Article Text

Download PDFPDF
Trajectories of phenotypical frailty over a decade in young–old community-dwelling adults: results from the Lc65+ study
  1. Sarah Fustinoni,
  2. Brigitte Santos-Eggimann,
  3. Yves Henchoz
  1. Department of Epidemiology and Health Systems, Center for Primary Care and Public Health (Unisanté), University of Lausanne, Switzerland
  1. Correspondence to Mrs Sarah Fustinoni, Center for Primary Care and Public Health, Lausanne, Switzerland; sarah.fustinoni{at}unisante.ch

Abstract

Background Few studies have examined the frailty trajectories of young–old adults using Fried frailty phenotype. Dropouts due to death were rarely taken into account. This longitudinal study aimed to identify trajectories with and without adjustment for non-random attrition and to analyse related factors.

Methods We used the first two samples of community-dwelling people in the Lausanne cohort 65+. Frailty phenotype was assessed at age 66–71 years and every third year over 10 years. A group-based trajectory modelling—first without and then with adjustment for non-random attrition—identified trajectories among all individuals with at least two observations (n=2286), excluding dropouts for reasons other than death. Multinomial logistic regressions estimated independent effects of participants’ baseline characteristics.

Results We identified three frailty trajectories (low, medium and high). Participants in the highest trajectory had a higher mortality over 10 years. (Pre)frailty at baseline was the main factor associated with adverse trajectories. Smoking, obesity, comorbidity and negative self-perceived health were associated with unfavourable trajectories independently of baseline frailty, while social engagement was related to the lowest frailty trajectory. Ignoring transitions to death attenuated the estimated effects of age on trajectories.

Conclusions Fried frailty phenotype should be assessed in individuals aged late 60s as it is strongly associated with frailty trajectories in the following decade of their life. Lifetime prevention of behavioural risk factors such as smoking and obesity is the strategy most likely to influence the development of frailty in older populations. Furthermore, our results underline social engagement as an important area of interest for future research.

  • epidemiology of ageing
  • longitudinal studies
  • cohort studies
  • mortality

Data availability statement

Data are available upon reasonable request. The data that support the findings of this study are deidentified participant data. Details of the data and how to request access are available from the principal investigator of the Lc65+ study (Yves.Henchoz@unisante.ch) at the Centre for Primary Care and Public Health (Unisanté), University of Lausanne, Switzerland.

Statistics from Altmetric.com

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.

Introduction

Twenty years after Fried et al had defined a frailty phenotype—first measured in the Cardiovascular Health Study1—a large body of research documented its association with risks of adverse outcomes in old age, including death.2 In ageing societies, preventing frailty from its earliest stages to avoid harmful consequences has become a public health and medical priority.3 Fried frailty phenotype and Rockwood’s frailty index appear to be the two most commonly used assessment tools in epidemiological research.4 Nevertheless, they are distinct and not substitutable instruments.5 Fried phenotype relies on a clear conceptual model based on the biology of the ageing process, which explains its fast spread in the medical field. Many epidemiological studies pointed to a low frequency of frailty in late middle age and an increasing prevalence in old age,6 reaching up to 77% among centenarians.7 Prefrailty, however, affects a large proportion of the youngest old people8 who could be an optimal target for early preventive actions.

The natural history of frailty is still under investigation.9 Since Gill et al have tackled frailty as a dynamic process in their early work in 2006,10 several studies of Fried phenotype have described transitions between two time points,11 12 particularly of the prefrail phenotype that can turn into both a non-frail or a frail phenotype. Individuals with frailty occasionally return to prefrailty, but their transition to non-frailty seems exceptional.10 11 Studies of frailty transitions that included transitions to death constantly found a markedly higher all-cause mortality among frail persons.11 As shown in a recent review, accounting for progression to death in studies of frailty transitions strongly lowers the estimate of remission rates.12

The long-term evolution of frailty was first described essentially based on the overall population mean of cumulated deficits.13 Results pointed to differences in successive cohorts,14–17 countries16 and population subgroups.14–19 More recently, a few studies have investigated the diversity of frailty trajectories followed by the members of the same cohort using repeated measurements of a frailty phenotype,20–24 index25–27 or scale.28 Previous studies based on a phenotypical approach of frailty identified trajectories in wide age range populations, and none of them focused specifically on young–old adults. These same analyses rarely adjusted for non-random attrition (predominantly mortality) when establishing the trajectories. Because intervals of several years separated the repeated measurements of the frailty phenotype in these studies,21–23 the states of severe frailty shortly preceding death may have escaped to observation and consequently to the construction of trajectories.

