Introduction

Frailty is a potentially reversible state of health characterized by decreased physiological reserve and increased vulnerability to adverse outcomes including falls, hospitalization and mortality (1–6). The pooled prevalence estimates of frailty are 10.7% in community settings (7), and 52.3% in nursing home (NH) settings (8). The International Association of Gerontology and Geriatrics has advocated for increased research in NHs (9). Timely identification of residents who are frail or at risk of frailty may assist with early implementation of interventions to prevent adverse outcomes, as well as improve care planning and resource use. There is emerging evidence that residents who are frail are more susceptible to adverse drug events (ADEs) and therefore have a different benefit-to-risk ratio for long-term preventative medications (10–12). Additionally, assessing residents’ frailty status may facilitate timely advance care planning and provision of palliative services to avoid unnecessary hospitalization and exposure to futile interventions at the end of life (13).

Currently, there is no international gold standard for screening or diagnosing frailty in NHs. Despite the plethora of frailty tools available, most have been developed for use in community or clinical settings (14). Fried’s phenotype (FP) (15) and the Frailty Index (FI) (16) remain among the most commonly used diagnostic measures but are challenging to administer in NHs. The FRAIL-NH scale is a 7-item screening tool specifically designed for NHs (17). It includes domains related to potentially reversible variables including: fatigue; resistance; ambulation; incontinence or illness; loss of weight; nutritional approach; and help with dressing (17). Each item is assigned a score of up to 2 points, generating a total score of 0 (non-frail) to 14 (most frail) (17). The FRAIL-NH scale is gaining popularity but no reviews have critically evaluated the literature on the use of the scale. The aim of this systematic review was to investigate the use, validity and adaptations of the FRAIL-NH scale.

Methods

The systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (18). The review protocol was published on PROSPERO (CRD42020185159) (19).

Search Strategy

MEDLINE, EMBASE, CINAHL, and Cochrane Library were searched from January 2015, when the FRAIL-NH scale was first published, to June 2021 (Supplement 1). For the purpose of this review, the term NH was considered synonymous with long-term care facilities and residential aged care services. Where relevant conference abstracts were identified, an author and title search were conducted to locate potential full-text articles. Reference lists of included articles were screened for additional relevant studies. A keyword search was conducted on PubMed and Google Scholar to identify additional studies.

Study Selection

Title and abstract screening were performed by two independent investigators (SJL, LAD), followed by a full-text review. Primary studies using the FRAIL-NH scale were eligible for inclusion. Although FRAIL-NH was developed for use in NHs, studies using FRAIL-NH in other settings including hospitals were not excluded. Non-primary studies (e.g. literature reviews) and studies that did not use the FRAIL-NH scale were excluded. There were no restrictions on study designs and publication language. For non-English studies, the original authors and/or bilingual researchers were contacted to extract data in English.

Data Extraction

Data extraction was performed independently by the same two investigators using a standardized table. Data extracted included study design, country, setting, and resident characteristics. The following FRAIL-NH data were extracted: cut-off points, prevalence, cross-sectional associations, predictive validity, concurrent validity, and adaptations. All study outcomes and variables that assessed for a relationship with FRAIL-NH were recorded. Cross-sectional associations examined the relationship between frailty and other variables of interest in the population. Predictive validity was defined as the ability of FRAIL-NH scores to predict health outcomes. Concurrent validity assessed the performance of FRAIL-NH in comparison to other scales. Cross-cultural adaptations were defined as translations and validations of FRAIL-NH for use in different linguistic, cultural and geographical regions. Authors were contacted in the event of missing data.

Quality Assessment

Quality assessment was conducted independently by the same two investigators using adapted versions of the Joanna Briggs Institute checklists for analytical cross-sectional studies and cohort studies (Tables S1 to S4) (20). The checklists were used to appraise the quality of how frailty was assessed using the FRAIL-NH scale (i.e. not the overall quality of study), irrespective of whether frailty was the primary focus of these studies. FRAIL-NH was deemed to be measured in a valid and reliable way where clear descriptions of FRAIL-NH application, or how FRAIL-NH was computed from other valid datasets were provided. Quality assessment was not applicable to studies that did not investigate cross-sectional associations or longitudinal outcomes according to frailty status. Any disagreements were resolved by discussion with additional investigators (JSB, SL).

Results

A total of 1,350 studies were retrieved, of which, 40 studies applied the FRAIL-NH scale (Figure 1). Studies were published in English (n=37), Chinese (n=1), Japanese (n=1), and Spanish (n=1) (Table 1).

