Trajectories of frailty among Chinese older people in Hong Kong between 2001 and 2012: an age-period-cohort analysis

Trajectories of frailty among Chinese older people in Hong Kong between 2001 and 2012: an... Abstract Background there is little evidence to suggest that older people today are living in better health than their predecessors did at the same age. Only a few studies have evaluated whether there are birth cohort effects on frailty, an indicator of health in older people, encompassing physical, functional and mental health dimensions. Objectives this study examined longitudinal trajectories of frailty among Chinese older people in Hong Kong. Methods this study utilised data from the 18 Elderly Health Centres of the Department of Health comprising a total of 417,949 observations from 94,550 community-dwelling Chinese people aged ≥65 years in one early birth cohort (1901–23) and four later birth cohorts (1924–29, 1930–35, 1936–41, 1942–47) collected between 2001 and 2012, to examine trajectories of the frailty index and how birth cohorts may have contributed to the trends using an age-period-cohort analysis. Results more recent cohorts had higher levels of frailty than did earlier cohorts at the same age, controlling for period, gender, marital status, educational levels, socioeconomic status, lifestyle and social factors. Older age, being female, widowhood, lower education and smoking were associated with higher levels of frailty. Conclusion more recent cohorts had higher levels of frailty than did earlier cohorts. Frailty interventions, coupled with early detection, should be developed to combat the increasing rates of frailty in Hong Kong Chinese. frailty, trajectories, cohort, age-period-cohort, Chinese, older people Introduction Life expectancy at age 60 has increased substantially in recent decades, which has been attributed to declining mortality rates at older ages [1]. However, there is little evidence to suggest that older people today are living in better health than their predecessors did at the same age. Consequently, the question of whether the extension in life expectancy is accompanied by longer/shorter healthy life expectancy has gained more importance, with far-reaching consequences to the individual, society and health care policies. Many studies have examined the trends in mortality and morbidity, with the aim of determining whether subsequent cohorts are healthier than those that have preceded them as well as the need for future care resources [1–4]. Although trends in disability according to the activities of daily living (ADL) of the older population are generally positive [5], some reports support the opposite trend. For example, findings from two American studies showed that ADL disabilities were worse in recent cohorts than in earlier cohorts [6, 7]. A Chinese cohort study also showed that the recent cohorts scored significantly worse for physical and cognitive functioning than did their earlier counterparts [4]. Compared with trends in mortality from specific causes of death, morbidity or levels of chronic disease indicators, trends in frailty (i.e. impaired capacity to withstand intrinsic and environmental stressors) may offer a more comprehensive perspective of ageing, as frailty develops as a consequence of age-related decline in multiple physiological systems, encompassing physical, functional and mental health dimensions [8, 9]. Depending on the used definition [8, 10], the prevalence of frailty at age 65 years and older varies from 4.0 to 59.1% [11–13]. Although frailty is considered an important indicator of an individual’s capacity for independent living and his/her risk of suffering from an adverse event, only a few studies have evaluated whether there are birth cohort effects on frailty, and with conflicting results [14–16]. Furthermore, these studies had limited waves of longitudinal data and/or did not adjust for the effects of