Unleashing Frailty from Laboratory into Real World: A Critical Step Towards Frailty-Guided Clinical Care of Older Adults

Abstract

Understanding patients’ degree of frailty is crucial for tailoring clinical care for older adults based on their physiologic reserve and health needs (“frailty-guided clinical care”). Two prerequisites for frailty-guided clinical care are: 1) access to frailty information at the point of care and 2) evidence to inform decisions based on frailty information. Recent advancements include web-based frailty assessment tools and their electronic health records integration for time-efficient, standardized assessments in clinical practice. Additionally, database frailty scores from administrative claims and electronic health records data enable scalable assessments and evaluation of the effectiveness and safety of medical interventions across different frailty levels using real-world data. Given limited evidence from clinical trials, real-world database studies can complement trial results and help treatment decisions for individuals with frailty. This article, based on the Thomas and Catherine Yoshikawa Award lecture I gave at the American Geriatrics Society Annual Meeting in Long Beach, California, on May 5, 2023, outlines our group’s contributions: 1) developing and integrating a frailty index calculator (Senior Health Calculator) into the electronic health records at an academic medical center; 2) developing a claims-based frailty index for Medicare claims; 3) applying this index to evaluate the effect of medical interventions for patients with and without frailty; and 4) efforts to disseminate frailty assessment tools through the launch of the eFrailty website and the forthcoming addition of the claims-based frailty index to the Centers for Medicare and Medicaid Services Chronic Conditions Data Warehouse. This article concludes with future directions for frailty-guided clinical care.

INTRODUCTION

Understanding older adults’ frailty is a core principle of geriatric medicine. Frailty is defined as a clinically identifiable state of diminished physiologic reserve and heightened vulnerability to adverse health outcomes.^1–4 Although a formal frailty assessment may not be performed in every clinical encounter, the concept of frailty fundamentally underpins the work of geriatricians. For robust individuals, geriatricians prioritize maintaining or enhancing physiologic reserve to delay disability and prolong life. For individuals with frailty, geriatricians prioritize managing conditions impacting their function and quality of life and preventing or minimizing the effect of stressors. By taking a patient’s frailty level into account, geriatricians avoid the pitfalls of making decisions based solely on chronological age or individual diseases.^5,6 This helps in avoiding treatments for patients who are unlikely to benefit and ensures that treatments are not withheld from those likely to benefit.⁷ From the perspective of the healthcare systems, older adults with frailty incur substantially higher total healthcare costs than those without frailty (mean annualized total Medicare costs increased from $5,724 for robust individuals to $44,586 for those with moderate-to-severe frailty).⁸ Moreover, those with frailty accounted for 51% of Medicare costs that could potentially be prevented.⁹ Identifying individuals with frailty can aid in the efficient allocation of resources and in targeting interventions that aim to reduce healthcare spending. Consequently, frailty assessment emerges as a critical tool for geriatricians and other healthcare providers, enabling them to deliver individualized, patient-centered care to older adults.

This special article highlights significant efforts to integrate frailty into the care of older adults led by our group and others in the field. It is not intended as a comprehensive literature review. The insights shared are derived from the lecture I presented at the American Geriatrics Society Annual Meeting in Long Beach, California, on May 5, 2023, after being honored with the Thomas and Catherine Yoshikawa Outstanding Scientific Achievement in Clinical Investigation Award. This article delineates the concept of frailty-guided clinical care, explores recent advancements in incorporating frailty within clinical practice and real-world database research, highlights initiatives to broaden the reach of frailty assessments beyond traditional geriatrics research, and concludes with next steps towards frailty-guided clinical care.

FRAILTY-GUIDED CLINICAL CARE

Randomized controlled trials (RCTs) have found several interventions to prevent or reverse frailty or improve associated outcomes, such as physical function, functional status, and falls, among older adults at risk for frailty or those living with frailty. These interventions include exercise and physical activity,^10–14 nutrition interventions,^13–15 comprehensive geriatric assessment (CGA)^16,17, and medication optimization.^18–21 Various geriatric care principles—such as Beers criteria for potentially inappropriate medications,²² framework for decision-making for older adults with multimorbidity,²³ time to benefit for preventive drug therapy^24–27 and cancer screening,^28,29 and patient-priorities care^30–32—along with models of patient-centered care like home-based primary care^33,34 and the Hospital-at-Home program,^35–37 are particularly useful for medically complex patients with frailty.

However, the heterogeneous health status of older adults complicates clinicians’ efforts to identify the most suitable interventions, care principles, and models of care for each patient. Despite the differences in participants across RCTs, it is possible to align the target population of each intervention along a continuum from fit to frail (Figure 1).³⁸ This facilitates a more streamlined approach to targeting interventions to the most suitable candidates.