Figure 1
figure 1

PRISMA flow diagram of literature search and study selection

Full size image
Table 1 Characteristics of Included Studies
Full size table

Study and Resident Characteristics

The FRAIL-NH scale was applied in studies across 20 countries; the majority in Australia (n=14) (21–34), followed by China (n=6) (35–40), Spain (n=3) (41–43), United States (US) (n=3) (44–46), Brazil (n=2) (47, 48), Japan (n=2) (49, 50), Singapore (n=2) (51, 52), Belgium (n=1) (53), Canada (n=1) (54), England (n=1) (50), Europe and Israel (n=1) (55), France (n=1) (56), Hong Kong (n=1) (57), Italy (n=1) (58), South Korea (n=1) (59), and Taiwan (n=1) (60). One study was an international multicenter study based on data from the Services and Health for Elderly in Long TERm care (SHELTER) study, conducted in 8 countries including Czech Republic, England, Finland, France, Germany, Israel, Italy, and the Netherlands (55).

Study designs included observational studies (n=31), experimental studies (n=6), qualitative studies (n=2), and a methodological study (n=1) (Table 1). Six articles utilized data from an Australian cohort study (23, 25, 26, 29–31), five from an Australian randomized controlled trial (RCT) (22, 27, 28, 32, 33), three from a Spanish longitudinal study (41–43), and three from a Chinese cross-sectional study (35–37). The number of participants ranged from 13 to 571,139, with a mean age range of 79.0 ± 8.8 to 89.4 ± 4.6 years. Of the 40 studies, 36 studies were conducted in NHs, three studies in hospital settings, and one study reported the scale translation (Table 1).

Frailty Prevalence

Frailty prevalence was reported in 30 of 40 studies, with FRAIL-NH cut-offs ranging from 1.5 to 11 (Table 2). The most frequent cut-off for defining frail and most frail residents were ≥2 (n=21) (21–23, 26–34, 37–41, 52, 56, 58, 60) and ≥6 (n=13) (21, 22, 25, 26, 29, 30, 32–34, 39, 56, 58, 60), respectively. When applying these definitions, between 15.1% and 79.5% of residents were frail, while 28.5% to 75.0% of residents were most frail. Other common cut-offs used to define frailty were ≥6 (n=5) (39, 42, 43, 48, 60) and ≥8 (n=8) (37, 44–47, 54, 55, 58). Ten studies assessed the sensitivity and specificity of different cut-offs (24, 29, 35–40, 47, 52), with ≥8 having the highest sensitivity (94.1%) and specificity (82.8%) for classifying residents as frail based on FI (37), while two studies reported an optimal cut-off of ≥2 using FI and FP, respectively (29, 40). Optimal cut-offs were primarily established based on FI, FP, or the Comprehensive Geriatric Assessment (CGA), and determined using Youden Index or receiver operating characteristic curves. Ten studies described how FRAIL-NH scores were computed from a combination of other validated scales and datasets (Table S5). All but two studies used the original score range of 0–14 and the complete seven domains (57, 59).

Table 2 Characteristics of FRAIL-NH, Frailty Prevalence, and Cross-Sectional Associations with Frailty
Full size table

Cross-Sectional Associations

Cross-sectional associations were described in 18 studies (Table 2). Fifteen studies reported univariate or bivariate associations (23, 26, 29, 31, 33, 39, 43, 45, 47, 52, 56–60); while four studies adjusted analyses for age, sex, and other variables (22, 35, 42, 55). FRAIL-NH scores were associated with older age (35, 39, 47, 55–57), female sex (29, 45, 55–57), multimorbidity (35, 43, 52, 57), cognitive impairment (43, 45, 52, 55, 57, 59), and functional dependence (43, 52).

A longitudinal study of 571,139 residents in 15,551 US NHs found an association between frailty and diabetes, cancer, stroke, heart failure, Parkinson’s disease, and depression (45). Residents experiencing pain, and those receiving anxiolytics or antidepressants at admission were more likely frail (45). Conversely, residents receiving hypnotics and antipsychotics were less likely frail (45). A cohort study of 383 residents in six Australian NHs found residents at high falls risk were more likely to be most frail compared with residents at low falls risk (31). Polypharmacy (42, 55), underprescription (42), preventive medication use (26, 55), and high-dose proton pump inhibitor use (23), were found to have no associations or were inversely associated with frailty.

Predictive Validity

Predictive validity was investigated in 14 studies (Table 3). FRAIL-NH scores were predictive of mortality (25, 30, 39, 44, 47, 52, 56, 57, 59), falls (44, 57), hospitalization (26, 30, 57), length of hospitalization (30, 41), and functional or cognitive decline (41, 52, 57, 58) over a median follow-up of 12 months. Among residents with no activities of daily living (ADL) dependence in a Hong Kong study (n=2,380), FRAIL-NH predicted falls, worsening ADL, hospitalization, and mortality (57). A Chinese study (n=329) reported that each one-score increment in FRAIL-NH increased the hazard ratio for 1-year mortality by 32% (39).