survey year (i.e. the year in which the surveys were carried out) to reflect changes in sociocultural, economic and environmental factors that may affect the entire population simultaneously. Although frailty is common in the older population in Hong Kong [13]; the trends in frailty in relation to cohort differences, in conjunction with evolving demographics, socioeconomic status, lifestyle and social factors, have not been studied. A better understanding of the longitudinal trajectories of frailty is important in predicting an individual’s health status at different stages of demographic transitions in Chinese populations, which are ethnically and culturally distinct from Caucasians and have different health and social care systems. Using data from the 18 Elderly Health Centres (EHCs, hereafter) of the Department of Health of Hong Kong, a longitudinal sample following multiple Chinese birth cohorts with different entry years during the period from 2001 to 2012 was analysed to describe the cohort effect on frailty among Chinese older people and to assess the modification of this effect by age, period, gender, marital status, educational levels, socioeconomic status, lifestyle and social factors using an age-period-cohort analysis. Methods Study sample We used data from the 18 EHCs of the Department of Health, which have been collecting longitudinal health data from a large population-based cohort in Hong Kong since 1998. All residents of Hong Kong aged 65 years and older can voluntarily enrol. Enrollees in the cohort receive standard medical examinations at baseline and are reassessed in subsequent years. The details of this cohort have been described elsewhere [17]. In this study, we retrieved longitudinal data from the EHCs of individuals who enroled between 2001 and 2012 (n = 427,790), and excluded those who were institutionalised (n = 9,490, 2.2%), those with missing data on living arrangement (n = 6, 0.0%), and those with data with less than 27 of the 30 frailty deficits (n = 345, 0.1%). Of these observations, we further cleaned the data with logical errors. Although it has been suggested that a minimum of 30 deficits is required to define frailty [18], a previous analysis shows no difference between using the full set of 32 deficits and a reduced set of 26 deficits when it comes to define frailty [19]. The study has been approved by the ethics committee of the Department of Health of the Government of the Hong Kong Special Administrative Region. Variable definitions Frailty index The multiple-deficits approach was used to construct the Frailty Index (FI), which consisted of 30 components covering chronic disease history and medication use, physical and cognitive functioning, psychological wellbeing, and geriatric symptoms, with a score of 1 representing a deficit for each component (Supplementary Table). The FI was calculated as the proportion of the number of deficits for an individual to the maximum total number of deficits. The deficits included in the FI satisfied the criteria as previously described [18]. All deficits of FI were associated with the health status of the individual, had less than 5% missing values, presented in at least 1% of the study population, and the prevalence of each deficit generally increased with age. In addition, the FI has a strong association with mortality. However, the risks for mortality vary across birth cohorts (data not shown). It is possible that the follow-up time may not be long enough to complete survival, particularly among the recent cohorts. Age and cohort Age was calculated by subtracting the survey year from the birth year. It was transformed by subtracting the value from the grand median and further being divided by 10 for mathematical convenience. The birth cohorts were categorised into the years 1901–23 (reference group), 1924–29, 1930–35, 1936–41 and 1942–47. Covariates Gender and the following time-varying covariates were