Article Text

Download PDFPDF

Assessing the measurement properties of a Frailty Index across the age spectrum in the Canadian Longitudinal Study on Aging
  1. David M Kanters1,
  2. Lauren E Griffith1,2,3,
  3. David B Hogan4,
  4. Julie Richardson5,
  5. Christopher Patterson6,
  6. Parminder Raina1,2,3
  1. 1 Department of Health Research Methods, Evidence, and Impact, McMaster University, Hamilton, Canada
  2. 2 McMaster Institute for Research on Aging, Hamilton, Canada
  3. 3 Labarge Centre for Mobility in Aging, McMaster University, Hamilton, Canada
  4. 4 Division of Geriatric Medicine, Department of Medicine, Cumming School of Medicine, University of Calgary, Calgary, Canada
  5. 5 School of Rehabilitation Science, McMaster University, Hamilton, Canada
  6. 6 Division of Geriatric Medicine, Department of Medicine, McMaster University, Hamilton, Canada
  1. Correspondence to Dr Lauren E Griffith, Department of Health Research Methods, Evidence, and Impact, MIP 309A, McMaster University, 175 Longwood Rd. S., Hamilton, ON L8P 0A1, Canada; griffith{at}


Background Frailty is a way to appreciate the variable vulnerability to declining health status of people as they age. No consensus for measuring frailty has been established. This study aimed to adapt a Frailty Index (FI) to the Canadian Longitudinal Study on Aging (CLSA) and evaluate its applicability in both younger and older adults.

Methods An FI was created based on 90 potential health deficits collected from adults aged 45–85 years at recruitment (N=21 241, 49.0% male). The construct validity of this instrument and the factor structure of the health deficits were evaluated.

Results The direction of associations between the FI and other variables were consistent with a priori hypotheses for construct validity. FI values were significantly associated with age (r=0.17; p<0.001), falls (r=0.12; p<0.001), injuries (r=0.12; p<0.001), formal home care (r =0.30; p<0.001), informal home care (r=0.32; p<0.001) and use of assistive devices (r=0.40; p<0.001). Values were negatively associated with male sex (r=−0.12; p<0.001), income (r=−0.34; p<0.001) and education (r=−0.17; p<0.001). Key factors among the health indicators were physical functioning, satisfaction with life and depressive symptoms. Results did not change when the sample was stratified by age and sex.

Conclusion The FI is a feasible method to evaluate frailty and capture frailty-related heterogeneity in populations aged 45–85 years. In this study, the FI had good construct validity in middle-aged and older adults, showing expected correlations with sociodemographic factors consistently across age groups. This method can be easily reproduced in similar datasets, making the FI a generalisable instrument.

  • CLSA
  • Measurement tool development
  • Ageing
  • Epidemiology of ageing

Statistics from


The Canadian population is ageing, with the proportion of adults over 65 years of age progressively increasing.1 As age increases, so does vulnerability to declining health status, and increased risk of institutionalisation and mortality. This vulnerability varies among people of the same age and is commonly referred to as frailty.2 Frailty is emerging as an important public health priority. Assessing frailty is felt by many to be necessary for evaluating the healthcare needs of ageing adults and to target a group for interventions aimed at preventing adverse health-related outcomes.3

No consensus operational definition or standard measure of frailty has emerged.4 Attempts to operationalise frailty measurement most commonly fall into one of two approaches. The frailty phenotype model proposed by Fried et al defines frailty as the presence of three or more of five criteria: exhaustion, weight loss, diminished activity level, slow walking speed and weak grip strength.5 The other commonly used measure is a Frailty Index (FI), based on a cumulative deficit approach as described by Rockwood and Mitnitski.6 7 Here frailty is measured as the proportion of age-related health deficits present in an individual, providing a continuous value indicating where the person stands on a fit-to-frail spectrum.6 The deficits considered should represent a variety of areas of health, including psychological, social and environmental factors. There is no prescriptive list of deficits. Indices composed of 30 or more deficits have been shown to produce a valid result, even when selected at random, as long as certain criteria are used in their selection.8 Researchers can adapt an FI approach to fit the patient population being studied and the data available.7 Both approaches have been shown to predict higher risks of future adverse health outcomes.9 Though interventions to mitigate frailty have yet to be shown effective, younger individuals have more opportunity to benefit from preventive and therapeutic approaches that will hopefully be developed.10 Examining and comparing important indicators of frailty in younger and older adults may provide important insights into the mechanisms of frailty and its progression as people age.