Figure 1. A Proposed Approach to Frailty-Guided Clinical Care of Older Adults.

From The New England Journal of Medicine, D.H. Kim and Rockwood K, Frailty in Older Adults, Vol. 391, Pages 538-548, Copyright © 2024 Massachusetts Medical Society. Reprinted with permission.

This figure is adapted from Kim DH, Rockwood K. Frailty in Older Adults. N Engl J Med. 2024. It outlines a general approach to applying evidence from randomized controlled trials and best practices in geriatric care by considering the degree of frailty. Clinical management should focus on increasing physiologic reserve and preventing long-term adverse health outcomes for robust and pre-frail individuals; preserving physiologic reserve and mitigating the effect of stressors for individuals with frailty; and providing comfort and dignity for those approaching end-stage frailty (i.e., those who exhibit all 5 features of Fried Physical Frailty Phenotype, a deficit accumulation frailty index near 0.7, or complete dependence in personal care activities). The proposed frailty score cutpoints are not intended as strict rules and may be adjusted to the clinical context.

The rationale of frailty-guided clinical care is that healthcare providers can tailor clinical care and treatments to older individuals’ physiologic reserve and health needs by understanding their degree of frailty and vulnerable health domains.³⁸ There are two prerequisites for delivering frailty-guided clinical care: 1) ensuring access to frailty information at the point of care, and 2) having the evidence necessary to inform clinical decisions based on the frailty information. The following sections detail significant advancements in incorporating frailty assessment from controlled research environments into clinical care and research using real-world data.^39–41

INCORPORATING FRAILTY ASSESSMENT INTO CLINICAL CARE

More than 20 years have passed since the seminal papers on the Fried Physical Frailty Phenotype⁴² and the deficit accumulation frailty⁴³ were published, yet the adoption of frailty assessment in clinical care has been gradual. Possible reasons include lack of a consensus definition, time and resource limitations, low awareness, insufficient evidence on interventions, and the current disease-centered model of care that disincentivizes frailty assessment.^44–47 When implementing these assessments in clinical practice, there are important caveats that healthcare providers must consider for accurate measurement and interpretation.

The Fried Physical Frailty Phenotype defines frailty as a syndrome, which is characterized by 3 or more features of exhaustion, weakness, slowness, physical inactivity, and weight loss.⁴² While this definition is most commonly used in research,⁴⁸ implementing measurements and scoring of gait speed, grip strength, and a physical activity questionnaire can pose challenges in clinical practice, particularly in acute-care settings. Modifying the components of the Fried Physical Frailty Phenotype (e.g., omitting physical performance tests, substituting with self-report, using population-dependent cutpoints) may result in misclassification of frailty status.⁴⁹ For example, in a cohort with 26.7% prevalence of the Fried Physical Frailty Phenotype according to the original definition, the prevalence varied from 12.5% when gait speed test was omitted, to 21.0% when a population-dependent cutpoint was used, and to 31.2% when self-report was used.⁴⁹

The deficit accumulation frailty defines frailty as a state of poor health that results from accumulation of age-related health deficits.⁴³ A deficit accumulation frailty index is the proportion of deficits present out of the deficits assessed. While it can range from 0 to 1, 99% of the population has a value less than 0.7.^50,51 A cutpoint between 0.20 and 0.25 has been used to define frailty.^52–54 A frailty index should contain at least 30 deficits representing multiple organ-physiologic systems.⁵⁵ Although the composition of a frailty index has little impact on its discriminatory ability for adverse health outcomes,⁵² different compositions (e.g., excluding functional status or performance tests) can influence the absolute value of the frailty index.^56,57 This is a critical consideration when clinical decisions depend on a frailty index threshold.

To enable a time-efficient and standardized frailty assessment in clinical care, several user-friendly web-based calculators and applications have been developed.⁵⁸ One such tool, developed by our group, is the Senior Health Calculator. This 50-item deficit accumulation frailty index based on routinely collected information from CGA (available at: https://www.bidmc.org/research/research-by-department/medicine/gerontology/calculator) has been integrated into the electronic health records (EHR) at Beth Israel Deaconess Medical Center in Boston, Massachusetts. The frailty index can be calculated without obtaining all performance tests, allowing use in acute-care settings and during telehealth visits. Telephone-based frailty assessment using this calculator could be successfully completed in 82.2% of older adults with serious chronic illness at an academic geriatrics clinic.⁵⁹ The average administration time was 26.1 minutes, reducing to 15.7 minutes when the Telephone Montreal Cognitive Assessment and the Geriatric Depression Scale were not performed.⁵⁹ Moreover, it provides patients’ level of impairment in each assessment domain and their frailty index value in reference to the age and sex-matched mean and 99^th percentile values in the US population.⁶⁰ As illustrated in two hypothetical patients (Figure 2 and Figure 3), this information is valuable in determining the target areas for interventions and in facilitating discussions about prognosis and decision-making before stressful treatments in both inpatient and outpatient settings. The Senior Health Calculator can aid in decision-making about transcatheter aortic valve replacement by providing the expected trajectory of functional status recovery.⁶¹

Figure 2. Results from Frailty Assessment of an 80-Year-Old Woman Patient A.

