New horizons in frailty identification via electronic frailty indices: early implementation lessons from experiences in England and the United States

Abstract

Frailty represents an integrative prognostic marker of risk that associates with a myriad of age-related adverse outcomes in older adults. As a concept, frailty can help to target scarce resources and identify subgroups of vulnerable older adults that may benefit from interventions or changes in medical management, such as pursing less aggressive glycaemic targets for frail older adults with diabetes. In practice, however, there are several operational challenges to implementing frailty screening outside the confines of geriatric medicine. Electronic frailty indices (eFIs) based on the theory of deficit accumulation, derived from routine data housed in the electronic health record, have emerged as a rapid, feasible and valid approach to screen for frailty at scale. The goal of this paper is to describe the early experience of three diverse groups in developing, implementing and adopting eFIs (The English National Health Service, US Department of Veterans Affairs and Atrium Health—Wake Forest Baptist). These groups span different countries and organisational complexity, using eFIs for both research and clinical care, and represent different levels of progress with clinical implementation. Using an implementation science framework, we describe common elements of successful implementation in these settings and set an agenda for future research and expansion of eFI-informed initiatives.

Key Points

• Electronic frailty indices (eFIs) represent an automated, valid and scalable approach to screen for frailty.

• We describe the early experience of three diverse groups in developing and adopting eFIs.

• We summarise common elements of successful eFIs implementation and set an agenda for future eFIs-informed research.

Introduction

Frailty provides a cohesive assessment of vulnerability and prognostication reflecting the heterogeneity of ageing [1–3]. Frailty is not a disease, but rather an integrative marker of risk, independent of age, that helps to identify vulnerable older adults across diverse healthcare settings. Compared to tailored clinical risk prediction models [4, 5], the appeal of frailty is that it represents a composite measure of risk. Frailty captures biologic ageing beyond chronologic age and is associated with numerous adverse outcomes beyond mortality, including incident disability, injurious falls, healthcare utilisation and post-operative complications [6].

Whilst there is consensus on the importance of routine frailty identification across care settings, including primary care, specialty clinics and prior to surgery [7–9], consensus is lacking on how routine frailty identification can be operationalised for population health management (PHM) for a health system or at a national level [10, 11]. Although in-person comprehensive geriatric assessment (CGA) is the gold-standard evaluation for frailty, it does not scale well given the severe shortage of geriatricians and professionals with necessary expertise [12]. Similarly, other frailty constructs, such as the frailty phenotype, [1] necessitate functional measures not routinely performed in most clinical settings. In contrast, electronic frailty indices (eFIs) based on the theory of deficit accumulation have emerged as a rapid, feasible and valid approach to screen for frailty at scale [13, 14]. An eFI can be run in the background on existing clinical data housed within the electronic health record (EHR), passively generating a score that identifies individuals more or less likely to be frail.

Whilst the feasibility and validity of computing an eFI has been demonstrated in multiple countries and settings [13, 15–18], building a risk score is just the first step in developing a comprehensive approach to routine clinical frailty identification and management. Older adults will only benefit from eFI-based risk stratification if there is a purposeful connection to clinical scenarios in which knowledge of frailty may improve medical decision-making or aid in identifying patients likely to need specialised care (Figure 1) [12, 19]. However, there are numerous challenges to identifying, prioritising and testing where frailty screening will be the most useful. The goal of this paper is to describe the early experience of three diverse groups in developing and adopting eFIs. These groups span different countries and healthcare systems, using eFIs for both research and clinical care, and represent different levels of progress with clinical implementation. We also utilise an implementation science framework to retrospectively explain what contextual factors and features of eFIs we feel influenced progress with implementation across three different health systems.

Figure 1.

Potential opportunities for clinical interventions informed by frailty screening. ACP denotes advanced care planning. Many of the potential interventions could be utilised across the spectrum of frailty. For example, physical activity or other exercise interventions could also be beneficial in moderate frailty.