included in the analysis: marital status, educational attainments, employment, housing, recipient of social assistance, smoking, regular alcohol consumption (drink at least once per week), regular exercise (exercise at least once per week) and participation in social activities, including volunteering and participating in leisure activities. Statistical analysis Descriptive statistics (mean (SD) or %) were used for summarising the primary FI outcome and other explanatory variable stratified by birth cohort. The repeated measurement data were analysed by a two-level hierarchical linear model (HLM) [14, 19]. In HLM, random effects of individual and time are able to capture the heterogeneity of baseline data and the changes in FI over time. It is also a modelling technique used to handle missing data and unequal time spaces which are typical in longitudinal observation studies [20]. In our study, the model is able to depict the change in FI with age for each individual (level 1) and the heterogeneity due to differences in gender and birth cohort amongst our study population (level 2). The model is specified as follows:   Level1Model:FIit=β0i+β1iAgeit+β2iAgeit2+∑jαjiZj,it+εit (1)  Level2Model for intercept:β0i=γ00+γ01Genderi+γ02Cohorti+u0i (2)  Level2Model for growth rate:β1i=γ10+γ11Genderi+γ12Cohorti+u1i (3) The level 1 model (equation 1) describes the change of FI of an individual i at time t ( FIit) given the population mean ( β0i), linear ( β1i) and quadratic effects ( β2i) of age, after controlling for the effects ( αji) of other demographics variables (i.e., marital status and education level), socioeconomic status, lifestyle and social factors that vary over time ( Zj,it, j = 1, 2,…). εit is the random error within individual i over time. The level 2 model for intercept describes the differences between individuals and is modelled as a function of gender, birth cohort, and their interaction. The model for the growth rate describes the linear rate of change with age from the variables of gender and cohort. The coefficients γ00−γ02andγ10−γ12 are the effects on mean FI and linear growth rate with age in FI, respectively. The age-by-cohort interaction is adopted as a proxy of period effect [21, 22]. The gender-by-cohort interaction was not included in the model due to insignificant effects (result not shown). The Model 1 of HLM was fitted using age, gender and cohorts only. On top of these variables, the Model 2 of HLM was fitted using other demographics variables (i.e., marital status and education level), socioeconomic status, lifestyle and social factors. The goodness of fit of the models were assessed by the 95% credible interval of residual errors, mean absolute error (MAE and mean absolute percentage error (MAPE). The HLM models were fitted using the HPMIXED procedure in SAS 9.4 software. Results The total number of observations was 417,949, which included 94,550 participants (female 64.3%). Table 1 shows the characteristics of the study population by birth cohort. The mean age of cohort at the time of study was 72 years and the mean FI was 0.14. The mean age of the study population decreased from 82 to 67 years across successive cohorts, indicating the presence of confounding by age on trends in frailty and providing justification for subsequent APC analyses. Table 1. Descriptive statistics for the study samples Variablesa  Birth cohort  All (n = 94,550)  1901–23 (n = 9,783)  1924–29 (n = 20,561)  1930–35 (n = 34,646)  1936–41 (n = 20,316)  1942–47 (n = 9,244)  Frailty index (mean ± SD)  0.14 ± 0.07  0.18 ± 0.07  0.16 ± 0.07  0.14 ± 0.07  0.13 ± 0.07  0.13 ± 0.07  Demographics   Age, years (mean ± SD)  72 ± 5.0  82 ± 3.4  75 ± 2.6  70 ± 3.2  69 ± 2.6  67 ± 1.2   Female (%)  64.3  63.7  62.5  62.4  65.8  72.5   Marital status (%)    Married  64.4  41.8  55.7  68.7  72.3  74.4    Never married  2.7  3.7  3.0  2.4  2.5  3.0    Widowed  30.1  52.9  39.2  26.2  21.9  18.2    