The ability of three different models of frailty to predict all-cause mortality: Results from the European Male Aging Study (EMAS)

Abstract

Few studies have directly compared the ability of the most commonly used models of frailty to predict mortality among community-dwelling individuals. Here, we used a frailty index (FI), frailty phenotype (FP), and FRAIL scale (FS) to predict mortality in the EMAS. Participants were aged 40–79 years (n = 2929) at baseline and 6.6% (n = 193) died over a median 4.3 years of follow-up. The FI was generated from 39 deficits, including self-reported health, morbidities, functional performance and psychological assessments. The FP and FS consisted of five phenotypic criteria and both categorized individuals as robust when they had 0 criteria, prefrail as 1–2 criteria and frail as 3+ criteria. The mean FI increased linearly with age (r² = 0.21) and in Cox regression models adjusted for age, center, smoking and partner status the hazard ratio (HR) for death for each unit increase of the FI was 1.49. Men who were prefrail or frail by either the FP or FS definitions, had a significantly increased risk of death compared to their robust counterparts. Compared to robust men, those who were FP frail at baseline had a HR for death of 3.84, while those who were FS frail had a HR of 3.87. All three frailty models significantly predicted future mortality among community-dwelling, middle-aged and older European men after adjusting for potential confounders. Our data suggest that the choice of frailty model may not be of paramount importance when predicting future risk of death, enabling flexibility in the approach used.

Introduction

Although specific definitions and models of frailty remain contested there is broad agreement that it describes a non-specific state reflecting age-related declines in multiple physiological systems, which in turn lead to an increased risk of adverse outcomes including morbidity, hospitalization, institutionalization and mortality (Rockwood, Mitnitski, Song, Steen, & Skoog, 2006). Frailty may be conceptualized as a state characterizing the broad health of individuals, facilitating risk classification across a wide range of people and conditions (Rockwood & Mitnitski, 2007). While this implies that frailty need not be operationalized as a clinical syndrome, extensive research has focused on identifying those older people with and without a frailty syndrome, defined in terms of specific criteria such as exhaustion, slowness, low or decreased activity, weakness and unintentional weight loss (Fried et al., 2001, Kuh and New Dynamics of Ageing Preparatory, 2007). An alternative approach has been to construct an index of frailty by summing the number of accumulated age-related symptoms or deficits a person has to derive a score to predict future risk of adverse outcomes (Rockwood & Mitnitski, 2007).

Currently, the most commonly used approaches to characterize frailty include the FI of Rockwood and colleagues (Mitnitski et al., 2005, Rockwood and Mitnitski, 2007), the FP described by Fried and colleagues (Fried et al., 2001), and the FS proposed by the International Academy of Nutrition, Health and Aging (IANA) (Abellan van Kan et al., 2008a, Abellan van Kan et al., 2008b), but its utility has not been fully explored. The phenotypic approach has the advantage of being relatively simple to administer, although the relatively restrictive set of criteria may not be applicable to all individuals. Conversely, while frailty indices may offer a broader coverage of deficits than simpler models and also allow identification of high functioning individuals, they are more time consuming in terms of data collection and may be less practical to apply in a clinical setting.

Although index based and phenotypic definitions of frailty have proved useful in predicting a range of deleterious health outcomes (Sternberg, Wershof Schwartz, Karunananthan, Bergman, & Mark Clarfield, 2011), there are few data describing how well the most commonly used frailty models predict mortality in older men living in different regions of the European Union. In this study we utilized the cohort of men participating in the European Male Aging Study to examine and compare the utility of three frailty models adapted from existing index and phenotypic approaches to predict all-cause mortality. A secondary aim was to investigate which of the individual component criteria of our frailty models were associated with mortality.