Settings for eFI implementation

The English National Health Service (NHS), publicly funded through general taxation, provides universal, free healthcare to 56 million residents. NHS England is the national organisation responsible for overseeing budgetary planning and operational delivery of NHS commissioning. Primary care, provided primarily by general practitioners (GPs), is the bedrock of the English NHS, with approximately 90% of all patient contact occurring in this setting. Alongside their core role of delivering primary care, GPs are also the gatekeepers for secondary care services. The majority of GPs operate under a contractual arrangement with NHS England. All primary care practices in England have well-established EHR systems, many since the late 1990s, with an individual NHS number serving as a unique national identifier for each patient. There are four commercial suppliers of primary care EHR systems in England—TPP, EMIS, Vision and Microtest.

The US Department of Veterans Affairs (VA) system is the largest integrated healthcare system in the USA, having served over 24 million veterans since its establishment in 1930 as a federal agency. Veterans who served in the active military, naval or air service and did not receive a dishonourable discharge are eligible to apply for Veterans Health Administration benefits. In 2023, the VA system includes 172 medical centres and 1,113 outpatient clinics, with over 350,000 staff serving over 9 million active patients. The VA has used a single electronic health record, CPRS (Computerized Patient Record System), since the 1980s across all healthcare settings. Records are readily accessible at any VA facility ensuring continuity of care regardless of where veterans are seen across the country. Efforts to modernise the current EHR system are ongoing.

Atrium Health—Wake Forest Baptist (AH-WFB) is a six-hospital health system in the Piedmont region of the southeastern USA. Its catchment area is a 24-county region including western North Carolina and southern Virginia. Primary care is organised through an integrated network of over 165 practices and more than 1,000 primary care providers. The majority of older adults with an affiliated primary care provider are covered under value-based contracts, typically part of accountable care organisations (ACOs). ACOs are voluntary collaborations amongst doctors, hospitals and other clinicians to provide coordinated care for Medicare and other insured patients [20]. When an ACO meets both quality and cost metrics, it shares in the savings it has achieved. As of April 2023, there were AH-WFB partners with over 20 ACOs, engaging with approximately 90,000 patients 65 years or older, which is over 80% of patients with a system-affiliated primary care provider in this age group. AH-WFB has used the Epic (Madison, WI) EHR system across its inpatient and outpatient settings since 2013.

History and progress with eFI implementation

England NHS

The well-established use of the EHR in the English NHS and related availability of large anonymised primary care research databases linked with secondary care data provided a strong foundation for the development, validation and national implementation of an eFI based on primary care EHR data. A 36-item eFI was developed using anonymised data from approximately 500,000 patients in the ResearchOne primary care research dataset, drawn from the TPP/SystmOne primary care EHR and working in full partnership with the commercial supplier, validated for the outcomes of hospitalisation, care home admission and mortality. The eFI was externally validated using data from about 500,000 patients in The Health Improvement Network (THIN) dataset, drawn from the Vision primary care EHR system [13]. The 3-year area under the curve (AUC) for mortality were 0.70 (internal validation) and 0.75 (external validation), with similar results for emergency hospitalisation and nursing home admission. Once validated [14], the strong links that were developed with the commercial supplier supported rapid implementation into the TPP/SystmOne EHR system, with subsequent implementation into the other three system suppliers over the subsequent 12 months. The eFI was made available to all the systems suppliers at no cost on the basis that a premium charge would not then be made to the end user—typically a GP or other primary care clinician.

VA

Similar to the English NHS, the long-term use of the national EHR in VA has allowed for large data analytics that pool clinical data with external claims data from the Centers for Medicare and Medicaid Services. Using these data sources, a 31-item eFI (the VA-FI) was developed and validated in over 3 million veterans with a primary care visit from 2002 to 2012 [15, 21]. The unadjusted 3-year AUC for mortality was 0.70 and was 0.71 for long-term institutionalisation [22]. The VA-FI was also validated against an in-person frailty index based on CGA and functional assessments, confirming construct validity [23]. Versions of the VA-FI based on either the EHR only, or the EHR linked to external claims, have been validated and updated to be compatible with the International Classification of Diseases, Tenth Revision (ICD-10) [24]. This has made the VA-FI a useful tool for both clinical and research needs.

