Skip to main content
PMC home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2009 Aug 20.
Published in final edited form as: Exp Aging Res. 2009 Jan–Mar;35(1):61–82. doi: 10.1080/03610730802545051

Change in Frailty and Risk of Death in Older Persons

A S Buchman 1,2, R S Wilson 1,2,3, J L Bienias 4,5, D A Bennett 1,2
PMCID: PMC2729435  NIHMSID: NIHMS128459  PMID: 19173102

Abstract

We developed and validated a continuous composite measure of frailty and examined its rate of change in 832 older persons with annual evaluations for up to 8 years. In generalized estimating equation models adjusted for age, sex, and education, there was a significant increase in frailty during follow-up. In a proportional hazards model controlling for age, sex, education and baseline frailty, each 1-unit increase in annual change in frailty was associated with an almost 5 times the risk of mortality. Using a continuous measure, we document that frailty is progressive in some older persons and that its rate of progression is associated with mortality.

Keywords: Frailty, Sarcopenia, Aging, Mortality

INTRODUCTION

Physical frailty is a heterogeneous condition characterized by impaired strength, gait and balance, body composition, and fatigue (Ferrucci et al., 2004). Frailty is thought to represent an age-related reduction in reserve and resistance to stressors leading to increased clinical vulnerability and adverse health outcomes. Cross sectional studies suggest that about 7% of persons older than 65 years are frail, and that the occurrence of frailty increases with age and may exceed 45% after age 85. Frailty is a common condition in elders, even after controlling for common chronic diseases, and is increasingly recognized as a major public health problem (Cohen, Harris, & Pieper, 2003; Ferrucci et al., 2004; Fried et al., 2001; Rockwood et al., 2004; Walston et al., 2002).

Evidence on the nature of the biological processes thought to contribute to frailty is limited, probably because of their complex and multifactorial nature (Ferrucci et al., 2004;Fried et al., 2001). In the absence of direct measures of these complex processes of dysregulation, investigators have used operational definitions of frailty to develop categorical measures that can identify older persons vulnerable to adverse health-related outcomes (Ferrucci et al., 2004; Fried et al., 2001; Ostir, Ottenbacher, & Markides, 2004; Puts, Lips, & Deeg, 2005; Rockwood et al., 2004). Categorical measures are useful to classify an individual person (e.g. frail, prefrail or at risk), but may not be the most advantageous approach for documenting the rate of change in frailty over time (Gill, Gahbauer, Allore, & Han, 2006; Ostir et al., 2004; Puts et al., 2005; Studenski et al., 2004). Furthermore, since frailty is a multidimensional construct, it is possible that its operational components change at different rates which may be difficult to capture with categorical measures. An alternative approach, employed effectively in other areas of aging research (Louis, Tang, Schupf, & Mayeux, 2005; Onder et al., 2002; Petersen et al., 2005; Schalk, Visser, Deeg, & Bouter, 2004; Schupf et al., 2005), has recently been applied to frailty and conceptualizes the operational components of frailty as continua which may be summarized into a continuous composite measure (Buchman, Boyle, Wilson, Tang, & Bennett, in press). A continuous composite measure of frailty may be a potent tool for examining the rate of change in frailty over time with greater precision and provide increased power to identify risk factors and the adverse health consequences associated with frailty.

Based previous operational definitions of frailty (Ferrucci et al., 2004; Fried et al., 2001), we constructed a continuous composite measure of frailty based on the four components: strength, gait, body composition, and fatigue. This composite measure of frailty has been reported to be associated with risk of AD and with the level and the rate of cognitive decline in the elderly. (Buchman et al., in press). The purpose of the current study was to provide further evidence demonstrating that this continuous composite measure of frailty complements its antecedent categorical measure of frailty and would be useful in longitudinal aging research. First we validated composite frailty by showing that baseline frailty predicted death and incident disability and was correlated with a categorical frailty measure used by other investigators. Next we used this composite measure of frailty to examine the rate change in frailty over up to eight years of follow-up, as well as the relation of this change to age, sex and education and whether change in frailty was affected by chronic disorders or use of medications. Then we examined the clinical consequences of documented change in frailty by examining if the rate of change in frailty was associated with risk of death. Since frailty is conceptualized as a multidimensional construct, we also analyzed change in each of the four components used to construct composite frailty. Finally, we considered a number of alternate strategies for summarizing frailty to determine the robustness of this approach.

METHODS

Participants

All subjects were participants in the Rush Memory and Aging Project, a longitudinal study of aging (Bennett et al., 2005; Wilson, Barnes, & Bennett, 2003; Wilson et al., 2005).The study was conducted in accordance with the latest version of the Declaration of Helsinki, and it was approved by the Institutional Review Board of Rush University Medical Center. Participants underwent a uniform annual structured clinical evaluation, which included a medical history, neurological examination, review of medications, and assessment of cognitive function (Bennett et al., 2005). One thousand and ninety participants completed their baseline evaluation at the time of these analyses. Eligibility for these analyses required the absence of clinical dementia based on the baseline clinical evaluation with a valid baseline and at least one follow-up composite frailty score since this study was designed to document change in frailty. We excluded 78 persons who met standard criteria for dementia at baseline (McKhann et al., 1984), leaving 1012. We excluded those participants without valid follow-up data including: 24 persons who died before their first follow-up examination, 120 who have not been in the study long enough for their first follow-up examination and 36 with missing follow-up data, leaving 832 (95.9% participation rate). Among the 832 persons with one or more follow-up evaluations, mean 3.7 (SD = 1.9; range 2 to 9); more than 75% had three or more evaluations; and valid frailty measures were missing in 127 of 3241 visits (3.9%) over the course of the study. The participants had a mean age of 80.4 years (SD = 6.9), a mean of 14.5 years of education (SD = 3.0); a mean Mini-Mental State Examination score of 27.9 (SD = 2.1); 74.4% were women and 91.3% were white and non-Hispanic. The mean duration of follow-up was 2.9 years (SD = 1.9).