The Senior Health Calculator is a 50-item deficit accumulation frailty index based on routinely collected information from comprehensive geriatric assessment (available at: https://www.bidmc.org/research/research-by-department/medicine/gerontology/calculator). The radar plot displays the level of impairment in each assessment domain, which can be useful for developing individualized interventions. The line plot shows the age and sex-specific mean values (men: blue solid line; women: pink solid line) and the 99^th percentile values (men: blue dash line; women: pink dash line). The 99^th percentile values are around 0.7, which means only 1% of the population has a value higher than 0.7. Patient A is an 80-year-old woman who has low medical complexity and moderately impaired muscle strength. Her frailty index of 0.133 (arrow) is lower than the mean value of women in the same age range.

Figure 3. Results from Frailty Assessment of an 80-Year-Old Woman Patient B.

Patient B is an 80-year-old woman who has high medical complexity, moderately impaired mobility, severely impaired muscle strength, mild activities of daily living disability, severe instrumental activities of daily living disability, severe cognitive impairment, and mildly impaired nutritional status. Her frailty index of 0.634 (arrow) is close to the top 1% value of women in the same age range. This patient is at high risk of dying and functional decline when she experiences stressful events, such as surgical procedures.

INCORPORATING FRAILTY ASSESSMENT INTO REAL-WORLD DATA

The gradual adoption of frailty assessment in clinical practice has motivated researchers to develop innovative ways to estimate the frailty level of patients in routine care using existing real-world data sources, such as administrative claims data and EHR data.^39–41 The premise is that, using diagnosis codes, procedure codes, and health service codes as proxies of frailty, it is possible to estimate an individual’s frailty level.⁶² The main advantage of database frailty scores is their scalability; that is, the frailty level can be estimated for a large population (e.g., members of a healthcare system or insurance plan) without in-person clinical assessments. Their main disadvantage is measurement error and misclassification of frailty status. This issue can result from several sources: 1) reliance on historical records or claims (e.g., treating a resolved diagnosis as an active condition, interpreting the absence of a diagnosis code as the absence of the condition); 2) lack of detailed clinical information in claims (e.g., functional status and laboratory test results); 3) differences in access to health care; 4) selective data capture (e.g., diagnostic tests and procedures are performed when there is suspicion for a diagnosis); and 5) incomplete measurement resulting from disenrollment from an insurance plan or fragmented care across multiple healthcare systems that use different EHRs. Database frailty scores tend to overestimate frailty in the lower end of the fit-to-frail spectrum and underestimate it in the upper end.⁶³

Over the past 10 years, several database frailty scores have been developed^39–41; Table 1 shows a representative (but not comprehensive) list of frailty score models for administrative claims and EHR data. The Fried Physical Frailty Phenotype cannot be directly measured because gait speed, grip strength, and physical activity are rarely available in real-world data. Drubbel et al.,⁶⁴ Clegg et al.,^65–67 Orkaby et al.,^68,69 and Pajewski et al.⁵⁴ directly implement the deficit accumulation method by counting the number of health deficits, which are defined by a curated list of related codes based on clinical knowledge. Pajewski et al.⁵⁴ included health deficits defined by vital signs, laboratory tests, and functional status from the Medicare Annual Wellness Visit in addition to diagnosis codes. Glibert et al. performed a cluster analysis of inpatient claims, identified a cluster that likely represented patients with frailty,^70,71 and then developed a simple scoring system for the probability of being in this frailty cluster. Davidoff et al.,⁷² Faurot et al.,^73,74 Segal et al.,^75,76 and our group^77–79 used a Medicare dataset linked to a survey or cohort in which a clinical frailty assessment was available to develop a regression model that estimates the probability of disability (as a proxy of frailty),^72–74 the probability of the Fried Physical Frailty Phenotype,^75,76 or the mean deficit accumulation frailty index score.^77–79 Ruiz et al. validated that the Care Assessment Need score, which was originally developed for predicting hospital admission or death based on the US veterans affairs EHR data, was correlated with clinical frailty measures.^80–82

Table 1.