AH-WFB

The eFI in AH-WFB began as a retrospective analysis demonstrating local feasibility and validity. This work demonstrated the feasibility of calculating the eFI in a more open health system (compared to the English NHS and VA), the strong association of the eFI with incident outcomes and the incremental benefit of incorporating functional assessments from the Medicare Annual Wellness Visit [16]. The unadjusted AUC was 0.75 for 1-year mortality, 0.72 for inpatient admissions and 0.75 for injurious falls. In parallel, potential clinical applications were informed by qualitative interviews with clinicians, indicating receptivity to the eFI. Over 80% of surveyed nurse navigators, primary care clinicians and surgeons agreed the eFI was feasible, acceptable and appropriate for use in their patient population. Over 75% also agreed that knowing the eFI score for a given patient would influence their clinical decision-making. Based on these data, in fall 2018, there was strong leadership support for implementing the eFI in the EHR. Over the next year, the eFI was built as an external application, leveraging nightly extracts of structured data from the Epic Clarity database and an Epic DataLink connection. The production version of the eFI was mapped to ICD-10 diagnosis codes and went live in October 2019. It is currently calculated for over 200,000 patients 55 years or older within the health system with at least two outpatient visits in the past 2 years.

Contexts for eFI implementation success

Whilst the eFI may be most commonly used as a frailty identification tool linked to a clinical intervention, we have framed the eFI as an intervention for the purposes of the discussion of contexts for eFI implementation success below. One of the most cited implementation frameworks is The Consolidated Framework for Implementation Research (CFIR) [25]. The CFIR aims to predict or explain barriers and facilitators to implementation effectiveness [26]. It can be used to help users make informed choices about which implementation strategies to use in the face of contextual determinants, used to prospectively guide predictions of implementation outcomes or used retrospectively to explain implementation outcomes by assessing differences in determinants across implementation settings [27], the latter being what we will attempt to do in the following sections, focusing on a subset of CFIR domains and constructs.

Innovation

Within the CFIR, constructs relating to the key attributes of an innovation may influence the success of implementation [28, 29]. The perceived innovation source will influence implementation success (Table 1). Since each eFI is homegrown and locally tailored to available EHR data (Table 2), a perception of internal development likely positively influenced implementation. In addition, the eFI’s adaptability takes into account variations in data and clinical populations in each manifestation. Because eFIs are automated, they are simple, easy to use and have low complexity. As a population screening tool that may target further assessment or intervention, an eFI could obviate the need to perform a resource intensive frailty assessment such as CGA on large groups of older adults, or target specific sub-groups for additional assessments or services, thus empowering decision-makers to optimise allocation of scarce resources. This recognition enhances relative advantage, comparing the utility of an intervention with other tools or programs in the same space. In contrast to tailored clinical prediction models, the eFI may be used for a variety of outcomes—it is independent of chronologic age and specific disease states and more focused on overall function and vulnerability. It is trialable, and its cost—in our systems—is negligible to the clinical end user. It saves clinical time that can be redirected to more resource-intensive assessments to identify frailty. Although there are potential time savings, the information technology resources needed, cultivation of frontline user buy-in, and resources required for new interventions targeting frailty all require up-front investment.

Table 1.

Relevant consolidated framework for implementation research domains and constructs

CFIR domain	Construct	NHS	AHWFB	VA
Innovation	Innovation source	Local geriatrics champions within each system. Also implementations begin with local refinement and validation, versus adopting an externally developed tool.
Innovation	Innovation relative advantage	Fully automated and applicable to a broad spectrum of settings, patient populations and outcomes.
Innovation	Innovation evidence base	Large number of publications demonstrating prognostic performance of frailty indices.
		Validation studies for both predictive and convergent validity.	Supported by several retrospective validation studies.	Prior use of a proprietary measure for frailty. VA-FI code publicly available, multiple validation studies.
Innovation	Innovation cost	Algorithm provided to EHR vendors at no cost, under condition of no premium charges to end users.	Generally low. eFI built external to EHR; some ongoing costs for further development and maintenance.	Existing enterprise-wide support for EHR-based phenotyping initiatives.
Innovation	Innovation trialability	Implemented across primary care.	Implemented as an EHR smartphrase for reporting, but not linked to specific best practice advisories or order sets.	Not yet implemented in the EHR, will be added to clinic dashboards in future.
Outer setting	Financing	Initial development supported by National Institute for Health Research funding.	Initial and ongoing development of eFI code base supported by research and innovation funding.	Supported by research programs.
Outer setting	External pressure	Contractual requirement to identify moderate to severe frailty.	Early adopter of value-based care contracts.	Changing demographics of veteran population. Higher prevalence of chronic conditions in veterans that use VA system for care.
Outer setting	Partnership and connections	Publication of joint British Geriatrics Society and Royal College of General Practitioners Frailty Guidelines in 2014.	Participant in age-friendly health system, though only through two acute care for the elderly units.	VA clinical operations and participation in the age-friendly health system.
Inner setting	Information technology infrastructure	Four commercial vendors, many systems in place since the late 1990s.	Epic used for inpatient and outpatient settings since 2013. Switching to new epic instance in 2024.	Used computerised patient record system since the 1980s.
Inner setting	Leadership culture and prioritisation of ageing	Geriatricians serving as the National Clinical Director for Older People.	Large clinical geriatrics section. In 2019, the health system president was an ageing researcher, and the newly founded innovation centre was led by a geriatrician.	Congressionally mandated geriatric research, education and clinical centres.
Implementation process	Assessing needs	Desire for proactive approach to frailty management combined with recognition that current instruments are not feasible for population screening.	Conducted qualitative interviews with clinicians during initial phase of implementation.	No specific needs assessment.
Implementation process	Engaging	Continued research productivity leveraging eFI implementations.
		Qualitative interviews of primary care clinicians around use of eFI.	Grand rounds presentations and ongoing engagement with leadership of population health programs.	Diverse research applications across a range of clinical specialties. Ongoing pilot programs.