Composite Frailty Measure

Composite measures have been widely used in longitudinal studies (Louis et al., 2005; Onder et al., 2002; Petersen et al., 2005; Schalk et al., 2004; Schupf et al., 2005; Wilson, Schneider, Bienias, Evans, & Bennett, 2003). They reduce random error, minimize floor and ceiling effects, and tend to be normally distributed. The primary measure of composite frailty used in this study has been used previously (Wilson et al., 2006; Buchman et al., in press) and was based on four components endorsed by previous investigators, including grip strength, timed walk, body composition and fatigue (Blaum, Xue, Michelon, Semba, & Fried, 2005; Boyd, Xue, Simpson, Guralnik, & Fried, 2005; Ferrucci et al., 2004; Fried et al., 2001; Onder et al., 2002; Ostir et al., 2004; Schalk, et al., 2004; Wilson et al., 2006). Strength was based on grip strength measured with the Jamar hydraulic hand dynamometer (Lafayette Instruments, Lafayette, IN). Two trials of grip strength were obtained for each hand. The four trials were averaged to yield strength (Buchman, Wilson, Bienias, & Bennett, 2005). Gait was based on the time to walk eight feet. Although most individuals in our cohort initially were able to walk, during follow-up some participants were no longer able to ambulate but still had other valid frailty components. Therefore, by using a six point scale from 0–6 and assigning a 0 to those unable to walk, we were able to include all participants so as to not lose data for the longitudinal analyses. During the course of the study, there were 128 / 3241 exams (3.9%) in which a participant was unable to complete the walk and received a score of zero. The data from the rest of the cohort who were able to walk and had walk times was converted to quintiles, with a score of 5 assigned to the fastest times and scores of 1 to the slowest times (Wilson et al., 2006). Body composition was based on body mass index (BMI). We used two questions from a modified version of the Center for Epidemiologic Studies-Depression (CES-D) Scale to assess fatigue (Radloff, 1977).

The four components used to construct a composite measure of frailty were structured so that higher values would indicate poorer performance (higher frailty) and lower values would indicate better performance (less frailty). For example, rather than larger grip value indicating a higher level strength, grip strength was multiplied by −1 so that larger values reflected less strength. The composite measure of frailty was constructed by converting the raw score from each of the four component measures to z scores using the mean and standard deviation from all participants at baseline. Z transforming requires multiplying all values by a constant value which can be obtained on the entire cohort or on a subgroup, In this particular cohort with an average age of >80 years it would be difficult to define what would constitute a healthy subgroup, instead we used the whole cohort at baseline. Therefore, although including impaired subjects at baseline will affect the absolute value, it should have minimal if any affect on the ranking of individuals since z scoring is the same as multiplying everyone by a constant. Pearson (r) or Spearman (rho) correlations are reported for the associations between the four components. The z scores of grip strength showed modest correlations with gait (r=0.30, p <.001), fatigue (rho=0.146, p< 0.001), and body composition (r=0.078, p = .051). Gait was associated with fatigue (rho=0.213, p <.001) and body composition (r=0.153, p < 0.001). Fatigue and body composition were modestly inversely correlated (rho=0.049, p=0.049). The, z scores for grip strength, gait, body composition and fatigue, described above were averaged to yield a composite measure of frailty.

Categorical Frailty Measure

Another way to validate the composite measure frailty is to examine its association with its antecedant categorical measure of frailty using data from the current cohort (Fried et al., 2001; Ostir et al., 2004; Bandeen-Roche et al., 2006). We dichotomized each of the five components used to construct categorical frailty (Fried et al., 2001). The lowest quintile of grip, gait, BMI and physical activity were defined as frail and any reports of fatigue were considered consistent with frailty. Due to level differences between men and women on performance measures, sex-specific quintiles were used for grip, gait and physical activity. Fried’s categorical measure of frailty was trichotomous with participants with 3 or more frail components considered frail, those with 1 or 2 frail components were prefrail and those with no frail components were considered not frail (Fried et al., 2001).

Other Outcomes and Covariates

Gender and race were recorded at the baseline interview. Race questions and categories were those used by the 1990 U.S. Census. Age in years was computed from self-reported date of birth and date of the clinical examination at which the strength measure was collected. Reported years of education were obtained at the time of the baseline cognitive testing. Participation in the Rush Memory and Aging Project includes agreeing to autopsy at the time of death. At the time of these analyses, 100% of the mortality data were complete and up to date.

Disability at the time of the interview was assessed using two standard disability measures which capture different aspects of daily functions. We included six items from the Katz Activities of Daily Living (ADL) Scale (Katz & Akpom, 1976): walking across a small room, bathing, dressing, eating, transferring, and toileting. A composite measure was created by summing the non-missing items with 1 = can do and 0 = cannot do [range 0 to 6] (Bennett et al., 2005).

In order to assess the influence of vascular risk factors and vascular disease burden on change in frailty, we summarized cumulative vascular risk factors and vascular disease burden (Boyle et al., 2005). We computed summary scores indicating each individual’s vascular risk factors (i.e. the sum of hypertension (51.9%), diabetes mellitus(11.9%), and smoking (39.5%), resulting in a score from 0–3 for each individual and vascular disease burden (i.e. the sum of heart attack (12.5%), congestive heart failure (4.5%), claudication (6.3%), and stroke (10.1%), resulting in a score from 0–4 for each individual. These summary scores were used as covariates in the analyses. Finally, we also controlled for psychotropic and antihypertensive medications which can affect mood and motor function, which are both components of frailty. A history of psychotropic and antihypertensive medications was based on participant report at the time of interview (Bennett et al., 2005). Physical activity was assessed using questions adapted from the 1985 National Health Interview Survey. Activities included walking for exercise, gardening or yardwork, calisthenics or general exercise, bicycle riding, and swimming or water exercise (McPhillips, Pellettera, Barrett-Connor, Wingard, & Criqui, 1989). Participants were asked if they had engaged in any of those activities within the past 2 weeks and, if so, the number of occasions and average minutes per occasion. Minutes spent engaged in each activity were summed and expressed as hours of activity per week, as previously described (Buchman, Boyle, Wilson, Bienias, & Bennett, 2007).