Our study aimed, first, to compute 10-year trajectories of a frailty phenotype starting at the age of 66–71 years; second, to identify baseline demographic, socioeconomic and health characteristics associated with frailty trajectories; and third, to explore the influence of non-random dropouts on the results by taking into account attrition due to death.

Methods

Study population

The Lausanne cohort 65+ is a population-based study of the manifestations, determinants and outcomes of frailty in community-dwelling older adults. Participants’ enrolment has previously been described in detail.29 In brief, 4731 randomly selected individuals aged 65–70 years, living in the city of Lausanne, Switzerland, were recruited in three successive samples (2004, 2009 and 2014) and invited to participate in a frailty phenotype baseline assessment conducted by trained medical assistants at the study centre during the following year. Follow-up assessments included annual questionnaires and triennial frailty phenotype assessments. The current research used a 10-year dataset from the first two samples (2005–2014 and 2010–2019, respectively).

Frailty assessment and score

Frailty was assessed according to the five components of Fried phenotype1 operationalised as follows29:

  1. Unintentional weight loss was defined as any reported unintentional weight loss over the past year.

  2. Self-reported exhaustion was defined as the response “much” to the question “Did you have feelings of generalised weakness, weariness, lack of energy in the last 4 weeks?”.

  3. Low physical activity was defined as reporting <20 min of sport activity once a week and <30 cumulated minutes of walking per day three times a week, and avoiding climbing stairs or carrying light loads in daily activities.

  4. Muscular weakness was defined as a low grip strength measure.

  5. Slow walking speed was defined as a low gait speed in a 20-metre walk test.

Low grip strength and slow walking were defined using Fried et al’s reference cut-off values observed in the Cardiovascular Health Study.1 The number of criteria fulfilled divided by the number of criteria evaluated, multiplied by 5, was considered as a score of frailty (0 to 5).

The frailty score of individuals for whom the five criteria could not be assessed but for whom at least three criteria could be measured (between 8% and 12% of scores depending on the year) was retained. For the others, the frailty score was considered missing for the given year. A score of 4 was assigned to participants admitted in a nursing home with a missing frailty score (until the next non-missing value, the end of follow-up or death), reflecting an advanced state of frailty of institutionalised participants. As a covariate in multivariable analyses of the whole sample, the baseline assessment of frailty was dichotomised into ‘non-frail’ (score 0) versus ‘(pre)frail’ (score >0).

Covariates

A set of commonly used covariates known to be associated with frailty were selected from the baseline assessment of population characteristics.30 31

Sociodemographic characteristics

Sociodemographics at baseline included sex; education, defined as the highest completed level of education (‘basic compulsory’ corresponding to levels 0–2 of the International Standard Classification of Education (ISCED),32 ‘apprenticeship’ or ISCED 3, and ‘postcompulsory’ or ISCED 4–8); living arrangement (‘alone’ vs ‘not alone’); and social engagement, defined as having a gainful or an extra-household volunteering activity over the last 3 months (‘yes’ vs ‘no’).

Health-related characteristics

Baseline health-related characteristics included self-rated health categorised as ‘very good or good’ versus ‘average, poor or very poor’; obesity defined as body mass index of ≥30 kg/m2 based on measured height and weight; depressive feelings (yes vs no), defined as a positive response to either of the following two questions: ‘During the past month, have you often been bothered by (1) feeling down, depressed or hopeless? (2) little interest in doing things?’33 ; cognitive complaints (yes vs no) defined by reports of any of the three following symptoms: memory loss, difficulty making decisions in daily life and difficulty concentrating; smoking status categorised into ‘current smoker’, ‘former smoker’ and ‘never smoker’; and number of chronic diseases defined by the count of reported lifetime diagnoses from a list of nine conditions (hypertension, coronary heart disease, other heart disease, stroke, diabetes mellitus, chronic respiratory disease, osteoporosis, arthritis and cancer), recoded into ‘0’, ‘1’, ‘2’ and ‘3 or more’.

Statistical analyses

Trajectory groups

Latent trajectory groups were determined by using group-based trajectory modelling (GBTM), a specialised application of finite mixture modelling that identifies clusters of observations with similar trajectories over discrete time periods.34 35 It assumes that individual differences in trajectories can be summarised by a finite set of polynomial functions of time. To model frailty criteria counts, we used the zero-inflated Poisson distribution, which is appropriate for count data and accommodates the scale minimum of zero. Time variable was defined according to follow-up time in years (1, 4, 7 and 10).