AH-WFB, Atrium Health—Wake Forest Baptist; eFI, electronic frailty index; EHR, electronic health record; NHS, National Health Service; VA, Veterans Affairs.

Table 2.

Commonalities and differences across electronic frailty index implementations

Outer setting

External policies and incentives such as the Age-Friendly Health System in the USA [30], Fit for Frailty in the UK [31], and progress towards value-based care in the USA, [32] each align with a need to better identify, characterise and provision care for a population of older adults at higher risk for disability, burdensome acute healthcare utilisation and mortality.

Inner setting

Though each setting differs, a few distinct constructs have supported the implementation of the eFI. First, at each site, information technology infrastructure has matured sufficiently to enable effective, efficient data element mapping and extraction; though each system employs a different EHR and different eFI implementation (Table 1). Recognition that the ageing population represents a high-utilisation, high-need subpopulation has enhanced each environment’s tension for change in seeking scalable solutions. Finally, each setting had geriatricians or ageing-focused clinician-leaders in positions of authority, who influenced the culture, the relative priority and assignment of available resources for adoption.

Implementation process

A driving force across all systems was the realised need for a scalable frailty screening approach.

Both the English NHS and AH-WFB involved multiple key individuals in the implementation process, including health system leadership, opinion leaders, internal implementation leaders and champions across multiple specialties. This engagement is also ongoing as specific care pathways guided by an eFI are tested. Across each system, there is a wealth of ongoing continued research, including large-scale surveillance efforts describing the temporal burden of frailty [33] and the relationship between frailty and neighbourhood disadvantage on acute healthcare utilisation [34].

Clinical eFI applications & use cases

Although there is modest evidence for a range of interventions to improve outcomes for older people with frailty [11], the underpinning evidence is unlikely to have been generated from a trial population selected using an eFI. Clinical applications based on eFIs have largely aligned with the concept of PHM, which is a tiered, integrated framework for designing and delivering interventions across different population groups. It involves the use of data and analytics to better understand the factors that are driving poor outcomes. This enables stratification of the population into tiered levels of risk and providing targeted interventions, often through centralised resources.

English NHS

Working in partnership with NHS England through the National Clinical Director for Older People (John Young and subsequently Martin Vernon—both geriatricians), the national availability of the eFI supported a major change to the GP GMS contract through the inclusion of identification and management of older people living with moderate and severe frailty as a new contractual requirement. This alignment of eFI implementation with a financial incentivisation scheme, in this case the national primary care GMS contract, was a key driver for wider adoption and spread. In the 12 months following national implementation of the eFI and the linked GMS contract change in 2017, NHS England data indicate over 2.5 million older people were assessed for frailty, with about 1 million living with moderate or severe frailty [33]. 25,570 people with frailty were identified as at risk of falling and referred to a falls service, and 210,687 received a medication review to optimise medications, reduce tablet burden and reduce potential side-effects. Following inclusion in the GMS contract, use of the eFI was cited in the 2019 NHS Long Term Plan, which set out the vision, priorities and strategy for NHS care delivery in England over the forthcoming 10 years [35]. The plan specifies supporting people to age well as a key objective, with a greater focus on proactive care based on population health management, including use of the eFI.