Table 3 Predictive validity of the FRAIL-NH scale
Full size table

An Australian study (n=383) reported that mild-moderately frail residents had higher numbers of hospitalizations and hospital days compared to non-frail residents, whereas most frail residents had lower numbers but were at higher risk of death (30). Over 12 months, more than 20% of most frail residents, but less than 3% of non-frail or vulnerable residents, died at the NH without hospitalization (30). Multiple antihypertensive use was associated with increased mortality among most frail residents in the same cohort (25). Statin use was associated with fall-related hospitalizations in mildmoderate and most frail residents (26). Conversely, among non-users of statins, fall-related hospitalizations were lowest in the frailest subset (26). Five studies reported contrasting outcomes whereby FRAIL-NH scores did not predict falls (53), hospitalization (44, 51), and mortality (41, 53).

Concurrent Validity

The FRAIL-NH scale has been compared to 16 other scales (Table 4). FRAIL-NH was correlated with FI (24, 29, 35, 44), FP (40), and FI-Lab (39). Four studies reported fair to moderate agreements between FRAIL-NH and FI (35, 37), FP (40), and CGA (38). A Chinese study (n=305) reported that FRAIL-NH, FRAIL, FP, Groningen Frailty Indicator, and Tilburg Frailty Indicator demonstrated similarly good diagnostic properties against CGA (38). Of these scales, FRAIL-NH, FRAIL and FP had moderate agreements with CGA at the optimal cut-offs (38). In a Singaporean geriatric hospital (n=210), FRAIL-NH and FRAIL had good diagnostic performance against FI (52). FRAIL-NH had less ceiling effect and had greater discriminatory ability for severe frailty than FRAIL (52).

Table 4 Concurrent validity of the FRAIL-NH scale
Full size table

Seven studies compared the predictive validity between FRAIL-NH and 12 other scales for mortality (30, 39, 41, 44, 52, 53, 59), falls (44, 53), hospitalization (30, 44), length of hospitalization (30, 41, 52), NH admission (52), functional and cognitive decline (41, 52). A US study (n=270) reported that FRAIL-NH was superior to FI at predicting death in frail residents, and falls in pre-frail residents at 6-month follow-up (44). Conversely, FI-Lab was a better predictor of 1-year mortality than FRAIL-NH in a Chinese study (n=329) (39). An Australian study (n=383) found that residents with mild-moderate frailty based on FRAIL-NH and FI had longer hospitalization than non-frail residents (30). In contrast, a Spanish study (n=110) reported that FRAIL-NH and Imputed Fried (IF) had similar predictive validity for shorter hospitalization, functional and cognitive decline (41). Although FRAIL was superior for in-hospital mortality and length of hospitalization in a Singaporean tertiary hospital study, FRAIL-NH better predicted mortality and NH admission up to 12-month follow-up (52).

Cross-Cultural Adaptations

At present, the FRAIL-NH scale has been translated into Brazilian Portuguese (47), Chinese (40), and Japanese (49). The scale was also applied to data from Italy (58), Korea (59), and Spain (41–43), although the translations and adaptations were not specifically reported. The Brazilian Portuguese version was generated through a process of translation by a geriatrician and a physical therapist, cross-cultural adaptation through consensus meetings, and back-translation by a native English speaker (47). The Chinese version was translated by two native bilingual speakers, back-translated by two US researchers, pilot trialed in a sample of NH residents (n=38), and subsequently conducted a post-study interview to ascertain comprehensibility of the translated version (40). Similarly, the Japanese version was translated by two geriatricians and one pharmacist, back-translated by a translation agency with designated expertise on geriatrics and gerontology, followed by a pilot NH trial, and a post-study evaluation (49).

Other Applications

The FRAIL-NH scale was used to investigate the impact of coronavirus disease 2019 (COVID-19) on changes in 152 Italian residents’ frailty status over time (58). COVID-19 accelerated deterioration in frailty by 21%, and COVID-19 survivors had a 4-fold higher chance of developing frailty (58). Frailty was associated with cognitive decline from pre- to post-COVID assessment (58). A large US study (n=571,139) reported that 23% of pre-frail residents at admission improved to robust by three months, whereas 30.5% of pre-frail residents transitioned to frail (45). At admission, residents with severe cognitive impairment were 74% more likely to be frail (45). By 6 months, those with severe impairment were over twice as likely to be frail (45). Other applications of FRAIL-NH include: establishing baseline comparability of participants in studies (24, 27, 28, 32, 33, 54); baseline screening as proxy to physical health for RCT recruitment (50); describing COVID-19 seroprevalence in frail residents (48); characterizing frailty status in residents interviewed on perceptions of frailty (21, 34); examining changes in FRAIL-NH scores at annual routine screening (46); and the prospective use of FRAIL-NH in acute care (52).