The identification of the optimal number of trajectory groups was based on the Bayesian information criterion and Bayes factors.34 Then, the shape of each group of trajectories (ie, linear, quadratic and cubic) was tested and maintained in the model when significant. Other supplemental criteria were also implemented for model selection, including (1) all trajectory groups encompassing a minimum of about 10% of the total sample, (2) the odds of correct classification based on the posterior probabilities of group membership being >5 for each group and (3) posterior probabilities of group membership being >0.7 for each group.34 Participants were assigned to the trajectory group for which they had the highest posterior probabilities of group membership.

Associated characteristics

In order to identify independent baseline factors significantly associated with trajectory group membership, a multivariable multinomial model including all characteristics was performed, taking the first group as the base outcome.

Model with dropout extension

As the missing at random assumption may generate biased estimates of trajectory group size, especially when individuals with the worst outcomes also tend to have higher missing rates, we performed an additional extended analysis to account for non-random attrition.36 37 While basic GBTM handles dropout as missing at random, this extension assumes attrition to be a dependent function of observed outcomes and unobserved trajectory group membership. In each follow-up wave, the dropout probability was modelled as a function of time and two prior observed outcomes using a logit distribution. It jointly estimated the trajectory of frailty phenotype and the probability of non-random dropping out due to death. Next, we compared the model with and without the dropout extension by checking the magnitude of change in the estimates, as well as in the group membership probabilities. Finally, we assessed whether there were differences in the baseline characteristics associated with the trajectory groups once non-random dropouts had been taken into account.

All analyses were performed using Stata/IC V.16.0. The GBTM was achieved using the TRAJ command,38 39 first without and then with the dropout option. Multivariable models were computed using the MLOGIT command.

Results

Overall, 3053 persons aged 65–70 years were included in the Lc65+ study in 2004 and 2009, of whom 3007 were invited to a baseline examination over the following year (online supplemental figure 1) after excluding 40 deaths, 2 end-of-life and 4 severe cognitive impairment situations. The frailty phenotype was assessed in 2783 participants at baseline. After excluding participants who were lost to follow-up for other reasons than death, the final sample included 2286 individuals (76.0% of invited persons) with a frailty assessment at baseline and with at least one follow-up measurement (online supplemental figure 1). Cumulative mortality was 4.4% in year 7 and 9.9% in year 10. The number of data points for frailty was 4 in 2011 cases (88.0%), 3 in 149 cases (6.5%) and 2 in 126 cases (5.5%).

Supplemental material

The dataset included 8743 observations of frailty status, of which 8645 (98.9%) resulted from measurement and 98 (1.1%) resulted from imputation due to institutionalisation.

Age at baseline ranged between 66.2 and 71.8 years (mean 68.9, SD 1.4). The baseline characteristics of the sample are presented in table 1. While frailty prevalence at baseline was low (<3%), about one in four participants was prefrail. The average frailty criteria count increased from 0.4 at baseline to 0.8 in year 10.

Table 1

Baseline characteristics of the study sample (N=2286)

Individuals excluded from the analyses (n=767, online supplemental table 1) were significantly older, more frequently male, current smokers, and reported less favourable ratings in terms of education, chronic conditions and depressive feelings (all p<0.05).

Frailty trajectories

The first stage of GBTM consisted in choosing the number of groups and shapes of trajectories to include in the model. The best fitting and most parsimonious model we identified was a three-group solution with two quadratic and one linear trajectories (table 2 and online supplemental table 2).

Table 2

Maximum likelihood estimates for frailty trajectories using a zero-inflated Poisson distribution (SEs in parentheses)

As shown in figure 1, cluster 1 (low-frailty trajectory, 50.2% of the sample) included participants who largely remained non-frail with minimal increase up to year 10. Cluster 2 (medium trajectory, 40.9%) started with a low risk of frailty and progressed to prefrailty with a score higher than 1 in year 10. Cluster 3 (high trajectory, 8.9%) started at prefrail level with a score close to 1.5 and showed a sharper increase towards frailty with a number of criteria close to 3 in year 10. More details about transition states according to group membership are presented in online supplemental table 3.

Figure 1

Estimated frailty trajectories over 10 years with 95% CIs (thin dotted line).