VA

The VA-FI has been made available for all investigators and VA operations users through the Centralized Interactive Phenomics Resource [36]. Prior to this, the VA had been using a proprietary measure, the JEN Frailty Index (JenFI), to understand frailty for clinical operations [37]. The JenFI measure has associated costs, and its proprietary nature makes it difficult to interpret and implement. Comparisons of the VA-FI to the JenFI within the VA have generally demonstrated similar performance for long-term institutionalisation and mortality, with the VA-FI performing slightly better [22]. We also considered using a Medicare claims–based FI in VA data [38]. However, this is limited by the need for insurance claims data, which are not available real time, may lag for several years and lack calibration to the VA system. The advantages of having a transparent and free tool, such as the VA-FI, have also been acknowledged. Work is ongoing to implement the VA-FI on primary care clinic dashboards directly within the EHR for clinical use.

AH-WFB

To date, use of the eFI has largely leveraged centralised resources available to ACO populations. The eFI was used to prioritise outreach by nurse navigators to engage vulnerable older adults around chronic disease management during the early stages of the COVID-19 pandemic. The eFI was similarly used to prompt navigator outreach for advanced care planning discussions, which was formally evaluated through a pragmatic trial using a pre-consent ‘Zelen’ randomisation design [39]. Based on these experiences, PHM teams continue to use the eFI as their front-line risk stratification tool for targeting outreach by navigators, social workers and community health workers. There are also several ongoing studies pragmatically testing care pathways in frail older adults identified via the eFI. The first is an ongoing trial also using the Zelen design to test a community health worker program for addressing social determinants of health in frail older adults (clinicaltrials.gov NCT05293730). Another direction is a recently completed a pilot study that leveraged centralised pharmacists to avoid intensive glycaemic control in frail patients [9, 40].

Looking to the future: evidence gaps and future directions

The highlighted early experiences to implement eFIs in three distinct health systems demonstrate how heterogeneity necessarily requires local adaptation and individualisation. There are several critical steps to facilitate eFI use, including the need for local validation of the particular eFI algorithm and EHR presentation, leadership and health system buy-in, technical support, education and clinical guidance. However, contrasting the three health systems in this work, it is clear that sustainable wide-reaching progress is also inevitably tied to financial or regulatory incentives. The English NHS was able to include its eFI within the GMS primary care contract as a requirement for frailty screening, which is the likely reason why progress in that system exceeds that of the other two. Despite movement to consider frailty as part of clinical guidelines, progress in the USA will likely necessitate eFI-based frailty being incorporated as part of metrics for quality of care or as enhancements to payment risk-adjustment models [41, 42].

Another universal need are initiatives to ensure clinicians, especially those outside geriatric medicine, understand what frailty based on an eFI means and have the resources to act upon it. A recent qualitative study of primary care providers in the English NHS highlights several challenges on this front [43]. As most providers don’t receive formal training on frailty, most still conceptualise it more in line with the phenotypic model [1], which, whilst related, will often be discordant with deficit accumulation [44]. There were also concerns about the resource implications of recognising, measuring and managing frailty, as well how and when to discuss the construct with patients. Whilst the British Geriatrics Society has developed a number of educational materials such as the Fit for Frailty Best Practice Guides [31], it is clear that more specific guidance, evidence and support will be needed to make frailty assessment a functional component of primary care.

Given this sobering qualitative study, the clinical application of frailty screening using eFIs raises several, as yet, unanswered questions:

• What level of risk and what populations should be the target of interventions based on an eFI? Much of the initial focus has been on targeting resources to frail older adults. However, there may be utility in targeting individuals indicated to be pre-frail or younger populations with elevated eFI scores as a means of delaying or preventing the progression to frailty. Integrated statistical and economic modelling could help target interventions across the frailty spectrum to maximise clinical, economic and health equity impact.