The Canadian Longitudinal Study on Aging (CLSA) has recruited a nationally representative sample of participants aged 45–85 years from all 10 provinces of Canada. The CLSA is one of the largest and most comprehensive research platforms in the world. It was designed to examine ageing through a number of different lenses by collecting a wide range of information over time about the changing biological, medical, psychological, social, lifestyle and economic characteristics of participants. The CLSA will follow participants for 20 years and will allow for the examination of midlife factors that may be associated with frailty in later life. Developing a measure for frailty that is interpretable and generalisable with good psychometric properties would be valuable for the study of frailty and evaluation of its role as a potential confounder in other studies using CLSA data. This study employed data from the CLSA Tracking Cohort, which include information from the first wave of data collected thorough telephone interview. The CLSA Comprehensive Cohort will be available in the future and will include prospective data collected through in-person interviews, physical assessments and biological specimens. Raina et al 11 provides an overview of the CLSA, and the full study protocol can be found on the CLSA website.12 This is a secondary analysis of baseline data collected on the CLSA tracking cohort.

The primary objective of this study was to adapt an FI to the CLSA using methodology based on the accumulated deficit approach,6 and to assess the construct validity and applicability of this measurement tool in adults aged 45–85 years. This was the most appropriate approach for frailty detection in the CLSA as comprehensive data, including many variables associated with frailty, were collected on participants but no physical measures, such as gait speed and grip strength were available on the Tracking Cohort used in this study. Other studies have demonstrated FIs using routinely collected health data.6 13 As a secondary objective, we examined the latent constructs of the FI using exploratory factor analysis (EFA), to identify which, if any, variables are key contributors to frailty.



The CLSA tracking cohort consists of a national stratified random sample of 21 241 adults between the ages of 45 and 85 years old at recruitment, of whom 49% were male and 42% were over the age of 65 years. This sample of community-dwelling adults was selected to be representative of the Canadian population for provincial estimates of health determinants, health status and health system utilisation.11 Those in long-term care, residents of Canadian territories or First Nations reserves, and full-time armed forces members were excluded from the study population.11 Significant cognitive impairment at baseline was a criterion for exclusion, as this can compromise informed consent and the interview responses.11 These eligibility criteria were adapted from the Canadian Community Health Survey (CCHS).14

Data collection

CLSA investigators collaborated with Statistics Canada on the sampling strategy. The first sampling frame for the CLSA was the CCHS.14 To reach the intended sample of 20 000 participants, tracking participants were also recruited through provincial healthcare registration databases and random digit dialling. Each participant provided written consent and completed a 60 min telephone interview.14

Frailty index

A wide array of self-reported indicators of health representing physical, psychological and social domains are collected in the CLSA. A literature review identified five systematic reviews of studies that developed an original measure for frailty using self-reported data.15–19 All CLSA variables previously used as part of a frailty measurement tool were selected for possible inclusion in our FI. Our research team reviewed this list for completion and relevance. The research team included physiotherapists, geriatricians and public health researchers, with an abundance of experience in the study of ageing, frailty and frailty measurement. This process resulted in a list of potential deficits for inclusion. The final selection of variables was based on the following criteria7: deficit associated with health status, becomes more common with ageing, does not saturate too early, and covers a range of systems. Deficits were transformed to a value from 0 (no deficit) to 1 (most severe deficit), and an individual’s FI value was calculated as the sum of deficits present divided by the total number considered. Many of the health measurement scales included in the CLSA contain multiple domains, some of which are common across scales, so individual items from health measurement scales were considered deficits for the FI. An FI was not calculated for individuals with >5% missing data. The 90 health deficits included in the FI we created are shown in online supplementary appendix 1.