Commonly Used Frailty Score for Real-World Data

First Author (Year)	Database	Predictor Selection	Predictor Assessment Period	Types of Predictors	Interpretation of the Score
Frailty Scores for Administrative Claims Data
Davidoff (2013)72	Medicare claims data, United States	Regression-based (Reference standard: disability)	12 months	• Sex • Diagnosis • Procedure & health service • Medicaid enrollment status • Health care utilization	Range: 0-1 0 to < 0.11: non-frail 0.11 to 1: frail
Faurot (2015)73,74	Medicare claims data, United States	Regression-based (Reference standard: disability)	8 months	• Age, sex, race • Diagnosis • Procedure & health service	Range: 0-1 0 to < 0.05: non-frail 0.05 to 1: frail
Segal (2017)75,76	Medicare claims data, United States	Regression-based (Reference standard: Fried frailty phenotype)	6 months	• Age, sex, race • Diagnosis • Comorbidity index • Health care utilization	Range: 0-1 0 to < 0.12: non-frail 0.12 to 1: frail
Kim (2018)77–79	Medicare claims data, United States	Regression-based (Reference standard: deficit accumulation frailty index)	12 months	• Diagnosis • Procedure & health service	Range: 0-1 0 to < 0.15: robust 0.15 to < 0.25: pre-frail 0.25 to < 0.35: mildly frail 0.35 to < 0.45: moderately frail 0.45 to 1: severely frail
Gilbert (2018)70,71	Inpatient claims data, England	Cluster analysis (No reference standard; prediction of the frailty cluster)	24 months	• Diagnosis	Range: 0-173.2 0 to < 5: low risk 5 to < 15: intermediate risk 15 to 173.2: high risk
Orkaby (VA-FI) (2019)68,69	Veterans Affairs claims data, United States	Clinical knowledge (No reference standard; direct implementation of deficit accumulation)	36 months	• Diagnosis • Procedure & health service	Range: 0-1 0 to ≤ 0.1: robust 0.1 to ≤ 0.2: pre-frail 0.2 to ≤ 0.3: mildly frail 0.3 to ≤ 0.4: moderately frail 0.4 to 1: severely frail
Frailty Scores for EHR Data
Drubbel (2013)64	Primary care EHR data, Netherlands	Clinical knowledge (No reference standard; direct implementation of deficit accumulation)	12 months	• Diagnosis • Polypharmacy	Range: 0-1 0 to 0.03: low risk 0.04 to 0.13: intermediate risk 0.14 to 1: high risk
Clegg (eFI) (2016)65–67	Primary care EHR data, United Kingdom	Clinical knowledge (No reference standard; direct implementation of deficit accumulation)	12 months	• Diagnosis • Procedure & health service • Pharmacy • Social determinants of health	Range: 0-1 0 to 0.12: fit 0.12 < to 0.24: mildly frail 0.24 < to 0.36: moderately frail 0.36 to 1: severely frail
Ruiz (Care Assessment Need score) (2018)80–82	Veterans Affairs EHR data, United States	Regression-based (Reference standard: hospital admission or death)	12 months	• Age, sex, marital status, service connection • Diagnosis • Vital signs • Health care utilization • Pharmacy • Laboratory tests	Range: 0-100 0 to < 95: low risk 95 to 100: high risk
Pajewski (2019)54	EHR data of an Accountable Care Organization, United States	Clinical knowledge (No reference standard; direct implementation of deficit accumulation)	24 months	• Diagnosis • Vital signs • Laboratory tests • Functional status • Self-rated health, smoking • Polypharmacy	Range: 0-1 0 to 0.10: fit 0.10 < to 0.21: pre-frail 0.21 to 1: frail

Abbreviations: BMI, body mass index; eFI, electronic frailty index; EHR, electronic health records.

Adapted from Kim DH, Park CM, Ko D, Lin KJ, Glynn RJ. Assessing the Benefits and Harms of Pharmacotherapy in Older Adults with Frailty: Insights from Pharmacoepidemiologic Studies of Routine Health Care Data. Drugs Aging. 2024. In Press.¹⁰⁸