Statistical Analyses

Pearson correlations were used to examine the associations between frailty, age and education. T-tests were used to compare men and women. Spearman correlations were used to examine the associations between frailty and categorical frailty. We used Cox proportional hazards models controlling for age, sex and education to examine the relation of composite and categorical frailty with mortality and incident disability. Generalized linear models fit using the method of generalized estimating equations (GEE) were used to summarize cross-sectional and longitudinal information about level of, and change in, frailty and its four components (Liang & Zeger, 1986; Zeger, Liang, & Albert, 1988). The models included a term for study time, measured as time in years from the study baseline. There were also terms for age (centered at the approximate mean of 80), sex (1 for male, 0 for female) and education (centered at the mean of 14) at baseline as well as terms for the interaction of each of these measures with study time. An identity link was used for the composite measure, grip, gait and BMI as they were all approximately normally distributed. A logit link was used for the 3 levels of fatigue (treating the measure of fatigue as “proportion fatigued” in an “events/trials” setting). After initial analyses of composite frailty, we added additional terms for other covariates as well as their interactions with time to see if they affected change in frailty. To investigate the clinical significance of change in frailty, we used a Cox proportional hazards models controlling for age, sex, education as well as baseline frailty to examine the relation of annual change in frailty with risk of death. For this analysis we used ordinary least squares regression to estimate the annual rate of change in frailty for each person. Finally we used identical generalized estimating equations to assess change in each of the four components of composite frailty. All models were examined graphically and analytically and assumptions were judged to be adequately met. Programming was done in SAS® (SAS Institue Inc, 2000).

RESULTS

Baseline Composite Frailty and Demographic Variables

The composite frailty measure showed an approximately normal distribution and ranged from −1.73 to 1.92 [10% percentile = −0.74, 90% percentile = 0.68; (mean= −0.03; SD = 0.56)] with higher scores indicating increased frailty. The relation of frailty to selected demographic and clinical measures is shown in Table 1. Frailty was positively related to age (r = 0.33, p<0.001) but as shown in Figure 2, while on average the level of composite frailty was significantly higher in older than younger persons, there was substantial heterogeneity across the ages. Frailty was inversely related to education (r = −0.16, p<.001), and men were less frail than women [men: −0.32 (SD=0.55) vs women: 0.06 (SD=0.54); t = 8.84 [830], p< 0.001]. Thus the associations of composite frailty with demographic variables were similar to those previously reported with categorical frailty (Fried et al., 2001).

Table 1.

Baseline Characteristics of Memory and Aging Participants*

Best Quintile Frailty Intermediate Quintile Worse Quintile
Age (years) 79.3 (6.5) 80.8 (5.8) 82.8 (7.0)
Women, % 52.1% 81.2% 84.9%
White, non Hispanic, % 88.5% 92.1% 90.9%
Education (years) 15.2(3.3) 14.2 (2.9) 14.0 (2.9)
MMSE score 28.5 (1.7) 28.0 (2.0) 27.2 (2.3)
Vascular Disease Burden 0.3 (0.5) 0.4 (0.7) 0.4 (0.6)
Vascular Risk Factors 1.2 (0.8) 1.2 (0.9) 1.2 (0.8)
Psychotropic Medication Use 4.9% 5.8% 9.2%
Hypertensive Medication 9.6% 9.1% 10.2%
Composite Frailty −0.80 (0.25) −0.06(0.07) 0.78 (0.30)
  Strength (lbs.) 69.1 (18.95) 45 2 (11.41) 34.4 (10.86)
  Gait (quintiles) 4.2 (0.99) 3.0 (1.30) 1.7 (1.06)
  Body Composition (BMI) 30.9 (6.60) 26.7 (4.73) 25.1 (4.25)
  Fatigue 2.0 (0.11) 1.9 (0.20) 1.5 (0.39)
0 Frail Components 92.1% 60.0% 0.6%
1 Frail Components 6.7% 37.6% 37.0%
2 Frail Components 1.2% 2.4% 51.0%
3 Frail Components – – 11.5%
Open in a new tab
*

The entire cohort was divided into quintiles based on composite frailty at baseline. The composite measure is constructed so that a more positive value is consistent with more frailty i.e poorer performance and a negative value is consistent with less frailty i.e. better performance. Therefore the fifth quintile has the lowest performance and the first quintile the best performance.

Figure 2. Composite Frailty and Age.

Figure 2

Open in a new tab

This figure shows the relationship between baseline composite frailty (vertical axis) and age (horizontal axis) for all participants.

Baseline Composite and Categorical Frailty

Another way to validate composite frailty is to examine its association with its antecedent categorical frailty measure. Categorical frailty was constructed by dichotomizing its individual component measures (Fried et al., 2001). As shown in Figure 1 increasing composite frailty is associated with an increasing number of frail components (rho = 0.72, p<0.001). Thus the participants with the worse quintile of baseline composite frailty scores had a higher number of frail components (Table 1). Composite frailty was associated with its antecedent categorical frailty measure (rho = 0.44, p<0.001).

Figure 1. Composite Frailty and Dichotomous Frail Components.

Figure 1

Open in a new tab

Composite frailty was highly associated the number of frail components (rho = 0.72, p<0.001).

Baseline Frailty and Adverse Health Consequences

Baseline Frailty and Mortality

Previous constructs of frailty have been associated with mortality. In a proportional hazards model controlling for age, sex, education and baseline frailty, we examined the relation of baseline frailty to risk of death. During the course of the study 106 participants died (12.7%). In a proportional hazards model controlling for age, gender and education, risk of death increased by about 85% for each 1-unit increase in composite frailty score at baseline (hazard ratio: 1.84; 95% CI: 1.28, 2.66). Similarly using the categorical frailty measure, a participant categorized as frail at baseline had an 85% increased risk of developing disability (hazard ratio: 1.85 95% CI: 1.25, 2.76).