Methodological Quality of Studies

Of the 18 cross-sectional studies, 15 studies provided clear inclusion criteria (Table S3) (22–24, 29, 31, 33, 35–38, 40, 42, 43, 45, 54). All studies described resident characteristics and settings in detail (22–24, 29, 31, 33, 35–38, 40, 42, 43, 45, 48, 54, 55, 60). Cross-sectional associations with frailty were not investigated in seven studies (24, 36–38, 40, 48, 54), therefore criteria related to confounding factors were not assessed in these studies. Residents with different frailty status were recruited from the same population, free of outcomes at the start of study, and follow-up time was reported in all 14 outcome studies (Table S4) (25, 26, 30, 39, 41, 44, 47, 51–53, 56–59). Confounding factors were identified and adjusted for in 13 outcome studies (25, 26, 30, 39, 44, 47, 51–53, 56–59). Overall, two studies did not provide adequate information on how FRAIL-NH was computed, which precluded assessment on whether the measurement was valid (48, 53).

Discussion

This was the first review of the use, validity and adaptations of the FRAIL-NH scale. FRAIL-NH has become widely used in Australia, Asia, Europe and North America. Frailty prevalence varied depending on the cut-off used. The optimal cut-off is dependent on the FRAIL-NH domains included, reference scale used, purpose of screening, setting, resident characteristics, structure and resources available. While a universal cut-off permits direct comparison across studies, it is not clear that optimal cut-offs are generalizable to all NHs. Although FI was commonly used as the reference for determining the optimal FRAIL-NH cut-off, FI scores were originally continuous instead of categorical, and the commonly applied FI cut-off of ≥0.22 was based on studies analyzing different deficits in community instead of NH settings (61). The wide range in proportion of frail residents likely reflects differences in NH services in different countries. Lower cut-offs with higher sensitivity could improve frailty detection, whereas higher cut-offs with higher specificity could reduce misdiagnosis and enhance resource allocation (38, 52).

FRAIL-NH was associated with resident and clinical characteristics indicative of higher care needs. One study included in the review suggested frail residents had the most complex medication regimens (22). This is important because frail residents are susceptible to ADEs, and regimen complexity has been linked to medication errors and hospitalizations (62–65). Frailty may also be overlooked in residents with chronic multimorbidity in conventional disease-based treatment decision-making (11, 62, 66). Use of a straightforward screening tool such as FRAIL-NH may assist in rapid identification of frailty to target clinical services (e.g. regimen simplification, medication reviews, CGA) to residents who may benefit most. Given that most NH residents have some degree of frailty, determining different levels of frailty (e.g. pre-frail, most frail) may allow assessment of residents most at risk of adverse outcomes for better care planning and resource allocation.

In addition, the FRAIL-NH scale may be useful for identifying residents at risk of various adverse outcomes such as hospitalization, functional decline, and ADEs. Among the five studies investigating all-cause hospitalization (26, 30, 44, 51, 57), three studies reported an association and two did not (44, 51), while another study reported FRAIL-NH scores further predicted fall-related hospitalization (26). Discrepancies in predictive validity may reflect different thresholds for hospitalization and differences in the extent to which clinical in-reach services are provided in NHs. Further studies on cause-specific hospitalizations are warranted to determine the reasons why frail residents are hospitalized and to develop preventive strategies. FRAIL-NH could guide development of individualized care plans to prevent falls, hospitalization and mortality. It is possible that the frailest residents may have advance care plans with do not hospitalize orders (30).

FRAIL-NH scores predicted functional decline in all three studies that examined this parameter (41, 52, 57). Frail residents with no baseline ADL dependence were at increased risk of worsening ADLs in a 9-year longitudinal study (57). Functional independence is an important outcome measure that has been prioritized by residents and carers (62). Using FRAIL-NH to detect pre-frail residents may help direct interventions to prevent functional dependence. FRAIL-NH can be applied on a regular or as needed basis to ensure care goals are updated in alignment with shifts in frailty transition. Two studies demonstrated that frailty modified the risk-benefit ratio of preventive medications (25, 26). Statin users with mild-moderate frailty or who were most frail were at higher risk of fall-related hospitalization (26), whereas residents who were most frail and using multiple antihypertensives were at increased risk of mortality (25). FRAIL-NH could be used to identify residents at increased risk of ADEs and assist with tailoring medication regimens accordingly.

