Introduction

Over the last decade, there has been exponential growth in the number of ‘FI studies’ published in peer-reviewed journals. The frailty index (FI) represents the accumulated deficit model of frailty [1] and is a continuous variable (ranging from zero to a theoretical maximum of one) derived from a list of potential health deficits [2]. Increasingly, FI scores are being used to assign individuals to frailty categories.

In their 2007 study, Rockwood and colleagues [3] found that an FI = 0.25 was the ‘crossing point’ of robust and frail groups (as measured by the phenotypic model of frailty) and predicted death and institutionalisation. These results were consistent with findings of an earlier study by this group. In 2005, Rockwood et al. [4] showed that the FI and Clinical Frailty Scale (CFS; a scale of increasing functional dependence) were highly correlated and independently predicted adverse outcomes, and that an FI = 0.25 lay between CFS category 4 (‘apparently vulnerable’, mean FI = 0.22) and CFS category 5 (‘mildly frail’, mean FI = 0.27). Together, Rockwood et al.’s studies demonstrated that an FI = 0.25 had construct and predictive validity to categorise community-dwelling older adults as frail or non-frail.

Nevertheless, a variety of FI categories have emerged in the literature. Our study had two key aims: firstly, to explore which FI categories (FI scores and labels) were being used in high impact studies of adults in the community, residential aged care and acute care; and secondly, why these categories were being chosen by study authors.

Methods

Protocol and registration

This systematic scoping review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) criteria [5]. The protocol was registered with the Open Science Framework Registry.

Search strategy

A search of Medline, Cochrane and EMBASE databases was conducted in May 2020 and again in March 2021. Search terms included ‘frailty index’, ‘acute care hospital’, ‘community’ and ‘residential care’. The full search strategy is included in the Appendix.

Eligibility criteria

Studies were eligible for inclusion if they used an FI that met the criteria as set out by Searle and colleagues [2] and the FI was categorised in some way (i.e., an FI score(s) delineated labelled sub-categories). Included studies could be of any design, but were to be conducted in a human adult population in one of three settings: community, acute care or residential aged care. Studies were excluded if they were not an original study (e.g., a protocol or review paper), if only the abstract was available or if there were not written in English.

Study selection

After removing duplicates, one reviewer (IK) independently screened the record titles and abstracts. Two reviewers (IK, NR) independently screened the full-text articles and disagreements were resolved by consensus and discussion with a third reviewer as required. Eligible studies were separated into the three settings of interest. A field-weighted citation impact (FWCI) score was calculated for each study. Sourced from SciVal, the FWCI compares the number of citations a publication receives to the average number of citations received by other similar publications in the Scopus database [6]. Similar publications are those that have the same publication year, publication type and discipline. Consequently, newer publications are not disadvantaged using this methodology. The ten studies with the highest FWCIs (i.e., the 10 ‘highest impact’ studies) from each setting underwent data extraction.

Data extraction and analysis

Three reviewers (IK, NR, EG) performed data extraction and any disagreements were resolved by consensus. Extracted study data included country, publication date, study design and sample size. FI data included mean, FI scores and labels and justification provided by the study author(s) for these FI categories.

Results

Study characteristics

The search strategy yielded 1512 studies and 303 were eligible for inclusion (Fig. 1). Of the 30 highest impact studies (i.e., 10 highest impact studies from each setting), 29 were published in the last decade (Table 1). Twenty-one studies were cohort design and seven were cross-sectional. The majority were conducted in North America. Study sample size ranged from 50 to 931,541. The mean FI of the populations described in the studies ranged from 0.07 to 0.42.

Fig. 1
figure 1

PRISMA diagram of study selection

Full size image
Table 1 Results: Community, acute care and residential aged care settings
Full size table

FI categories

In studies of community-dwelling adults, an FI = 0.25 delineated frail and non-frail individuals in three studies [10, 14, 15], all of which referenced Rockwood and colleagues’ 2007 study [3]. An FI = 0.21 was used in three studies [9, 11, 13]. One referenced Rockwood et al.’s CFS validation study [4] and the other two referenced Hoover and colleagues’ study [12], which demonstrated the predictive validity of this FI cut-off in older community-dwellers. In a large cohort study using the electronic FI (eFI), Clegg et al. [7] used quartiles to define fit (FI < 0.12) versus frail (FI > 0.12) categories. Subsequently, two high impact UK studies adopted these eFI categories for their analyses [16, 17].

In the acute care setting, an FI = 0.25 was the most common score used to determine frailty [18,19,20, 25, 27]. One study referenced Rockwood and colleagues’ community-dweller study [4]. The other studies either provided no justification, referenced studies that did not use FI categories or referenced other papers written by the same authors. Incident adverse outcomes were used to delineate frailty severity (i.e., less or more frail; least frail and least fit) in two studies [26, 28].