Separated/divorced/others  2.7  1.6  2.1  2.6  3.2  4.4   Educational levels (%)    No education  35.4  47.2  45.6  38.5  23.3  15.1    Primary  38.3  36.1  37.7  38.9  37.9  41.2    Secondary  19.3  12.5  13.4  17.1  26.3  32.9    Post-secondary  6.9  4.3  3.4  5.5  12.5  10.8  Socioeconomic status   Employment (%)    Unemployed/retired/others  95.2  98.5  96.8  94.1  94.7  93.4    Working full-time  2.7  0.9  1.8  3.4  3.0  3.2    Working part-time  2.1  0.7  1.5  2.5  2.3  3.4   Housing (%)    Public and subsidised  43.9  41.4  40.6  41.6  48.4  52.5    Private  54.1  55.6  56.6  56.3  50.5  46.6    Temporary/others  2.1  3.0  2.9  2.1  1.1  0.9   Recipient of social assistance (%)  12.1  24.0  15.5  10.5  8.1  6.3  Lifestyle   Smoking (%)    Never-smoker  73.1  64.4  67.8  72.9  78.6  83.4    Ex-smoker  19.7  28.4  24.1  19.0  15.8  11.5    Non-daily smoker  0.5  0.1  0.2  0.3  1.0  1.6    Daily smoker  6.7  7.2  8.0  7.8  4.6  3.6   Regular drinker (%)  3.8  2.9  3.8  4.4  4.1  2.3   Regular exercise (%)  90.4  89.0  90.4  89.7  91.9  91.7  Social support/social factor   Participation in social activities (%)  69.6  65.4  63.7  61.7  81.6  90.6  Variablesa  Birth cohort  All (n = 94,550)  1901–23 (n = 9,783)  1924–29 (n = 20,561)  1930–35 (n = 34,646)  1936–41 (n = 20,316)  1942–47 (n = 9,244)  Frailty index (mean ± SD)  0.14 ± 0.07  0.18 ± 0.07  0.16 ± 0.07  0.14 ± 0.07  0.13 ± 0.07  0.13 ± 0.07  Demographics   Age, years (mean ± SD)  72 ± 5.0  82 ± 3.4  75 ± 2.6  70 ± 3.2  69 ± 2.6  67 ± 1.2   Female (%)  64.3  63.7  62.5  62.4  65.8  72.5   Marital status (%)    Married  64.4  41.8  55.7  68.7  72.3  74.4    Never married  2.7  3.7  3.0  2.4  2.5  3.0    Widowed  30.1  52.9  39.2  26.2  21.9  18.2    Separated/divorced/others  2.7  1.6  2.1  2.6  3.2  4.4   Educational levels (%)    No education  35.4  47.2  45.6  38.5  23.3  15.1    Primary  38.3  36.1  37.7  38.9  37.9  41.2    Secondary  19.3  12.5  13.4  17.1  26.3  32.9    Post-secondary  6.9  4.3  3.4  5.5  12.5  10.8  Socioeconomic status   Employment (%)    Unemployed/retired/others  95.2  98.5  96.8  94.1  94.7  93.4    Working full-time  2.7  0.9  1.8  3.4  3.0  3.2    Working part-time  2.1  0.7  1.5  2.5  2.3  3.4   Housing (%)    Public and subsidised  43.9  41.4  40.6  41.6  48.4  52.5    Private  54.1  55.6  56.6  56.3  50.5  46.6    Temporary/others  2.1  3.0  2.9  2.1  1.1  0.9   Recipient of social assistance (%)  12.1  24.0  15.5  10.5  8.1  6.3  Lifestyle   Smoking (%)    Never-smoker  73.1  64.4  67.8  72.9  78.6  83.4    Ex-smoker  19.7  28.4  24.1  19.0  15.8  11.5    Non-daily smoker  0.5  0.1  0.2  0.3  1.0  1.6    Daily smoker  6.7  7.2  8.0  7.8  4.6  3.6   Regular drinker (%)  3.8  2.9  3.8  4.4  4.1  2.3   Regular exercise (%)  90.4  89.0  90.4  89.7  91.9  91.7  Social support/social factor   Participation in social activities (%)  69.6  65.4  63.7  61.7  81.6  90.6  aBaseline records were used for time-varying covariates. Table 2 shows the model estimates for all cohorts. Quadratic age trajectories of the FI were found, indicating accelerated increases in the accumulation of health deficit with age. Controlling for the effect of age, positive cohort effects on FI were observed. An interaction between birth cohort and age was observed, suggesting the existence of a cohort effect independent of age and period effects (Model 2). The HLM shows acceptable goodness of fit in which the 95% credible interval, MAE and MAPE of the full model are (−0.05 to 0.06), 0.026 and 22%, respectively. Table 2. Parameter estimates (95% confidence intervals) from HLM   Model 1  Model 2  Demographics   Age  0.0460 (0.0429, 0.0490)  0.0449 (0.0418, 0.0481)   Age2  0.0130 (0.0119, 0.0141)  0.0127 (0.0116, 0.0138)   Female  0.0234 (0.0225, 0.0242)  0.0234 (0.0224, 0.0244)   Age*female  −0.0012 (−0.0023, 0.0000)  −0.0011 (−0.0024, −0.0001)   Cohorts (Reference: 1901–1923)    1924–29  0.0162 (0.0140, 0.0184)  0.0164 (0.0142, 0.0186)    1930–1935  0.0243 (0.0221, 0.0265)  0.0257 (0.0235, 0.0279)    1936–41  0.0293 (0.0271, 0.0315)  0.0321 (0.0298, 0.0344)    1942–1947  0.0502 (0.0448, 0.0556)  0.0533 (0.0479, 0.0587)   Age*cohorts    Age*1924–29  0.0071 (0.0044, 0.0097)  0.0082 (0.0055, 0.0109)    Age*1930–1935  0.0136 (0.0104, 0.0169)  0.0148 (0.0115, 0.0182)    Age*1936–1941  0.0268 (0.0228, 0.0308)  0.0277 (0.0237, 0.0318)    Age*1942–47  0.0490 (0.0411, 0.0569)  0.0495 (0.0415, 0.0574)   Marital status (reference: married)    Never married    −0.0005 (−0.0027, 0.0017)    Widowed, separated/divorced, or others    0.0015 (0.0009, 0.0022)   Educational levels (reference: no education)    Primary    −0.0012 (−0.0018, −0.0006)    Secondary or above    −0.0029 (−0.0038, −0.0021)  Socioeconomic status   Working full-time or part-time    −0.0051 (−0.0062, −0.0039)   Private housing    −0.0005 (−0.0010, 0.0001)   Recipient of social assistance    0.0082 (0.0074, 0.0091)  Lifestyle   Smoker    0.0036 (0.0027, 0.0045)   Regular drinker    −0.0022 (−0.0033, −0.0010)   Regular exercise    −0.0047 (−0.0054, −0.0040)  Social support/social factor   Participation in social activities    −0.0023 (−0.0027, −0.0019)    Model 1  Model 2  Demographics   Age  0.0460 (0.0429, 0.0490)  0.0449 (0.0418, 0.0481)   Age2  0.0130 (0.0119, 0.0141)  0.0127 (0.0116, 0.0138)   Female  0.0234 (0.0225, 0.0242)  0.0234 (0.0224, 0.0244)   Age*female  −0.0012 (−0.0023, 0.0000)  −0.0011 (−0.0024, −0.0001)   Cohorts (Reference: 1901–1923)    1924–29  0.0162 (0.0140, 0.0184)  0.0164 (0.0142, 0.0186)    1930–1935  0.0243 (0.0221, 0.0265)  0.0257 (0.0235, 0.0279)    1936–41  0.0293 (0.0271, 0.0315)  0.0321 (0.0298, 0.0344)    1942–1947  0.0502 (0.0448, 0.0556)  0.0533 (0.0479, 0.0587)   Age*cohorts    Age*1924–29  0.0071 (0.0044, 0.0097)  0.0082 (0.0055, 0.0109)    Age*1930–1935  0.0136 (0.0104, 0.0169)  0.0148 (0.0115, 0.0182)    Age*1936–1941  0.0268 (0.0228, 0.0308)  0.0277 (0.0237, 0.0318)    Age*1942–47  0.0490 (0.0411, 0.0569)  0.0495 (0.0415, 0.0574)   Marital status (reference: married)    Never married    −0.0005 (−0.0027, 0.0017)    Widowed, separated/divorced, or others    0.0015 (0.0009, 0.0022)   Educational levels (reference: no education)    Primary    −0.0012 (−0.0018, −0.0006)    Secondary or above    −0.0029 (−0.0038, −0.0021)  Socioeconomic status   Working full-time or part-time    −0.0051 (−0.0062, −0.0039)   Private housing    −0.0005 (−0.0010, 0.0001)   Recipient of social assistance    0.0082 (0.0074, 0.0091)  Lifestyle   Smoker    0.0036 (0.0027, 0.0045)   Regular drinker    −0.0022 (−0.0033, −0.0010)   Regular exercise    −0.0047 (−0.0054, −0.0040)  Social support/social factor   Participation in social activities    −0.0023 (−0.0027, −0.0019)  To compare cohort differences at the same age, we compared predicted age trajectories of the FI across birth cohorts (Figure 1). We found that more recent cohorts had higher levels of frailty than did earlier cohorts at the same age, controlling for demographics, socioeconomic status, lifestyle and social factors. We also found that, on average, being female was associated with 0.02 units increase in FI, after adjusting for age and cohort effects (Table 2, Model 1). At the same age, recent cohorts were at a higher risk of having an increased FI that did earlier cohorts in both men and women (Supplementary Figure). Figure 1. View largeDownload slide Age trajectories of the frailty index (FI) with 95% confidence intervals of the mean FI (red lines). Figure 1. View largeDownload slide Age trajectories of the frailty index (FI) with 95% confidence intervals of the mean FI (red lines). Table 2 (Model 2) also shows the associations of demographics, socioeconomic status, lifestyle and social factors with frailty. Individuals who were widowed had a 0.002 unit higher FI on average. Completing one’s secondary education or higher decreased the FI by 0.003. Working full-time/part-time