The claims-based frailty index (CFI) we developed (Kim CFI) for Medicare claims estimates a deficit accumulation frailty index using 93 variables defined by International Classification of Diseases codes, Current Procedural Terminology codes (nursing facility care), and Healthcare Common Procedure Coding System codes (e.g., hospital beds, wheelchairs, and transportation services) in the past year (available at: https://dataverse.harvard.edu/dataverse/cfi).^77–79 The Kim CFI outperforms other Medicare database frailty scores in identifying the Fried Physical Frailty Phenotype (C statistic: 0.78 vs 0.73-0.74) and correlates more strongly with a deficit accumulation frailty index (correlation: 0.59 vs 0.22-0.45).⁷⁹ The prespecified CFI cutpoint of 0.25 offers 62% sensitivity and 78% specificity for the Fried Physical Frailty Phenotype and 60% sensitivity and 86% specificity for a clinical deficit accumulation frailty index ≥0.25 (Table 2).⁶³ It was inversely correlated with gait speed and grip strength.⁷⁸ Although both the Kim CFI and the Charlson Comorbidity Index were associated with mortality and hospitalization, only the CFI was independently associated with institutionalization, disability, and prolonged skilled nursing facility stay.⁷⁸ In older adults who were hospitalized and then discharged to skilled nursing facility before returning home, nearly 60% of those with CFI <0.20 recovered by 45 days of home health, whereas only 33% of those with CFI ≥0.30 did.⁸³ Moreover, changes in the CFI over 1 year were associated with total healthcare costs in the subsequent year,⁸⁴ and adding the CFI to the Centers for Medicare and Medicaid Services (CMS) Hierarchical Condition Category model improved the cost prediction over the current CMS model.⁸

Table 2.

Performance of the Kim Claims-Based Frailty Index Against Clinical Frailty Assessments at Selected Cutpoints

Clinical frailty Assessment	Cutpoint Characteristic	CFI Cutpoint	Sensitivity	Specificity	PPV	NPV
Fried Physical Frailty Phenotype (≥ 3)	Prespecified	0.250	0.62 (0.58-0.67)	0.78 (0.76-0.79)	0.33 (0.31-0.36)	0.92 (0.91-0.93)
	Optimal	0.237	0.71 (0.66-0.75)	0.74 (0.72-0.75)	0.32 (0.30-0.35)	0.93 (0.92-0.95)
	80% sensitivity	0.207	0.80 (0.76-0.84)	0.61 (0.59-0.63)	0.27 (0.25-0.29)	0.94 (0.93-0.96)
	80% specificity	0.257	0.59 (0.55-0.64)	0.80 (0.79-0.81)	0.34 (0.32-0.37)	0.92 (0.90-0.93)
Deficit accumulation frailty index (≥ 0.25)	Prespecified	0.250	0.60 (0.57-0.63)	0.86 (0.85-0.88)	0.66 (0.63-0.70)	0.83 (0.81-0.84)
	Optimal	0.227	0.71 (0.68-0.74)	0.78 (0.77-0.80)	0.60 (0.57-0.63)	0.86 (0.84-0.87)
	80% sensitivity	0.200	0.80 (0.77-0.83)	0.66 (0.63-0.68)	0.51 (0.48-0.54)	0.88 (0.86-0.90)
	80% specificity	0.232	0.68 (0.65-0.71)	0.80 (0.78-0.82)	0.61 (0.58-0.64)	0.85 (0.83-0.86)

Abbreviations: CFI, claims-based frailty index; PPV, positive predictive valve; NPV, negative predictive value.

Adapted from Sison SDM, Shi SM, Oh G, Jeong S, McCarthy E, Kim DH. Claims-Based Frailty Index and Its Relationship with Commonly Used Clinical Frailty Measures. J Gerontol A Biol Sci Med Sci. 2024. doi: 10.1093/gerona/glae094.⁶³

EVALUATING THE OUTCOMES OF MEDICAL INTERVENTIONS BY FRAILTY IN REAL-WORLD DATA

The uptake of newly approved treatments is delayed in older adults with frailty.^85–87 Clinicians may be concerned about treatment-related harms, unclear time-to-benefit, and insufficient evidence due to under-representation of individuals with frailty in RCTs. Although patients with frailty may have been enrolled in RCTs,⁸⁸ a frailty subgroup analysis is not possible without a formal frailty assessment. Lately, researchers constructed a deficit accumulation frailty index using variables collected at baseline to evaluate treatment effects by frailty levels in RCTs.^89–94 However, these frailty subgroup analyses were often underpowered and subject to multiple testing.⁹⁵ Because it is unknown how post-hoc frailty scores from RCTs relate to clinical frailty measures, clinicians may not easily identify patients to whom trial results can apply.⁹⁶ Real-world database studies utilizing database frailty scores might overcome these limitations and complement RCT post-hoc analyses.

