. Nevertheless, as the Johns Hopkins CFI associated with multiple individual markers of frailty and disability, it remains valid for use in the valvular disease population. Third, our analysis was limited to a specific group of studies of TAVR linked to Medicare claims. Thus, whether similar results would be observed with studies of other populations or non‐Medicare claims is unclear. Fourth, because of overlapping age distributions between CFI tertiles, the effect of age cannot be removed despite adjustment. Fifth, though linked and nonlinked study participants were overall similar across a broad range of characteristics, it is possible that they are different across unmeasured characteristics that could influence the generalizability of the study results.

Conclusions

In linked Medicare and CoreValve study data, a CFI, anchored to a well‐developed construct of phenotypic frailty, consistently identified individuals with greater impairments in in‐person measures of frailty, disability, cognitive dysfunction, and malnutrition. The CFI identified individuals at higher risk of death, MACCE, hospitalization, and bleeding, independent of age and conventional risk metrics. These results that, while not a replacement for validated frailty assessments using in‐person measurement, the use of CFIs to retrospectively ascertain frailty status in studies of aortic valve disease missing these assessments may be valid and capture prognostically relevant information over and above traditional risk metrics.

Sources of Funding

The project was funded by grants from the National, Heart, Lung, and Blood Institute (1R01HL136708‐01 [Dr Yeh]; 1K23HL144907 [Dr Strom]).

Disclosures

Dr Strom reports additional grant funding from Edwards Lifesciences, Ultromics, HeartSciences, and Anumana; consulting for Bracco Diagnostics; and speaker fees from Northwest Imaging Forums, unrelated to the submitted work. Dr Yeh reports additional grant support from Abbott Vascular, AstraZeneca, BD Bard, Boston Scientific Cook, Philips Medical Medtronic and Zoll, and consulting fees from Abbott, AstraZeneca, Boston Scientific, Edwards Lifesciences, Medtronic, Shockwave Medical, and Zoll, outside the submitted work. Dr Shen is an employee of Biogen. The remaining authors have no disclosures to report.

Tables S1–S12

Figure S1

Click here for additional data file.