In studies of adults residing in residential aged care, there was even greater variability. One study defined frailty as an FI ≥ 0.25 [39] and referenced studies that did not evaluate the validity of this cut-off. Four studies utilised an FI = 0.21 to define frailty [42,43,44] and all referenced (directly or indirectly) the community-dweller study by Hoover et al. [12]. Three studies defined frail as an FI > 0.30 [35, 37, 45]. Two referenced other papers written by the same authors and one referenced a study that demonstrated the predictive validity of similar FI categories in community-dwellers [38].

Across the settings, additional categories such as robust, pre-frail, mildly, moderately and severely frail were defined inconsistently. Methods included examining data spread (such as FI quartiles) [7, 8, 16, 17, 29, 32] and sensitivity/specificity analyses (in relation to adverse outcomes) [26, 28]. Three studies [11, 33, 44], two of which were conducted in residential aged care, adopted the categories that Hoover et al. [12] validated.

Discussion

This scoping review demonstrated variability in FI categorisation in high impact studies of community-dwellers, acute care patients and adults living in a residential aged care. An FI = 0.25 was the most commonly used score to determine frailty, although this was used in less than half of all studies. Greatest variability was seen in residential aged care studies. The rationale for using particular FI categories was reported in most studies, but was not always relevant.

Fourteen studies referenced Rockwood et al. [3, 4] and Hoover et al. [12] as justification for a variety of FI cut-offs and labels. Researchers used the mean FI values reported in Rockwood and colleague’s CFS study [4] to define FI categories, but not all in the same way. While some categories (e.g., frail = FI > 0.21 versus FI > 0.25) were similar, others (e.g., frail = FI > 0.45 versus most frail = FI ≥ 0.45) probably captured different groups of adults. In their 2013 study, Hoover and colleagues [12] tested the predictive validity of published cut-offs (including FI > 0.21 [4], > 0.25 [3] and > 0.35 [38]) in an older community-dwelling population. Using stratum-specific likelihood ratios for hospital-related outcomes, they identified four frailty categories (non-frail = FI < 0.1, pre-frail = 0.1 < FI ≤ 0.21, frail = FI > 0.21 and most frail = FI ≥ 0.45). These categories align with Rockwood et al.’s study [4], where the mean FIs of very fit (CFS 1) and severely frail (CFS 7) adults were 0.09 and 0.43, respectively.

Some FI categories validated in community-dwelling populations have been used in studies of adults in acute and residential aged care. It is debatable whether FI categories should vary by setting. Certainly, in these care settings, a greater proportion of adults are frail and, as a result, dichotomizing the FI into frail and non-frail is suboptimal. For example, in their recent cross-sectional study of Australian aged care residents, Ambagtsheer and colleagues [42] found that using an FI score of 0.21 to delineate frail and non-frail residents yielded a frailty prevalence rate of 43.6%. Thus, the heterogeneity of almost half of the residents’ health statuses would not be captured using this categorisation.

Frailty prevalence rates are also high in the acute setting. For example, Joseph and colleagues [18] found that 44% of geriatric trauma patients were frail (FI > 0.25). In a previous study by our group [46], the negative predictive value for an FI > 0.40 was high (84–98%) for all adverse outcomes, including individual geriatric syndromes, in older inpatients. This study was not included in this scoping review as the authors did not use this FI value to define FI categories (such as FI > 0.4 = more frail or FI < 0.40 = less frail). Nevertheless, two studies included in this review yielded similar results: an FI > 0.46 and an FI > 0.40 predicted adverse outcomes in elderly patients in intensive care and rehabilitation, respectively [26, 28]. These data indicate that an FI ≥ 0.40 is a valid cut-off for severe frailty in the acute care setting. Overall, further data are required to validate mild, moderate and severe categories and to determine whether these categories are applicable across settings.

The major limitation of this scoping review is that data were extracted from 11% of eligible studies. The decision to extract data from the studies with the highest FWCIs was primarily pragmatic. This study not only aimed to describe which FI categories were being used in the literature but also aimed to examine why these categories were being chosen. It was not feasible to extract and present data with this degree of granularity from over 300 studies. Studies with the highest FWCIs are most likely to influence and to have influenced adoption of FI categories in clinical practice and research. Therefore, extracting and synthesising data from these studies generates meaningful results relevant to both spheres. Overall, this methodology yielded highly heterogeneous results and it is unlikely that extracting data from more studies would have resulted in consensus regarding FI categorisation. An additional limitation of this systematic scoping review is that only one reviewer screened titles and abstracts.

In summary, this scoping review demonstrated that high impact studies vary in the way they categorise the FI and while there is some evidence in the community-dweller literature, FI categories have not been well validated in acute and residential aged care. For the time being, the FI should be reported as a continuous variable wherever possible. It is important to continue working towards defining frailty categories - it may be desirable for researchers to recruit only mildly frail community-dwellers for an intervention study or it may be preferable for hospital-based clinicians to provide severely frail patients with an alternative model of care to mildly frail patients. Variable, unvalidated FI categorisation impacts the ability to synthesise results and to translate findings into clinical practice.