Baseline Frailty and Incident Disability

Next to validate composite frailty, we examined the relation of composite frailty to incident disability using measures of basic activities of daily living.

Katz ADL Disability

Of 731 persons without disability on the Katz scale at baseline, 181 developed incident disability. In a proportional hazards model controlling for age, gender and education, risk of disability on the Katz scale more than doubled for each 1-unit increase in composite frailty score at baseline (hazard ratio: 2.10; 95% CI: 1.56, 2.81). Similarly using the categorical frailty measure , a participant categorized as frail at baseline had about a 70% increased risk of developing disability (hazard ratio: 1.69 95% CI: 1.26, 2.26).

IADL Disability

Of 411 persons who were independent in all instrumental activities of daily living at baseline, 225 became dependent in one or more items. In a proportional hazards model controlling for the confounding age, gender and education, each one-unit increase in composite frailty at baseline was associated with a more than 75% increase in the risk of developing disability in instrumental activities of daily living (hazard ratio: 1.76; 95% CI: 1.30, 2.40). Similarly using the categorical frailty measure, a participant categorized as frail at baseline had a 40% increased risk of developing disability (hazard ratio: 1.37 95% CI: 1.06, 1.78).

Documenting the Rate of Change in Composite Frailty

In the previous analyses we validated composite frailty by showing the association of baseline composite frailty with demographic variables, risk of death, incident disability (both IADL and ADL), as well as its correlation with its antecedent categorical frailty measure. While previous studies have examined the change from being categorized as not frail to frail over time, categorical measures do not lend themselves to documenting the rate of change in frailty. A generalized linear model fit using the approach of generalized estimating equations (GEE) was used to summarize the cross-sectional and longitudinal associations of frailty with age. On average, for each additional year of age at baseline, frailty was nearly 0.03 point higher (Table 2), Further, composite frailty increased by about a tenth of a unit per year during follow-up (Time, Table 2). Overall, about 2/3 of the cohort show increased frailty over the course of this study and the other third showed no change or slight improvement. The longitudinal effect of aging on the progression of frailty (estimated effect of Time = 0.08) is more than 2.5 times larger than what would be expected based solely on the cross-sectional effect at baseline (estimated effect of Age: 0.03, Table 2). Figure 3 illustrates both the cross-sectional and longitudinal effects of age. The dotted line shows the strong cross-sectional effect of older age at baseline on frailty at baseline, plotted for a person with average values of the other covariates. Contrast the cross-sectional effect of age, the dotted line, with the longitudinal effect, as indicated by the three solid lines, where each solid line shows the trajectory of change in frailty over time for a person who was 65, 75, or 85 years old, respectively, at baseline. The rate of increase shown in the solid lines is steeper as compared to the dotted line, which illustrates the strong effect of Time in the model. There was no significant difference in the rate of change in frailty during follow-up in men compared to women (Sex*Time, Table 2). Education did not modify the annual rate of change in frailty (Education *Time, Table 2).

Table 2.

Change in Frailty and Its Subcomponents and Their Association with Baseline Demographic Measures [Estimate (std. error, p-value)]*

Term Frailty Grip Strength Gait Fatigue Body Composition
Time 0.077(0.008, p<0.001) 0.089(0.008, p<0.001) 0.128 (0.013, p<0.001) 0.061 (0.022, p=0.006) 0.020 (0.007, p=0.006)
Age 0.028 (0.0031, p<0.001) 0.046 (0.004, p<0.001) 0.032 (0.005, p<0.001) 0.020 (0.009, p=0.031) 0.030 (0.006, p<0.001)
Age*Time 0.002(0.001, p=0.069) 0.002 (0.001, p=.009) 0.004 (0.002, p=0.027) −0.003 (0.004, p=0.461) 0.001 (0.001, p=0.207)
Sex −0.431 (0.040, p<0.001) −1.604(0.072, p<0.001) −0.170 (0.072, p=0.019) −0.436(0.149, p=0.004) 0.026(0.067, p=0.695)
Sex*Time 0.013 (0.014, p=0.357) 0.071 (0.014, p<0.001) 0.019 (0.024, p=0.419) 0.038 (0.046, p=0.414) 0.003 (0.012, p=0.821)
Education −0.021 (0.006, p<0.001) −0.028 (0.008,p=0.001) −0.039 (0.010, p<0.001) −0.109 (0.022, p<0.001) 0.026 (0.012, p=0.037)
Ed*Time −0.001(0.002, p=0.779) 0.002 (0.002,p=0.298) −0.001 (0.004, p=0.835) 0.008 (0.006, p=0.186) −0.001 (0.002, p=0.791)
Open in a new tab
*

Generalized Estimating Equation Models with age (centered at 80 years), gender (male = 1, female =0) and education (centered at 14 years). An identity link was used for composite frailty, grip strength, gait and body composition measures; a logit link was used for fatigue measure. Estimates are the coefficients from the linear predictor, i.e., they are not odds ratios for fatigue. All components are coded such that higher values represent worse function (i.e., higher values mean more frailty, less grip strength, slower gait, more fatigue, and lower body composition).

Figure 3. Cross Sectional Versus Longitudinal Effects Of Age And Composite Frailty.

Figure 3

Open in a new tab

This figure shows the predicted cross sectional effect (dotted line) and longitudinal effect (solid line) of age (x axis) on composite frailty (y axis).

Change in Composite Frailty and Other Covariates

In a series of secondary analyses, we used the same analytic approach described above but we also added terms for several potentially confounding variables including vascular risk factors, vascular disease burden and use of psychotropic medications or antihypertensive medications and their interactions with time. Change in frailty was not affected by the addition of terms for vascular risk factors or diseases or the use of psychotropic or antihypertensive medications (coefficient estimates for frailty after adjusting for each term in separate models were 0.08, 0.08, 0.07 and 0.08 respectively).