Supplementary Material

Supplementary Appendix 1

Health status measures included self-rated physical and mental health, sensory impairment, height, weight and chronic conditions. Functional status was measured using 14 items from Framingham Disability Study, Established Populations for Epidemiologic Studies of the Elderly study and the Nagi and Rosow-Breslau scales.20–23 Activities of daily living are measured using the Older Americans Resources and Services (OARS) Multidimensional Assessment Questionnaire.24 Two domains of cognition are assessed in the CLSA tracking cohort, memory and executive function. Memory is assessed using the Rey Auditory Verbal Learning Test25 and executive function using the mental alternation test (MAT)26 and animal naming test of verbal fluency. Depressive symptoms are measured using the Center for Epidemiologic Studies Short Depression Scale (CES-D 10).27 The Satisfaction with Life Scale (SWLS) is included to measure positive mood state or life satisfaction, an important measure for self-assessment of health and well-being.28 Social support is measured using the 19-item MOS Social Support Survey.29 CLSA tracking includes eight items on the frequency of participation in a variety of social activities, and participants’ desire to participate more in social activities.

Construct validation

The construct validity of the FI in the CLSA was evaluated through comparison with sociodemographic and healthcare utilisation variables in the dataset. These included sex, age, number of chronic conditions, social participation, injuries, falls causing injury, formal and informal (including unpaid) home care, and the use of assistive devices for mobility, such as a cane, walker or wheelchair. We hypothesised that FI values should increase with age, female sex, number of chronic conditions, injuries, falls, assistive device use and home care received, and that values would be lower in those with more education and higher annual household income. If these hypotheses were met, the study would support both our conceptualisation of frailty and the FI as a method of measurement.30

Exploratory factor analysis

EFA was used to estimate latent factor structure of the 90 observed variables comprising the FI. Polychoric bivariate correlations were used, as many of the included variables were binary.31 The principal axes method of factor estimation was used to extract factors. The number of factors retained was determined using the Cattell scree plot of eigenvalues, which helps interpret the additional variation in the data explained by each successive factor.32 Orthogonal varimax rotation was used as many of the factors were strongly correlated.33 The minimum factor loading cut-off for an item to be considered part of a factor was 0.30, based on convention.33 A sample size of 21 241 is sufficient for this analysis, far exceeding the recommended minimum sample based on a subject to item ratio of 10:1.33


Frailty index

An FI value was calculated on 20 874 participants (98% of total potential number), with a mean of 0.14 and SD of 0.07. The distribution of FI values approximated a normal distribution with a skewness of 1.55 and kurtosis of 3.8. The minimum FI for the full sample was 0.003 while the maximum was 0.677. Using an established cut-off of 0.25 for detecting frailty based on a comparison to the phenotype approach,34 1449 of 20 874 (6.94%) participants were frail, including 950 of 10 654 (8.92%) female and 499 of 10 220 (4.88%) male participants. Although a smaller proportion of participants aged 45–54 years old were frail (315 of 5781, 5.45%) compared with those aged 75 or more years (431 of 4078, 10.57%), these results show that frailty affects both middle-aged and older adults. Each age group had a higher proportion of women compared with men with an FI value above 0.25 and this difference increased with age (table 1). Figure 1 displays the linear regression of the relationship between FI value and age. FI values increase with age and the rate is higher in women compared with men.

Table 1

Frequency and proportion of participants with frailty index (FI) ≥0.25

Figure 1

Linear regression of age and frailty index with sex interaction.

Construct validation

Pearson correlations between the FI and other health indicators were examined (table 2). In the full sample, these correlations were statistically significant at the p<0.001 level and in the direction hypothesised. The FI in the entire population and when stratified by age was positively correlated with age, fall status, injuries, assistive device use, and home care, and negatively correlated with income, education, and male sex. Correlations between FI and age, sex, home care, injuries and assistive devices were strongest in the ≥75 age group and became stronger with age, while correlations between FI and education and income weakened as age increased.

Table 2

Frailty index Pearson correlation with demographic and healthcare variables

Exploratory factor analysis

The strongest associations between items were found within scales: SWLS (0.50–0.74), CES-D 10 (0.14–0.74), OARS (0.11–0.91) and MOS Social Support (0.69–0.83). The Cattell scree plot (figure 2) showed that three factors should be retained. These three factors accounted for 33%, 17% and 7%, respectively, of the variance of the observations (ie, total of 56% variance explained). The factor loadings of the three retained factors after rotation are shown in table 3. The minimum cut-off was 0.30 for a variable to be considered a defining part of a factor. When items with low factor loadings were removed, the factor structure remained unchanged.