Associated factors

Bivariate associations between trajectory group membership and baseline characteristics are shown in table 3. Mean age and proportions of women, persons with a lower educational level, living alone, smoking, rating their health negatively, reporting a high number of chronic conditions, with obesity, with cognitive complaints, with depressive feelings, and prefrail or frail at baseline examination increased from cluster 1 to clusters 2 and 3. Social engagement decreased across clusters.

Table 3

Bivariate relationships between baseline characteristics and 10-year frailty trajectories (N=2286)

Table 4 displays the results of multivariable analyses. Status of (pre)frailty at baseline was strongly associated with clusters 2 and 3. Trajectories of increasing frailty were further associated with current smoking, obesity, negative self-rated health, a high number of chronic conditions and low social engagement. Age and cognitive complaints had an additional effect selectively on cluster 3.

Table 4

Multivariable relationships between baseline characteristics and 10-year frailty trajectories (N=2095)

Model with dropout extension

GBTM estimated the probability of cumulative dropout in year 7 at 2.0% in the low-frailty trajectory, compared with 4.1% in the medium trajectory and 19.7% in the high trajectory. In year 10, it was estimated at 2.1%, 4.8% and 22.2%, respectively.

The comparison between the basic model and the extended model that accounted for subjects’ non-random attrition showed that estimates and group membership probabilities for the three trajectory groups were almost identical (online supplemental table 4). A multivariable logistic regression showed that male sex (OR=1.79, p=0.001), former or current smoking (OR=1.75 and OR=2.95, respectively, p<0.001), negative self-rated health (OR=1.94, p<0.001) and baseline (pre)frailty status (OR=1.82, p=0.001) were significantly associated with dropout due to death.

Multivariable associations of baseline characteristics with trajectories were comparable with the basic model (see online supplemental table 5).

Discussion

In this Swiss cohort of community-dwelling adults aged 66–71 years at their first assessment, we studied clusters of frailty evolution and their associated factors in 10-year trajectories of frailty. In line with other studies,20 21 23 analyses identified three clusters corresponding to trajectories starting at different levels of frailty. According to our results, increasing frailty within the next decade concerns half of the persons at an age between 66 and 71 years, some of whom experience rapid progression.

To prevent frailty and its harmful consequences, understanding risk and protective factors is of major importance. The prefrailty or frailty status at baseline was the factor most strongly associated with the medium increasing and high increasing frailty trajectories. Previous papers exploring trajectories of Fried phenotype also reported major differences in the baseline level of frailty between trajectories.20–24 However, adjustment for the baseline frailty score was never applied to factors identified as associated with trajectories. Given the importance of frailty at baseline, reaching older age with optimal physiological resources appears to be a key factor for healthy ageing.

Negative self-ratings of health, the number of chronic conditions, smoking and obesity at baseline were additional risk factors for an adverse evolution, regardless of the presence of prefrailty or frailty at baseline, while social engagement was protective. Age had a selective effect on the least favourable trajectory. Few previous studies investigated risk factors for adverse trajectories of phenotypical frailty. Howrey et al20 found that trajectories of increasing frailty were related to increasing age, low education and selected diseases. Sex and obesity had no significant effect on trajectories, whereas church attendance was protective. Taniguchi et al reported specific chronic diseases (hypertension, cerebrovascular disease and diabetes) and poor ratings of subjective health as factors associated with an adverse evolution of the frailty phenotype.24 Peek et al23 examined factors associated with frailty within three identified trajectories. They highlighted a significant association between a high number of chronic conditions and increasing frailty. Their results also suggested a protective effect of social support on the evolution of frailty in a group characterised by progressive moderate frailty. Education was protective only in the progressive high frailty trajectory. In the present study, we observed a significant association between low education and poor frailty trajectories in bivariate analyses but not in a multivariable model in which frailty at baseline was the strongest associated factor. Hence, the protective effect of education on frailty trajectories previously suggested may occur mainly by reaching old age with low-frailty scores, but to a lesser extent through preventing frailty development in old age.