Our group and others applied a database frailty score to evaluate the effectiveness and safety of drug therapy by patients’ frailty levels (Table 3). Some studies revealed meaningful variations in treatment effects by frailty levels.^97–101 In Medicare beneficiaries with atrial fibrillation, dabigatran and rivaroxaban were associated with lower rates of clinical events compared with warfarin only among non-frail patients; however, apixaban was associated with lower event rates compared with warfarin in both non-frail and frail patients.⁹⁷ Patients treated with rivaroxaban or warfarin instead of apixaban were more likely to lose days at home over 1 year; this difference was more pronounced in those with frailty than those without.⁹⁸ In older veterans with multiple myeloma, DuMontier et al. compared a 3-drug regimen with a 2-drug regimen.⁹⁹ They found that the mortality reduction associated with the 3-drug regimen, compared with the 2-drug regimen, was somewhat more evident in those with frailty than in those without frailty. In older veterans with metastatic prostate cancer, Deol et al. showed that the mortality reduction associated with enzalutamide, relative to abiraterone, appeared to be greater in frail patients compared to the effect observed in non-frail patients.¹⁰⁰ In a comparative study of mRNA vaccines for Coronavirus Disease 2019 (COVID-19), robust Medicare beneficiaries who received the mRNA-1273 vaccine experienced lower rates of facial nerve palsy and thrombocytopenic purpura, and possibly lower rates of myocarditis, pericarditis, and COVID-19 diagnosis, compared to those who received the BNT162b2 vaccine; however, these differences were attenuated in frail individuals.¹⁰¹

Table 3.

Real-World Data Studies Evaluating the Heterogeneity of Treatment Effects of Pharmacotherapy by Frailty Status

Author	Study Population	Frailty	Outcome	Treatments	Effect Estimate (95% CI)
					Non-frail	Frail
Cardiovascular Medications
Kim (2021)97	Medicare beneficiaries with non-valvular AF	Kim CFI	Composite of death, stroke, or major bleeding	Dabigatran vs warfarin	HR: 0.81 (0.68, 0.97)	1.09 (0.96, 1.23)
				Rivaroxaban vs warfarin	HR: 0.88 (0.77, 0.99)	0.96 (0.89, 1.04)
				Apixaban vs warfarin	HR: 0.61 (0.52, 0.71)	0.73 (0.67, 0.80)
Lin (2023)98	Medicare beneficiaries with non-valvular AF	Kim CFI	Home time loss ≥ 14 days in 1 year	Rivaroxaban vs apixaban	RD: 0.8% (0.5, 1.2)	2.7% (1.9, 3.4)
				Warfarin vs apixaban	RD: 1.7% (1.2, 2.2)	3.6% (2.8, 4.3)
Anderson (2023)102	Hospitalized veterans with elevated BP	VA-FI	Composite inpatient adverse events	Intensification of BP regimen vs no intensification	OR: 1.26 (1.13, 1.41)	1.29 (1.15, 1.45)
Orkaby (2023)103	Veterans with no prior CVD	VA-FI	Death	Statin vs no statin	HR: 0.61 (0.61, 0.62)	0.63 (0.62, 0.64)
			MACE	Statin vs no statin	HR: 0.87 (0.86, 0.88)	0.90 (0.88, 0.92)
Diabetes Medications
Dave (2019)105	Adults with type 2 diabetes	Kim CFI	Severe UTI	SGLT2i vs DPP4i	HR: 1.00 (0.46, 2.15)	0.84 (0.49, 1.43)
				SGLT2i vs GLP1RA	HR: 1.12 (0.58, 2.16)	0.64 (0.40, 1.02)
Zhuo (2021)106	Medicare beneficiaries with type 2 diabetes	Kim CFI	Fractures	SGLT2i vs DPP4i	HR: 1.28 (0.73, 2.25)	0.93 (0.62, 1.41)
				SGLT2i vs GLP1RA	HR: 1.33 (0.74, 2.39)	1.09 (0.69, 1.70)
Htoo (2023)107	Medicare beneficiaries with type 2 diabetes	Kim CFI	Severe hypoglycemia	SGLT2i vs DPP4i	HR: 0.78 (0.59, 1.03)	0.83 (0.69, 1.00)
				SGLT2i vs GLP1RA	HR: 1.03 (0.79, 1.35)	0.87 (0.74, 1.04)
Kutz (2023)104	Medicare beneficiaries with type 2 diabetes	Kim CFI	Death or MACE	SGLT2i vs DPP4i	HR: 0.74 (0.68, 0.81)	0.79 (0.70, 0.89)
				GLP1RA vs DPP4i	HR: 0.75 (0.68, 0.82)	0.83 (0.76, 0.91)
			Composite safety endpoints	SGLT2i vs DPP4i	HR: 0.90 (0.74, 0.87)	0.89 (0.78, 1.01)
				GLP1RA vs DPP4i	HR: 0.89 (0.82, 0.97)	1.01 (0.92, 1.10)
Antineoplastic Agents
DuMontier (2023)99	Veterans with multiple myeloma, ineligible for bone marrow transplant	VA-FI	Survival	Triplet vs doublet regimen	HR: 0.86 (0.67, 1.10)	0.74 (0.56, 0.97)
Deol (2023)100	Veterans with castration resistant metastatic prostate cancer	VA-FI	Survival	Enzalutamide vs abiraterone	HR: 0.93 (0.86, 1.01)	0.85 (0.77, 0.93)
Vaccines
Harris (2023)101	Community-dwelling Medicare beneficiaries with no recent COVID-19	Kim CFI	Facial nerve palsy	mRNA-1273 vs BNT162b2	RR: 0.86 (0.75, 0.99)	1.14 (0.89, 1.45)
			Thrombocytopenic purpura	mRNA-1273 vs BNT162b2	RR: 0.89 (0.80, 0.99)	0.96 (0.77, 1.20)
			Pulmonary embolism	mRNA-1273 vs BNT162b2	RR: 0.94 (0.88, 1.00)	1.00 (0.92, 1.08)
			Myocarditis or pericarditis	mRNA-1273 vs BNT162b2	RR: 0.76 (0.56, 1.03)	0.96 (0.58, 1.61)
			COVID-19 diagnosis	mRNA-1273 vs BNT162b2	RR: 0.85 (0.82, 0.88)	0.94 (0.89, 0.99)