Table 3

Exploratory factor analysis rotated factor pattern 

Figure 2

Exploratory factor analysis Cattell scree plot.

Factor 1 had high loadings (0.623–0.974) from functional status and self-rated health and moderate loadings for some chronic conditions. Two items from the CES-D 10 depression scale loaded on factor 1: frequency ‘feel everything is an effort’ (0.42) and frequency ‘have trouble get going’ (0.33). The self-rated vision item also had a small loading on factor 1 (0.30).

Factor 2 included high loadings from the satisfaction with life scale (0.86–0.91)and some CES-D 10 depression items, including frequency feel ‘happy’ (0.71) and ‘hopeful for the future’ (0.73). Self-rated mental health (0.59), social support availability (0.38–0.48), and social participation (0.35–0.41) items also loaded on factor 2.

Factor 3 had high loadings from CES-D 10 depression items that did not load strongly on either of the previous factors (0.56–0.83). Other mental health items also had moderate loadings, including mood disorder (0.510), anxiety (0.47) and self-rated mental health (0.47).

When the sample was stratified by age and sex, the structure of the first three factors remained constant, with only small variation in the variable loadings. These three factors—labelled functional status, life satisfaction and depressive symptoms— accounted for the bulk of the variance in the measurement of frailty-related variables in the sample regardless of age or sex.


The FI in this sample was free of floor or ceiling effects. No participant, not even the youngest members of the sample, had a value of 0. Our data suggest that this is an important measure for capturing frailty-related heterogeneity in groups aged 45–85 years. This tool can differentiate between levels of frailty, which allows us to study the mechanisms of frailty in different age groups. The distribution FI values, including the maximum of 0.68, was consistent with other studies measuring an FI in community-dwelling adults including the English Longitudinal Study on Ageing35 and Canadian Multicentre Osteoporosis Study.36 This consistent maximum value in relatively healthy samples collected in large population-based studies of community-dwelling adults may be a theoretical maximum FI value for an individual who has not experienced adverse health outcomes, such as mortality or institutionalisation, which would preclude inclusion in this type of study. The continued study of the FI in the CLSA as longitudinal data become available will provide a clearer picture of frailty measurement, and the capacity of the FI to accurately measure how frailty changes as people age and predict adverse health outcomes. A measurement tool that can identify individuals early in the progression of frailty presents an opportunity to change their trajectory.

The direction of the correlations between the FI and other participant characteristics was consistent with what would be expected for a measurement of frailty, providing evidence that frailty is the construct being measured. These results should be interpreted cautiously as, although all associations were statistically significant, the magnitude of several correlations was small (r<0.30).37 Only income, injuries, falls, formal home care, informal home care and assistive device use had moderate correlations (r≥0.30). The direction of these correlations was consistent across age groups, while the strength showed that the factors associated with frailty may be different in younger and older adults. Income and education were strongly associated with higher values in younger age groups while age, sex, home care use and assistive devices were more strongly associated with higher values in older age groups. Statistically significant association between frailty, measured using the FI, and injuries has been shown previously in young-aged and middle-aged adults.36 Further research is required to properly ascertain the reliability and construct validity of this measurement model using prospective data.

Three key factors were found in the EFA. The first and strongest factor in terms of proportion of variance among observations explained was the Physical Frailty factor. This supports the supposition that physical measures, such as strength, mobility and self-care are the most important contributors to frailty.38 The frailty phenotype, for example, is based on Physical Frailty measures and is a common approach to frailty in population-based studies.16 The second and third overlapping factors identified, Life Satisfaction and Depressive Symptoms, do indicate there is an important mental health component to frailty. Life Satisfaction includes satisfaction with life and items dealing with mood, hopefulness, mental health and social participation. The Depressive Symptoms consisted of items primarily from a depression questionnaire. Our findings are consistent with research suggesting that psychological vulnerability may be an important component of frailty and items from this domain should be included in the FI.39 In stratified analyses, the same factor pattern was seen in each age group, which indicates the FI has the same factor structure in both middle-aged and older adults.