Although all deaths do not necessarily result from frailty, mortality after the age of 65 years is overwhelmingly due to degenerative conditions reducing physiological reserves such as cancer, and cardiovascular or cerebrovascular diseases. In Switzerland, less than 4% of deaths between 65 and 84 years result from suicide or accident.40 Overlooking periods of transition to death could be particularly problematic while studying factors associated with frailty trajectories if mortality disproportionately affects some subgroups of the population, as is the case in men of middle–old age. Results have shown higher attrition rates in the least favourable trajectories, reflecting the fact that dropout is informative of group membership. Although attrition due to mortality was more frequent among men and in the presence of several vulnerability characteristics, adjustment for non-random dropout did not produce any meaningful differences in the main results. The relatively low attrition rate in our sample (less than 10%) can probably explain these findings. Indeed, Zimmer et al37 showed a significant contribution of dropout to the study of trajectories related to the numbers of limitations in activities of daily living. However, this analysis involved an older population with a cumulative mortality greater than 75% in year 7. Howrey et al used a similar approach in the study of frailty and cognitive decline trajectories with a mortality rate reaching 72% in year 18.20

Strengths and limitations

The strengths of this study are its longitudinal scope, the use of validated tools for assessing the frailty phenotype and the attempt to account for dropouts due to death. Furthermore, its focus on persons enrolled shortly after retirement age contributes to understanding the significance of Fried phenotype in the youngest of older adults. Our results, however, are limited to a community-dwelling population aged 66–71 years at inclusion. Other trajectories and risk factors may characterise older populations. Another limitation was the use of baseline measurements to characterise each trajectory group, while some characteristics may vary over time. The relationship observed between social engagement and trajectories may result from uncontrolled health and functional characteristics. Additional limitations include the observational nature of the study, the relatively large time intervals separating the repeated measurements of the frailty phenotype and the more vulnerable profile of participants excluded from the analyses.

Conclusions

Our results highlighted three main trajectories of the frailty phenotype in a population of community-dwelling individuals initially aged 66–71 years. The most unfavourable trajectories were associated with a higher risk of death. By showing that a higher number of frailty criteria at baseline are strongly associated with the most adverse trajectories, our study stresses the importance of assessing the frailty phenotype in the youngest–old and identifies target groups for the prevention of progressive frailty. Health promotion strategies should be encouraged at all life stages so as to reach old age with optimal physiological reserves and free from smoking and obesity. Finally, reinforcing social engagement early after retirement is an area of interest for future research.

What is already known on this subject

  • Since Fried et al have operationalised the frailty phenotype, several studies have described the transition between the different frailty states in older adults, but few of them have used repeated measurements to identify frailty trajectories and associated factors.

  • To date, no study has yet investigated trajectories based on the frailty phenotype in the young–old community-dwelling adults.

  • Additionally, despite the strong association between frailty and mortality, transitions to death were rarely taken into account in the study of frailty trajectories. Overlooking transition to death could be problematic in the study of factors associated with frailty trajectories, particularly when mortality disproportionately affects some subgroups of the population.

What this study adds

  • By taking transitions to death into account, this study includes the most vulnerable people in the analysis of the factors associated with an adverse evolution. Our results point to the significance of Fried phenotype at an age between 66 and 71 years as a chief factor associated with frailty trajectories, and the additional relationship with smoking and obesity. Thus, they support a strategy of frailty prevention through actions on adverse health behaviours over the life course and of (pre)frailty detection in the youngest–old individuals.

Data availability statement

Data are available upon reasonable request. The data that support the findings of this study are deidentified participant data. Details of the data and how to request access are available from the principal investigator of the Lc65+ study (Yves.Henchoz@unisante.ch) at the Centre for Primary Care and Public Health (Unisanté), University of Lausanne, Switzerland.

Ethics statements

Patient consent for publication

Ethics approval

The study protocol and informed consent were approved by the ethics committee for human research of the Canton Vaud (19/04).

Acknowledgments

We are deeply grateful to all Lausanne cohort 65+ participants and to the team members involved in data collection, entry and management.

References

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

Footnotes

  • Contributors BS-E created and conducted the Lausanne cohort 65+ project. BS-E, SF and YH designed the study; SF performed the data analysis; SF and BS-E drafted the manuscript. All authors interpreted the results, provided critical revisions and approved the final version of the manuscript.

  • Funding The Lc65+ study has been supported by the University of Lausanne Centre for Primary Care and Public Health (Unisanté); University of Lausanne Hospital Centre; Canton de Vaud Department of Public Health; City of Lausanne; Loterie Romande (research grants 2006–2008 and 2018–2019); Lausanne University Faculty of Biology and Medicine (multidisciplinary research grant 2006); Swiss National Foundation for Scientific Research (grant 3247B0-120795/1); and Fondation Médecine Sociale et Préventive, Lausanne. Sponsors did not intervene in the design, execution, analysis and interpretation of the data, nor in the writing of this study.

  • Competing interests None declared.

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

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.