Peter Hanlon, Silje A Welsh, Nicholas R Evans, Constructing a quality frailty index: you get out what you put in, Age and Ageing, Volume 53, Issue 1, January 2024, afad248, https://doi.org/10.1093/ageing/afad248
Navbar Search Filter Mobile Enter search term Search Navbar Search Filter Enter search term SearchThe number of papers on frailty has risen exponentially over the last decade, with an ever-increasing diversity of researchers investigating the role of frailty in their area of interest. This rise has been matched by heterogeneity in how and when frailty is assessed. Over this period, the frailty index has become a stalwart method for enabling quantification of frailty, where its robust yet flexible design allows a frailty index to be applied in diverse settings and across a range of datasets. However, this variation and flexibility has also meant that the use of the frailty index has varied considerably in the application of the principles set out in Searle et al’s 2008 [ 1] description of the standard approach.
In updated guidance, Theou and colleagues identify 10 steps to guide the creation of a valid and reliable frailty index [ 2]. Using cohort data from the Health and Retirement Study to illustrate its construction and underlying principles, the authors demonstrate the decision-making processes involved in selecting and scoring candidate variables to produce a robust frailty index that may be used as an exposure, predictor, or control variable, as well as an outcome measure. This explicit methodological and analytical guidance is welcome and has the potential to guide greater consistency in the use of the frailty index approach, as well as a means to assess the quality of frailty indices.
This methodological guidance is important and timely, and reinforces that the validity (and indeed value) of any frailty index is dependent on the underlying data from which it is derived. This includes both the quantity of available deficits meeting the criteria for inclusion, and the quality of the data on which they are based. This can lead to challenges in constructing a valid frailty index when the range of available deficits may be relatively narrow (e.g. in trial cohorts), when data is collected for other purposes (e.g. routine healthcare data), or when time and resource are barriers to assessing a wide range of deficits (e.g. in clinical practice).
In the absence of a universally-agreed form, the frailty index has appeared in several iterations, born from divergence across datasets. While the expected standards of a frailty index includes a minimum of 30 multi-domain variables [ 1], a limiting factor to such a comprehensive approach is that it can feel burdensome for practical clinical application. Consequently, several more concise iterations have been produced, such as the 11-item modified frailty index [ 3] used in spinal surgery [ 4], urology [ 5] and vascular surgery [ 6]. Furthermore, in vascular surgery research alone, a total of 20 different frailty indices have been identified, of which 16 had less than 30 variables, yet predictive value persisted [ 6]. While a ‘mini’ frailty index feels more practical for clinical application, and so may be viewed as an opportunity to bridge the gap between research and clinical practice, the obvious challenge in constructing a short-form frailty index for a heterogeneous syndrome will be selecting the most accurate deficits. As a result, the construct validity of such short-form frailty indices may naturally come in to question. Yet, it is recognised that even some well-constructed indices adhering to sound theoretical principles may not always perform equally well in certain clinical contexts [ 7]. This additional guidance will be helpful in reducing some of this variability further. As the frailty concept continues to gain momentum across research and clinical settings, it may be of value to receive guidance on the preferred number of frailty-related domain-deficits to incorporate into future frailty indices. Certainly this would enhance interpretability and comparison, as well as control growing divergence in frailty indices, particularly when considering any application to clinical practice.
The frailty index is increasingly used to assess frailty in secondary analyses of randomised controlled trials [ 8–10]. Trial populations are often highly selected, based on explicit (and often restrictive) inclusion/exclusion criteria [ 11]. This may limit the applicability of such trials to older people living with frailty, who are often under-represented or excluded. This is compounded by the fact that trials rarely quantify frailty explicitly or directly. Applying the frailty index to trial data offers opportunities to evaluate the representativeness of trials, and to explore heterogeneity in the efficacy and safety of treatments across the frailty spectrum. However, quantifying the frailty index in trial data offers specific challenges, for example the data collected in the trial process (including comorbidities, symptom questionnaires, or clinical measurements) are generally selected for their relevance to the trial condition and study outcome, rather than for explicitly measuring a broad range of health measures across multiple systems. This has led to criticism of some applications of the frailty index in trial data, where deficits have largely reflected (for example) cardiovascular deficits in the context of hypertension [ 12]. This point is raised effectively by the authors, who discuss important criteria for standardising this process. In particular, they highlight that using highly-selected populations may alter the distribution of the frailty index or the properties of the deficits themselves (e.g. their relationship with age), and using external data to assess the validity of a given deficit may be advantageous [ 8]. They also highlight the need for deficits to span a wide range of physiological systems, though the point at which a given range of deficits is ‘broad enough’ remains a matter of judgement.
The authors rightly acknowledge that there remain important uncertainties that need to be addressed. In particular, this approach does not account for differences in the duration of a deficit and how to accommodate this in a frailty index. This also has implications for the timing of data collection, particularly in relation to a major life event. Frailty is a dynamic process and the timing of assessment will be key. For example, multiple deficits across domains will likely develop quickly following a stroke, and will likely change over the course of rehabilitation [ 13]. At what time point the frailty index should be collected for prognostication or as a co-variable remains unclear and an area for future work.
The recognition of frailty as a research priority across a range of conditions is likely to result in a renewed focus on measuring frailty explicitly and prospectively in clinical studies [ 14]. As with all research, quality of study design and data collection is critical, and the principles and structured approach to formulating a frailty index outlined in this article are likely to prove invaluable to researchers, particularly those evaluating frailty for the first time.
Clinical Research Training Fellowship from the Medical Research Council (Grant reference: MR/S021949/1) to Peter Hanlon. Stroke Association Senior Clinical Lectureship [SA-SCL-MED-22\100006] and NIHR Cambridge Biomedical Research Centre (NIHR203312) to Nicholas Evan.