Rate of Change in Composite Frailty and Risk of Mortality

Since the rate of change in frailty was relatively small (about a tenth of a unit per year), its clinical significance may be questioned. In a proportional hazards model controlling for age, sex, education and baseline frailty, we examined the relation of change in the rate of frailty and risk of death. During the course of the study 106 participants died. Baseline frailty and annual change in frailty were relatively independently associated with risk of death. The risk of death more than doubled with each 1-unit increase in baseline frailty (hazard ratio: 2.29; 95% CI: 1.58, 3.32). and the risk of death was increased more than five times with each 1-unit increase in annual change in frailty (hazard ratio: 4.97; 95% CI: 3.08, 8.02). Thus a participant at the 75th percentile of change in frailty (rapidly increasing frailty) had about a 1.25 times the risk of death as compared to a participant at the 25th percentile of annual change in frailty.

Rate of Change in the Components of Composite Frailty

We conducted similar analyses for each of the four individual components used to construct composite frailty. Each model included terms for age, sex, education and their interactions with time. Overall, grip strength impairment increased by about a tenth of a point per year (Time, Table 2). Older age was associated with a more rapid decline in strength (e.g., 2% faster per year, for a person who was 81 years of age at baseline versus a person who was 80) [Age*Time, Table 2]. Gait slowed at more than a tenth of a unit per year (Time, Table 2) and increased age was associated with a more rapid decline in gait speed (Age*Time, Table 2). The odds of fatigue increased by about 0.06 points per year [Time, Table 2; exp (0.06) = 1.06] with no association between age and change in fatigue (Age*Time, Table 2). Body composition increased by about 0.02 point per year during the study but did not change more rapidly in older participants (Table 2). Grip strength declined almost 80% more rapidly over time in men compared to women (Sex*Time, Table 2). The rate of change in the other three components was similar in men and women (Sex*Time, Table 2). Education was not associated with the rate of change in any of the four frailty components (Table 2).

Alternative Composite Measures of Frailty

Our primary measure of frailty was constructed from four components including grip strength, gait, body composition and fatigue. We examined the robustness of the approach by considering the rate of change in alternative composite measures of frailty based on different metrics and numbers of component variables. For these analyses, we again used generalized estimating equations with an identity link, and all models included terms for time, plus age, sex, education and their interactions with time.

Because there has been a particular emphasis on loss of weight in frailty, we examined the results of composite frailty in which we dichotomized BMI such that a BMI of less than 20 was considered frail (Sergi et al., 2005). This composite measure of frailty (mean −0.03, SD=0.59) showed significant change in frailty of about 13% of standard deviation per year (estimate 0.080; S.E. 0.010; p < 0.001). These results were qualitatively similar to those changes observed with composite frailty constructed using a continuous measure of BMI (Time, Table 2).

Next, we assessed the results of a composite measure of frailty using the actual timed walk instead of converting walk times into quintiles. For these analyses, those participants who could not walk were considered missing. This composite measure of frailty (mean −0.02, SD=0.58) showed significant change in frailty of about 11% of standard deviation per year (estimate 0.061; S.E. 0.007; p < 0.001). These results are qualitatively similar to those changes observed with composite frailty based on conversion of walk times to quintiles (Time, Table 2). Whereas there were few persons unable to walk in the study at this time, we wanted to ensure that the composite frailty measure was useful in the future at a time when an increasing number of participants would no longer be ambulatory. Therefore, we retained the quintile measure for the primary analyses.

Finally because there may be concerns that assessment of fatigue based on two fatigue questions rather than a continuous measure may skew the results, we first constructed a composite measure of frailty without the fatigue component, using only strength, gait and body composition. This measure of frailty based on three components (mean –0.04, SD=0.63) showed significant change in frailty of about 16% of standard deviation per year (estimate 0.10; S.E. 008 ; p < 0.001). These findings suggest that the present domain of fatigue based on two fatigue questions does not contribute much to the composite measure of frailty and may actually dilute the sensitivity of the composite measure (Time, Table 2). We also assessed frailty using the entire CES-D in place of the two fatigue questions. This measure of frailty (mean −0.03, SD=0.56) showed significant change in frailty of about 14% of a standard deviation per year (estimate 0.078; S.E. 0.007; p < 0.001), which was very similar to the composite measure with the two fatigue questions. However, using the entire CES-D would represent mood rather than fatigue, making it difficult to study the relationship of mood and frailty. We therefore retained the fatigue component based on the two CES-D questions despite its acknowledged metric limitations, so that our primary frailty variable would be most consistent with the published literature on frailty.

DISCUSSION

We used clinical data from more than 800 older persons without dementia who were examined annually for up to eight years to test whether a continuous composite measure of frailty based on strength, gait, body composition and fatigue could document the rate of change in frailty. First, we validated this continuous composite measure of frailty by showing that 1) the associations of baseline composite frailty with demographic variables were similar to those previously reported; 2) that it predicted mortality and incident disability; and 3) that it was correlated with a categorical measure of frailty used by other investigators. Next we have shown that composite frailty increased during the course of this study and that its rate of change varied among its individual components. Finally we have shown that even after controlling for baseline level of frailty, the rate of change in frailty was associated with risk of death. These data provide strong empirical evidence that frailty can be progressive in old age and that its level and the rate at which it progresses are both relatively independently associated with increased risk of death.

Most data on clinical frailty come from studies which assessed frailty at one point in time (Bandeen-Roche et al., 2006; Boyd et al., 2005; Ferrucci et al., 2004; Rockwood et al., 2004; Walston et al., 2002). Many cross sectional studies suggest that frailty increases with age and that women and those with less education are more frail (Boyd et al., 2005; Cohen et al., 2003; Fried et al., 2001; Mitnitski, Mogilner, MacKnight, & Rockwood, 2002; Newman et al., 2001; Ostir et al., 2004; Puts et al., 2005; Rockwood et al., 2004; Walston et al., 2002). Based on previous operational definitions of frailty (Boyd et al., 2005; Ferrucci et al., 2004; Fried et al., 2001; Ostir et al., 2004), we used measures of strength, gait, body composition and fatigue to construct a continuous composite measure of frailty. This composite measure behaved similarly to its antecedent categorical measure of frailty in previous studies, showing similar cross sectional associations with age, sex and education, and predicting death and incident disability. Most previous studies of frailty have employed categorical measures to classify an individual person and such classification may be necessary for treatment trials or interventions. In this study we have suggested a second approach using the same data that has been used to construct categorical measures in earlier studies (i.e. grip strength) but which conceptualizes each of the operational components of frailty as continua that may be added together and summarized as a composite measure of frailty. Since both categorical and composite frailty are correlated and predicted adverse health consequences these approaches are complementary.