Overall, the FRAIL-NH scale demonstrated good agreement with other well-established but more complex frailty scales. FRAIL-NH had similar diagnostic and predictive properties as FI, FI-Lab, FP, IF, and FRAIL. Unlike FP, FRAIL-NH does not require use of specific instruments (e.g. dynamometer to measure handgrip strength) and measurement of gait speed which can be challenging in NHs. As many residents have functional dependence, assessing walking speed may be impractical. Although FI may be more comprehensive for care planning, it is resource intensive to manually code residents’ deficits in the absence of electronic software. In contrast, FRAIL-NH is a simple and rapid scale that utilizes routinely collected data in NHs. The ease of administration makes FRAIL-NH a good candidate as a routine screening tool in NHs. Specialist training is not required to administer FRAIL-NH. This is supported by 24 studies reporting that care workers, technicians, or nurses effectively applied the scale (21–37, 39, 41–43, 49, 56, 59). Furthermore, FRAIL-NH can often be retrospectively applied to existing datasets as demonstrated by nine studies in this review (44, 45, 51–53, 55–57, 59). This is feasible because most items in FRAIL-NH can be adapted from other validated scales that are routinely applied in NHs (Table S5).

Strengths, Limitations and Implications for Future Research

A key strength is the inclusion of all full-text primary studies from 20 countries that have used the FRAIL-NH scale from inception to present irrespective of study designs and publication language. The review provides a snapshot of the use of FRAIL-NH globally, excluding results from a Japanese abstract reporting that FRAIL-NH predicted NH admissions (67), and a Mexican abstract using the Spanish version of FRAIL-NH (68). To our knowledge, full-text articles for these abstracts have not been published. Extensive variability in study samples and FRAIL-NH cut-offs limited comparability of included studies. Findings from individual studies are not necessarily representative or generalizable to all frail NH residents in the corresponding countries. Cross-sectional study designs limit the prospect of exploring causal relationships. A meta-analysis was not performed due to heterogeneity in study objectives and outcomes investigated.

None of the RCTs in this review investigated frailty as the primary outcome. However, an emerging application of FRAIL-NH may be to monitor changes in frailty status over time as an outcome in RCTs and cohort studies. This highlights the dynamic state of frailty and assists with identifying risk factors influencing shifts in frailty transitions. Future studies are needed to explore whether targeted interventions can delay or reverse frailty transitions. Similarly, FRAIL-NH may be used to describe frailty status of RCT participants. This is important because frail older people are often under-represented in RCTs (12, 62). FRAIL-NH can facilitate by establishing frailty profiles and examining potential differences in characteristics between intervention versus control groups. Since poor health status among frail residents may also contribute to lower rates of follow-up, another potential use of FRAIL-NH is to model characteristics of residents who drop-out of prospective intervention studies.

Given that the FRAIL-NH scale is predictive of hospitalization, mortality, and other important clinical outcomes, it may be useful as a routine screening tool in NHs. According to the World Health Organization’s guide on screening programs (69), FRAIL-NH fulfills the criteria for a clinically meaningful screening method to reduce the incidence and severity of a condition by early detection and treatment. Mass population screening of frailty may not be the most cost-effective approach and requires further research to establish its feasibility and effectiveness at improving outcomes (66). However, frailty screening in adults ≥70 years when accessing health care is recommended by major international frailty consensus groups across Europe, North America and Asia-Pacific (1, 70, 71).

The New Zealand Frailty Care Guides recommended the use of the FRAIL-NH scale to assess gradual deterioration in residents (72). Future studies should explore the potential integration of FRAIL-NH screening into routine care planning and whether this translates to improved outcomes for residents and NHs. The Australian National Aged Care Classification describes frailty as a factor driving NH resource use (73). Consideration of frailty status can assist clinicians and NH providers to identify residents at risk of harm, and implement strategies to prevent the onset and development of frailty. FRAIL-NH may have a potential role in identifying and accounting for care burden associated with frailty to better estimate funding and staff allocation.

Conclusion

In seven years, the FRAIL-NH scale has been applied in 20 countries and adapted into three languages. Despite being applied with a range of cut-offs, FRAIL-NH was associated with various factors indicative of higher care needs. FRAIL-NH demonstrated good agreement and had similar predictive properties as other well-established but more complex frailty scales. FRAIL-NH was predictive of falls, cognitive or functional decline, hospitalization, and mortality across different settings. The association with adverse health outcomes highlights its value in guiding care for frail residents in NHs.