Abbreviations: BP, blood pressure; CFI, claims-based frailty index; CI, confidence interval; COVID-19, Coronavirus Disease 2019; CVD, cardiovascular disease; DPP4i, dipeptidyl peptidase-4 inhibitor; GLP1RA, glucagon-like peptide-1 receptor agonist; HR, hazard ratio; MACE, major adverse cardiovascular events; OR, odds ratio; RD, risk difference; RR, risk ratio; SGLT2i, sodium-glucose cotransporter-2 inhibitor; UTI, urinary tract infection; VA-FI, Veterans Affairs Frailty Index.

Adapted from Kim DH, Park CM, Ko D, Lin KJ, Glynn RJ. Assessing the Benefits and Harms of Pharmacotherapy in Older Adults with Frailty: Insights from Pharmacoepidemiologic Studies of Routine Health Care Data. Drugs Aging. 2024. In Press.¹⁰⁸

Other studies found no evidence for the heterogeneity of treatment effects by frailty levels (Table 3).^102–107 In a cohort of older hospitalized veterans with elevated blood pressure readings within 48 hours of admission, intensification of antihypertensive regimen was associated with an increased risk of inpatient adverse events in both frail and non-frail patients.¹⁰² In another study of older veterans without history of cardiovascular disease, statin was associated with a similar reduction in death and cardiovascular events, regardless of frailty levels.¹⁰³ Kutz et al. examined the effects of sodium-glucose cotransporter-2 inhibitors and glucagon-like peptide-1 receptor agonists, compared with dipeptidyl peptidase-4 inhibitors in Medicare beneficiaries with type 2 diabetes.¹⁰⁴ Their results suggest that sodium-glucose cotransporter-2 inhibitors and glucagon-like peptide-1 receptor agonists are more effective and safer than dipeptidyl peptidase-4 inhibitors, regardless of frailty levels. Studies evaluating severe urinary tract infection, fractures, and severe hypoglycemia were inconclusive due to the low event rates.^105–107

The lessons from real-world data studies of comparative effectiveness and safety of drug therapy are summarized as follows: 1) the rate of adverse clinical events increases as the frailty level rises; 2) despite the elevated risk of adverse events, being frail does not imply lack of treatment benefit; and 3) because of the high event rate, the absolute event reduction may be greater in individuals with frailty. These results should be interpreted within the limitations inherent in real-world database studies, such as unmeasured confounding and measurement error in assessing frailty and clinical events in real-world data.¹⁰⁸ Nonetheless, given the scarcity of RCT evidence for older adults with frailty, real-world database studies hold the potential to guide medication management in this population and to generate hypotheses for future RCTs.