While there have been important prospective studies of frailty (Gill et al., 2006; Ostir et al., 2004; Puts et al., 2005; Studenski et al., 2004), these studies have assessed the change from not being frail to being frail. One study used a categorical definition of frailty measured at two points in time and reported that about 20% of men and women became frail, and frailty was associated with an increased risk of mortality only among women (Puts et al., 2005). We are unaware of any prior work that used a continuous measure of frailty to document change in frailty over time. In the present study, we found that the annual rate of change in frailty was about 2.5 times the estimate of the effect of age from the cross-sectional analysis, suggesting that cross-sectional studies may under-estimate the progressive nature of frailty over time. This finding is consistent with a prior study that reported differences between cross-sectional estimates and longitudinal change in upper extremity sensorimotor performances in older persons (Desrosiers, Hebert, Bravo, & Rochette, 1998). More importantly not only were we able to document progression of frailty, but that the rate of change in frailty is clinically significant since it is associated with risk of mortality. The current findings provide support for the validity of composite frailty and reinforce a recent report in this same cohort, which showed that composite frailty is associated with incident AD and the rate of cognitive decline in the elderly (Buchman et al., in press).

The present study suggests an alternate approach to measuring frailty which conceptualizes each of the components of frailty as continua that may be added together and summarized as a composite measure of frailty. The use of a continuous measure is more likely to document change and allow assessment of the individual components of frailty which, as seen in this study, change at different rates. The various alternative methods used to summarize the different components of frailty all yielded composite measures that were all able to document change in frailty. These results underscore that despite the metric limitations of some of the components used to construct frailty, the resulting composite measures are robust and all can be used to identify change in frailty. Our results also show that frailty is not an inevitable consequence of aging and some persons show improvement in physical function over time. Recent work has underscored that frailty and disability can fluctuate over time, underscoring the need for analytic models which can account for non-linear changes in frailty over time (Gill et al., 2006). The present results augment the studies that have effectively employed composite measures to document change in cognitive and motor abilities, and on functional performance in both cross sectional and longitudinal studies (Louis et al., 2005; Onder et al., 2002; Petersen et al., 2005; Schalk et al., 2004; Schupf et al., 2005; Wilson et al., 2003). The results of this study suggest that composite measures, may also be used to study frailty to investigate associations and risk factors crucial for understanding its biologic substrate.

Composite frailty was constructed from individual components which themselves changed at different rates over time. Assessment of these components may also inform on change in frailty over time. For example, little data are available regarding the effect of age on change in frailty (Fried et al., 2001; Puts et al., 2005). Although baseline age was not associated with change in composite frailty, we did observe a steeper decline in both grip strength and gait speed among older persons at baseline. We did not observe changes in BMI during follow-up, but longer periods of time may be required to document change with this component (Elia, 2001). Similarly, the measure of fatigue employed in this study was also very crude. Thus, the extent to which fatigue increases in old age is uncertain. Overall, these observations raise the possibility that we underestimated the effect of age on change in frailty.

This study has several limitations: Since we excluded those participants who died before their first follow-up exam this would underestimate true changes in frailty overtime. Results are based on a selected group of participants willing to make organ donation at death, so replication of these results in a defined population will be important. Another limitation is the rather crude measure for fatigue. BMI does not differentiate between fat and muscle mass so other methods such as bioimpedance or DXA would add additional precision for assessment of fat free mass. Thus, more quantitative measures of body composition and fatigue, in addition to other aspects of frailty such as leg or respiratory strength or balance, might increase the sensitivity for measuring change in frailty.

Several factors increase confidence in the findings from this study: the composite measure of frailty was based on a uniform evaluation and accepted performance measures which are reliable and and easy to use in community-based studies; persons with dementia were excluded from analyses reducing the impact of cognitive impairment on our results; a relatively large number of older persons with a wide spectrum of education were examined increasing statistical power to identify the associations of interest while controlling for potentially confounding demographic variables. The availability of three or more repeated measurements per individual, for almost 75% of participants in these analyses, and the use of generalized estimating equations approach to fit our longitudinal generalized linear models, gives us confidence in our inferences. To our knowledge, these results provide the most direct evidence to date that frailty can be a progressive disorder in older persons.

ACKNOWLEDGMENTS

This research was supported by National Institute on Aging grants R01AG17917 and R01AG24480, the Illinois Department of Public Health, and the Robert C. Borwell Endowment Fund. We are indebted to the residents from the following groups participating in the Rush Memory and Aging Project: Fairview Village, Wyndemere, Luther Village, The Holmstad, Windsor Park Manor, Covenant Village, Bethlehem Woods, King-Bruwaert House, Friendship Village, Mayslake Village, The Moorings, Washington Jane Smith, Victory Lakes, Village Woods, Franciscan Village, Victorian Village, The Breakers of Edgewater, The Oaks, St. Paul Home, The Imperial, Frances Manor, Peace Village, Alden Waterford, Marian Village, The Birches, Elgin Housing Authority, Renaissance, Holland Home, Trinity United Church Of Christ, St. Andrews-Phoenix, Green Castle, Kingston Manor, Lawrence Manor, Community Renewal-Senior Ministry, Garden House, and the residents of the Chicago metropolitan area.

We thank Traci Colvin, Tracy Hagman, and Tracy Nowakowski for project coordination; Barbara Eubeler, Mary Futrell, Karen Lowe Graham, and Pamela Smith for participant recruitment; George Dombrowski and Greg Klein for data management; Liping Gu and Woojeong Bang for statistical programming, and the staff of the Rush Alzheimer’s Disease Center and Rush Institute for Healthy Aging.