• What is the ideal interval of frailty assessments? Is frailty best used as a one-time metric of risk or as a dynamic measure of health status/biological ageing over time? The optimal time scale for assessing change with eFIs remains to be determined, though it is likely to be in the order of years. eFIs commonly employ look-back windows over several years and may not purely reflect current status. Additionally, eFIs incorporate deficits like major comorbidities that do not tend to change once developed, or may not be removed from the EHR even if resolved (e.g. anaemia), although this can potentially be addressed by using time constraints for deficits that have potential to improve.

• Does an eFI score in one care setting translate to another? Different settings will lend themselves to different uses of an eFI. A pre-operative clinic may use frailty to help prognosticate and mitigate risk prior to a procedure. Operational needs for PHM may find uses for both a snapshot of frailty to identify high-risk, high-need patients and examining change over time to evaluate the influences of interventions on frailty status. Nevertheless, having a single eFI used across settings within and across organisations can standardise assessments and avoid situations with varying scores across specialist areas.

• What happens when data are missing [45], for example, a patient does not receive regular care in a hospital system, but comes in for emergency or specialty care? Lack of data may misclassify a patient as non-frail, highlighting the need for clinical oversight, interpretation and potentially options for on-demand frailty screening [46].

• Implementing an eFI may lead to yet another label for vulnerable adults, who may not recognise or actively reject the language of frailty. An ethical challenge is to ensure that ‘frailism’ does not merely become another form of ‘ageism’.

• What is the relationship between resilience and frailty? Why may two individuals with the same level of frailty respond differently to the same intervention [47]? As a marker of risk, frailty is a powerful tool that may help guide appropriate therapy. Despite this prognostic utility, trials integrating frailty into their design are sparse, providing limited data to guide drug therapy and other interventions in a generalisable population of frail patients. As an example, a large analysis of routine UK primary care EHR data indicated elevated risks of antihypertensive treatment in adults with moderate or severe frailty, identified using eFI [48]. Conversely, randomised trials have indicated the largest absolute cardiovascular benefit with antihypertensive treatment in more frail adults given their elevated risk of experiencing cardiac events [49, 50], a result that has also been shown with respect to anticoagulation and heart failure–specific treatments [51–53].

• How does frailty change when measured in a dynamic healthcare system? Do PHM initiatives impact frailty?

Conclusion

To meet the needs of the ageing population, it is time for frailty to be integrated in the clinical environment to personalise care in real time. eFIs represent a promising and feasible way to scale frailty across health systems, though it is clear that they will not be sufficient in and of themselves to drive the recognition and management of frailty. Implementation of eFIs require flexibility as these tools are introduced across diverse health systems, although general principles described in the CFIR framework are applicable in all settings. Achieving the goal of eFI-informed care will require trials to demonstrate that the use of eFIs as a risk stratification tool improves the efficiency of care delivery, provides value to individual providers and ultimately improves patient outcomes.

Declaration of Conflicts of Interest:

Wake Forest University Health Sciences has an interest in making electronic screening tools such as the electronic frailty index (eFI), a commercial product in the future so that other hospitals and clinics could use it for their patients. Therefore, N.P. and K.C. and Wake Forest University Health Sciences could financially benefit from future sales of an eFI application. A.O. and J.D. led the development of the VA frailty index, which is freely available. A.C. led the development and national implementation of the eFI in England, which is licenced to suppliers of electronic health record systems and risk stratification software at no cost on the basis that a premium charge is not then subsequently applied to the end National Health Service user.

Declaration of Sources of Funding:

A.O. was supported by grant number R03-AG060169 from the National Institute on Aging (NIA) and by VA CSR&D CDA-2 award IK2-CX001800. N.P. was supported by grant numbers P30AG021332 and U54AG063546 from the NIA. K.C. was supported by NIA grant number K76AG059986. A.C. is part-funded by the National Institute for Health and Care Research (NIHR) Applied Research Collaboration Yorkshire & Humber, the NIHR Leeds Biomedical Research Centre and Health Data Research UK, an initiative funded by UK Research and Innovation Councils, NIHR and the UK devolved administrations and leading medical research charities. The views expressed in this publication are those of the author(s) and not necessarily those of the NIHR, NHS or the UK Department of Health and Social Care. Some of this material is the result of work supported with resources and the use of facilities at the Veterans Affairs Boston Healthcare Centers and the New England Geriatric Research Education and Clinical Center. The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the United States government.