The FI has predominantly been studied in populations over 65 years.16 The present study differs from this research by studying a relatively younger population of community-dwelling adults aged 45–85 years. The longitudinal follow-up of frailty conceptualised as a continuum in this population could provide new insights on the development, regression, and progression of frailty as this sample ages using this tool. It will also allow the comparison of factors associated with frailty in relatively older and younger populations. For example, previous work on the allied concept of disability in a population-based study of those 65 years of age and older showed a more important role for disease in the development of disability among those 65–84 compared with subjects 85 and older.40 The results of our analyses of the FI derived from CLSA data were consistent in both middle-aged and older adults. While primarily used in older populations, this work suggests that frailty indices may be a meaningful measure in middle-aged adults as well.

A number of limitations must be acknowledged. CLSA participants at the time of enrolment were aged 45–85 years, residing in the community, and able to provide informed consent. Recruitment excluded those over 85 years of age, living in long-term care facilities and/or unable to provide informed consent. As a result, the sample may be subject to a healthy participant bias by excluding those most likely to be frail (ie, the very old, residents of long-term care facilities and sufferers of dementia). This limits the generalisability of our results to the entire ageing population in Canada and may impair our ability to compare our findings to previous research on the FI, which has tended to focus on older (ie, ≥65 years) populations. As noted above, this limitation is balanced by our ability to look at preclinical and prodromic stages of frailty by the inclusion of a middle-aged population and the opportunities that will arise because of the longitudinal nature of the CLSA. We will be able to investigate frailty trajectories and contrast the components of the frailty syndrome at different ages using this measurement tool. The current study is limited by the use of cross-sectional data. The available data could not be used to assess the predictive validity of the FI for adverse health outcomes, a critical criterion for any frailty measurement.

Strengths other than the ones already noted include the population-based nature of our study. In future analyses, prospective CLSA data will be used for validation of the FI through its predictive capacity for adverse health outcomes such as mortality and institutionalisation. The CLSA was designed to be comprehensive and is collecting a wide array of data on participants. We were able to map potential indicators for frailty identified in the literature to the data being obtained on participants of CLSA.


An FI that underwent construct validation was developed from the data available in the CLSA. The widespread adoption of frailty indices in population-based research supports the generalisability of this instrument. The continued validation and exploration of the FI will promote better understanding of the determinants and outcomes of frailty as well as allow the use of this as an explanatory or confounding variable in other studies using CLSA data.

What is already known on this subject

As people age, they experience changes in their vulnerability to declining health status, which is commonly called frailty. In countries with ageing populations, frailty is gaining prominence as a healthcare priority. However, there is no consensus operational definition or standard measure of frailty.

What this study adds

An FI was created to measure the frailty of participants in the Canadian Longitudinal Study on Aging. This study contributes to the growing body of evidence that an FI is a valid measure of frailty in middle-aged and older adults and is feasible to create and interpret in population-based studies.


LEG is supported by a Canadian Institutes of Health Research New Investigators Award and the McLaughlin Foundation Professorship in Population and Public Health. PR holds a T1 Canada Research Chair in Geroscience and the Raymond and Margaret Labarge Chair in Research and Knowledge Application for Optimal Aging. The CLSA is led by PR and CP.



  • Contributors This research was conducted by DMK under the supervision of LEG. DBH, CP, PR, and JR contributed to the conception of the study and acted as scientific advisors, providing feedback on the design and clinical relevance of the study. All authors contributed to the analysis and interpretation of data and preparation of the manuscript.

  • Funding Funding for the CLSA is provided by the Government of Canada through the Canadian Institutes of Health Research (CIHR) under grant reference: LSA 94473 and the Canada Foundation for Innovation.

  • Competing interests None declared.

  • Ethics approval Hamilton Integrated Research Ethics Board.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement This study employed data from the CLSA. The CLSA has recruited a nationally representative sample of participants aged 45–85 years from all 10 provinces of Canada. The CLSA is one of the largest and most comprehensive research platforms in the world. It was designed to examine ageing through a number of different lenses by collecting a wide range of information over time about the changing biological, medical, psychological, social, lifestyle and economic characteristics of participants. Researchers can learn more about the CLSA and submit a data access request through

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.