REFERENCES

  1. Bandeen-Roche K, Xue Q-L, Ferrucci L, Walston J, Guralnik JM, Chaves P, Zeger SL, Fried LP. Phenotype of Frailty: Characterization in the Women's Health and Aging Studies. Journals of Gerontology: Medical Sciences. 2006;61A:262–266. doi: 10.1093/gerona/61.3.262. [DOI] [PubMed] [Google Scholar]
  2. Bennett DA, Schneider JA, Buchman AS, Mendes de Leon C, Bienias JL, Wilson RS. The Rush Memory and Aging Project: study design and baseline characteristics of the study cohort. Neuroepidemiology. 2005;25:163–175. doi: 10.1159/000087446. [DOI] [PubMed] [Google Scholar]
  3. Blaum CS, Xue QL, Michelon E, Semba RD, Fried LP. The association between obesity and the frailty syndrome in older women: the Women's Health and Aging Studies. Journal of the American Geriatrics Society. 2005;53:927–934. doi: 10.1111/j.1532-5415.2005.53300.x. [DOI] [PubMed] [Google Scholar]
  4. Boyd CM, Xue QL, Simpson CF, Guralnik JM, Fried LP. Frailty, hospitalization, and progression of disability in a cohort of disabled older women. American Journal Medicine. 2005;118:1225–1231. doi: 10.1016/j.amjmed.2005.01.062. [DOI] [PubMed] [Google Scholar]
  5. Boyle PA, Wilson RS, Aggarwal NT, Arvanitakis Z, Kelly J, Bienias JL, Bennett DA. Parkinsonian signs in subjects with mild cognitive impairment. Neurology. 2005;65:1901–1906. doi: 10.1212/01.wnl.0000188878.81385.73. [DOI] [PubMed] [Google Scholar]
  6. Buchman AS, Wilson RS, Bienias JL, Bennett DA. Gender differences in motor performance of older persons. Geriatrics Gerontology International. 2005;5:59–65. doi: 10.1111/j.1447-0594.2005.00266.x. [DOI] [PubMed] [Google Scholar]
  7. Buchman AS, Boyle PA, Wilson RS, Bienias JL, Bennett DA. Physical activity and motor decline in older persons. Muscle Nerve. 2007;35:345–362. doi: 10.1002/mus.20702. [DOI] [PubMed] [Google Scholar]
  8. Buchman AS, Boyle PA, Wilson RS, Tang Y, Bennett DA. Frailty Is Associated with Incident Alzheimer’s Disease and Cognitive Decline in the Elderly. Psychosomatic Medicine. doi: 10.1097/psy.0b013e318068de1d. (in press). [DOI] [PubMed] [Google Scholar]
  9. Cohen HJ, Harris T, Pieper CF. Coagulation and activation of inflammatory pathways in the development of functional decline and mortality in the elderly. American Journal of Medicine. 2003;114:180–187. doi: 10.1016/s0002-9343(02)01484-5. [DOI] [PubMed] [Google Scholar]
  10. Desrosiers J, Hebert R, Bravo G, Rochette A. Comparison of cross-sectional and longitudinal designs in the study of aging of upper extremity performance. Journal of Gerontology: Medical Sciences. 1998;53A:B362–B368. doi: 10.1093/gerona/53a.5.b362. [DOI] [PubMed] [Google Scholar]
  11. Elia M. Obesity in the elderly. Obesity Research. 2001;9:244S–248S. doi: 10.1038/oby.2001.126. [DOI] [PubMed] [Google Scholar]
  12. Ferrucci L, Guralnik JM, Studenski S, Fried LP, Cutler GB, Jr, Walston JD. Designing randomized, controlled trials aimed at preventing or delaying functional decline and disability in frail, older persons: a consensus report. Journal of the American Geriatrics Society. 2004;52:625–634. doi: 10.1111/j.1532-5415.2004.52174.x. [DOI] [PubMed] [Google Scholar]
  13. Fried LP, Tangen CM, Walston J, Newman AB, Hirsch C, Gottdiener J, Seeman T, Tracy R, Kop WJ, Burke G, McBurnie MA. Frailty in older adults:evidence for a phenotype. Journal of Gerontology:Medical Sciences. 2001;56A:M146–M156. doi: 10.1093/gerona/56.3.m146. [DOI] [PubMed] [Google Scholar]
  14. Gill TM, Gahbauer EA, Allore HG, Han L. Transitions between frailty states among community-living older persons. Archives of Internal Medicine. 2006;166:418–423. doi: 10.1001/archinte.166.4.418. [DOI] [PubMed] [Google Scholar]
  15. Katz S, Akpom CA. A measure of primary sociobiological functions. International Journal of Health Services. 1976;6:493–508. doi: 10.2190/UURL-2RYU-WRYD-EY3K. [DOI] [PubMed] [Google Scholar]
  16. Lawton MP, Brody EM. Assessment of older people: self-maintaining and instrumental activities of daily living. Gerontologist. 1969;9:179–186. [PubMed] [Google Scholar]
  17. Liang KY, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika. 1986;73:13–22. [Google Scholar]
  18. Louis ED, Tang MX, Schupf N, Mayeux R. Functional correlates and prevalence of mild parkinsonian signs in a community population of older people. Archives of Neurology. 2005;62:297–302. doi: 10.1001/archneur.62.2.297. [DOI] [PubMed] [Google Scholar]
  19. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM. Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's Disease. Neurology. 1984;34:939–944. doi: 10.1212/wnl.34.7.939. [DOI] [PubMed] [Google Scholar]
  20. McPhillips JB, Pellettera KM, Barrett-Connor E, Wingard DL, Criqui MH. Exercise patterns in a population of older adults. American Journal of Preventive Medicine. 1989;5:65–72. [PubMed] [Google Scholar]
  21. Mitnitski AB, Mogilner AJ, MacKnight C, Rockwood K. The mortality rate as a function of accumulated deficits in a frailty index. Mechinisms of Ageing and Development. 2002;123:1457–1460. doi: 10.1016/s0047-6374(02)00082-9. [DOI] [PubMed] [Google Scholar]