ENHANCING THE ACCESSIBILITY OF FRAILTY ASSESSMENTS BEYOND GERIATRICS RESEARCH

Inspired by the geriatrician researchers who developed the ePrognosis website to disseminate prognostic models, our group has developed eFrailty (available at: efrailty.org) in an effort to make frailty assessments more accessible to clinicians and researchers. The eFrailty website helps users choose a frailty tool based on their purpose and clinical context (screening or brief risk assessment, comprehensive assessment and care planning, or risk assessment before stressful treatments) and availability of cognitive and physical performance tests. It hosts 15 validated frailty assessments and web-based calculators (10 tools for general populations and 5 tools for specific patient populations), including the Fried Physical Frailty Phenotype,⁴² the Senior Health Calculator (a CGA-based deficit accumulation frailty index),⁵⁹ the Clinical Frailty Scale,^109,110 the Fatigue, Resistance, Ambulation, Illnesses, and Loss of Weight scale,^111,112 the Essential Frailty Toolset before aortic valve replacement,^61,113 the Risk Analysis Index before major surgery,^114,115 and the Liver Frailty Index for patients with end-stage liver disease.^116,117 A crosswalk has been created to facilitate comparison across 9 commonly used frailty tools.¹¹⁸ This resource can be valuable to trainees and clinicians who are unfamiliar with various frailty assessment tools. Moreover, the U.S. Department of Health and Human Services Assistant Secretary for Planning and Evaluation recommended that the Kim CFI be added to the CMS Chronic Conditions Data Warehouse for its dissemination and use by CMS, researchers, and other stakeholders.¹¹⁹

FUTURE DIRECTIONS TOWARDS FRAILTY-GUIDED CLINICAL CARE

For frailty-guided clinical care to become a reality, the concept and measurement of frailty, which have traditionally been the interest of a select group of geriatricians and researchers, need to be “unleashed” to patients, the broader clinical and research communities, and policymakers. The development of web-based frailty calculators and database frailty scores provides more options of frailty assessment modalities for use in clinical care and population screening. While increasing numbers of healthcare systems are incorporating database frailty scores in their EHR,^{54,65–67,80–82} the most time- and resource-efficient method for integrating database frailty scores in clinical workflow remains to be determined. Because of the measurement errors in database frailty scores, it is crucial to test their validity (beyond predictive validity) and performance (including sensitivity and specificity) against clinical frailty assessments before their clinical application. Given that several clinical frailty assessments exist, which clinical assessment was used as a reference standard for the validation of a database frailty score should be reported. Measurement errors could potentially be reduced by additionally integrating unstructured EHR data through natural language processing and deep learning.^120–123 Substituting clinical frailty assessments with database frailty scores remains unlikely until their validity is confirmed. If critical clinical decisions hinge on a frailty assessment, patients with a high database frailty score should undergo further evaluation with a clinical frailty assessment to avoid false positive and false negative classifications. There is growing interest in the use of wearable technology, sensors, and artificial intelligence to measure physical frailty and its characteristics (e.g., physical activity, gait parameters, and postural transition).^124–127 This modality may offer a less intrusive way to screen and monitor frailty outside clinical settings.

The impact of frailty screening and assessment on care quality, health outcomes, and healthcare costs depends on how clinicians and healthcare systems utilize frailty information to optimize clinical care and reduce waste in healthcare delivery. The utility of frailty assessment is threefold: 1) it identifies patients at high risk for nearly all age-related poor health outcomes and increased healthcare spending (strong evidence); 2) it enables the development of individualized care plans to improve frailty and its associated outcomes (moderate evidence); and 3) it informs decisions regarding medical interventions (weak evidence). More RCTs are needed to evaluate interventions that can prevent or reverse frailty, and reduce functional decline, falls, and long-term institutionalization, and to compare various strategies for implementing these proven interventions in clinical practice. Additionally, routine use of database frailty scores in real-world database studies can provide evidence and hypotheses regarding the comparative effectiveness and safety of medical interventions across different levels of patients’ frailty, thereby informing treatment decision-making.¹⁰⁸ Lastly, education and training for clinicians and healthcare system leaders are essential to translate research findings into improved clinical care and population management.

Key points.

• Understanding patients’ degree of frailty is crucial for tailoring clinical care for older adults based on their physiologic reserve and health needs.

• The development of web-based frailty calculators and database frailty scores provides more options of frailty assessment modalities for use in clinical care and population screening.

• Given limited evidence from clinical trials, real-world database studies can provide timely evidence and generate hypotheses to inform clinical management for older adults with frailty.

Why does this paper matter?

The potential utility of frailty assessment is threefold: 1) it identifies patients at high risk for poor health outcomes; 2) it enables the development of individualized care plans; and 3) it informs decisions regarding medical interventions. The real impact of frailty assessment will depend on how clinicians and healthcare systems utilize frailty information to optimize clinical care and reduce waste in healthcare delivery. This article delineates the concept of frailty-guided clinical care, explores recent advancements in incorporating frailty within clinical practice and real-world database research, highlights initiatives to broaden the reach of frailty assessments beyond traditional geriatrics research, and concludes with next steps towards frailty-guided clinical care.