  22. Newman AB, Gottdiener JS, McBurnie MA, Hirsch CH, Kop WJ, Tracy R, Walston JD, Fried LP. Associations of subclinical cardiovascular disease with frailty. Journal of Gerontology:Medical Sciences. 2001;56A:M158–M166. doi: 10.1093/gerona/56.3.m158. [DOI] [PubMed] [Google Scholar]
  23. Onder G, Penninx BW, Lapuerta P, Fried LP, Ostir GV, Guralnik JM, Pahor M. Change in physical performance over time in older women: the Women's Health and Aging Study. Journal of Gerontology: Medical Sciences. 2002;57A:M289–M293. doi: 10.1093/gerona/57.5.m289. [DOI] [PubMed] [Google Scholar]
  24. Ostir GV, Ottenbacher KJ, Markides KS. Onset of frailty in older adults and the protective role of positive affect. Psychololgy and Aging. 2004;19:402–408. doi: 10.1037/0882-7974.19.3.402. [DOI] [PubMed] [Google Scholar]
  25. Petersen RC, Thomas RG, Grundman M, Bennett D, Doody R, Ferris S, Galasko D, Jin S, Kaye J, Levey A, Pfeiffer E, Sano M, van Dyck CH, Thal LJ. Vitamin E and donepezil for the treatment of mild cognitive impairment. New England Journal of Medicine. 2005;352:2379–2388. doi: 10.1056/NEJMoa050151. [DOI] [PubMed] [Google Scholar]
  26. Puts MT, Lips P, Deeg DJ. Sex differences in the risk of frailty for mortality independent of disability and chronic diseases. Journal of the American Geriatrics Society. 2005;53:40–47. doi: 10.1111/j.1532-5415.2005.53008.x. [DOI] [PubMed] [Google Scholar]
  27. Radloff LS. The CES-D scale: A self-report depression scale for research in the general population. Applied Psychological Measurement. 1977;1 385-340. [Google Scholar]
  28. Rockwood K, Howlett SE, MacKnight C, Beattie BL, Bergman H, Hebert R, Hogan DB, Wolfson C, McDowell I. Prevalence, attributes, and outcomes of fitness and frailty in community-dwelling older adults: report from the Canadian study of health and aging. Journal of Gerontology: Medical Sciences. 2004;59A:1310–1317. doi: 10.1093/gerona/59.12.1310. [DOI] [PubMed] [Google Scholar]
  29. SAS. SAS/STAT® user's guide, (version 8) Cary, NC: SAS Institute Inc; 2000. [Google Scholar]
  30. Schalk BW, Visser M, Deeg DJ, Bouter LM. Lower levels of serum albumin and total cholesterol and future decline in functional performance in older persons: the Longitudinal Aging Study Amsterdam. Age Ageing. 2004;33:266–272. doi: 10.1093/ageing/afh073. [DOI] [PubMed] [Google Scholar]
  31. Schupf N, Tang MX, Albert SM, Costa R, Andrews H, Lee JH, Hogan DB, Wolfson C, McDowell I. Decline in cognitive and functional skills increases mortality risk in nondemented elderly. Neurology. 2005;65:1218–1226. doi: 10.1212/01.wnl.0000180970.07386.cb. [DOI] [PubMed] [Google Scholar]
  32. Sergi G, Perissinotto E, Pisent C, Buja A, Maggi S, Coin A, Grigoletto F, Enzi G. An adequate threshold for body mass index to detect underweight condition in elderly persons: the Italian Longitudinal Study on Aging (ILSA) Journal of Gerontology: Medical Sciences. 2005;60A:866–871. doi: 10.1093/gerona/60.7.866. [DOI] [PubMed] [Google Scholar]
  33. Studenski S, Hayes RP, Leibowitz RQ, Bode R, Lavery L, Walston J, Duncan P, Perera S. Clinical Global Impression of Change in Physical Frailty: development of a measure based on clinical judgment. Journal of the American Geriatrics Society. 2004;52:1560–1566. doi: 10.1111/j.1532-5415.2004.52423.x. [DOI] [PubMed] [Google Scholar]
  34. Walston J, McBurnie MA, Newman A, Tracy RP, Kop WJ, Hirsch CH, Gottdiener J, Fried LP. Frailty and activation of the inflammation and coagulation systems with and without clinical comorbidities: results from the Cardiovascular Health Study. Archives of Internal Medicine. 2002;162:2333–2341. doi: 10.1001/archinte.162.20.2333. [DOI] [PubMed] [Google Scholar]
  35. Wilson RS, Barnes LL, Bennett DA. Assessment of lifetime participation in cognitively stimulating activies. Journal of Clinical and Experimental Neuropsychology. 2003;23:634–642. doi: 10.1076/jcen.25.5.634.14572. [DOI] [PubMed] [Google Scholar]
  36. Wilson RS, Schneider JA, Bienias JL, Evans DA, Bennett DA. Parkinsonian like signs and risk of incident Alzheimer disease in older persons. Archives of Neurology. 2003;60:539–544. doi: 10.1001/archneur.60.4.539. [DOI] [PubMed] [Google Scholar]
  37. Wilson RS, Barnes LL, Krueger KR, Hoganson G, Bienias JL, Bennett DA. Early and late life cognitive activity and cognitive systems in old age. Journal International Neuropsychological Society. 2005;11:400–407. [PubMed] [Google Scholar]
  38. Wilson RS, Buchman AS, Arnold SE, Shah RC, Tang Y, Bennett DA. Harm avoidance and disability in old age. Experimental Aging Research. 2006;32:243–261. doi: 10.1080/03610730600699142. [DOI] [PubMed] [Google Scholar]
  39. Zeger SL, Liang KY, Albert PS. Models for longitudinal data: a generalized estimating equation approach. Biometrics. 1988;44:1049–1060. [PubMed] [Google Scholar]

RESOURCES