similarly decreased the FI by 0.005. Recipients of social assistance had a 0.008 unit higher FI on average. Regular drinking and regular exercise were associated with a lower risk of frailty. Smokers including ex-smokers, daily and non-daily smokers, had an increased FI of 0.004 units. Participation in social activities was associated with a lower risk of frailty. The inclusion of these predictors did not attenuate the cohort effect. Discussion Using a large longitudinal dataset of community-dwelling Chinese people aged ≥65 years, we found that more recent cohorts had higher levels of frailty than did earlier cohorts, after adjusting for age, period, gender, marital status, educational levels, socioeconomic status, lifestyle and social factors. This finding indicates that the increase in life expectancy has negative implications, as it is associated with concurrent increases in levels of frailty. This may lead to greater costs for medical care, social services and long-term care. The results from the current study are consistent with a previous British analysis, which also reported higher levels of frailty in recent cohorts compared with earlier cohorts (for cohorts aged over 70 in 2002) [16]. An US-based study examining cohort-specific trajectories of frailty also reported higher levels of frailty in recent cohorts compared with earlier cohorts [14]. Similar findings were also observed in the US for cohort-specific trajectories of self-reported illness [23, 24]. In contrast to these findings, a Sweden-based study examining cohort-specific trajectories of frailty reported higher levels of frailty in earlier cohorts compared with recent cohorts [15]. The discrepancy of the findings may reflect different social, cultural and health policies at different stages of demographic transitions [25]. The use of different types of data (repeated cross-sectional or longitudinal) and definitions of frailty across studies can also influence results. In particular, some previous studies did not control for period effects, which may significantly cofound the cohort effects. Although the mechanisms that cause frailty to increase amongst more recent birth cohorts in our study population are not well understood, one possibility is that frail individuals in more recent cohorts had a better chance of surviving into old age than their predecessors, most likely due to improvements in living conditions and medical treatments over their lifetimes. These improvements imply that older people in more recent cohorts are expected to live longer, but the improved survival rates of these people may be associated with an increase in chronic diseases and impaired physical and cognitive functioning (as observed in a Chinese cohort) [4], which, in turn, increase the risk of frailty. Longitudinal trajectories of lifestyle and social factors due to demographic changes may also partly explain the differences in frailty risk across successive birth cohorts. For example, the economic reforms and high-tech revolution that have taken place over the past few decades in Hong Kong may have exposed individuals in recent cohorts to sedentary occupations (non-agriculture-/non-manufacturing-related jobs) that require less daily energy expenditure than earlier cohorts. This has been accompanied by a decline in functional capacity and health, which might have promoted frailty. In addition, the establishment and the mobility of nuclear families, as reflected by the trends in declining household size and the rising number of older people living alone [26, 27], might have had adverse impacts on the social networks/connections of older people, which may accelerate frailty. Findings from a Mexican–American study showed that social support was protective against worsening in frailty among individuals with moderate frailty [28]. Analyses of EHCs’ data also demonstrated that older people who had participated in social activities had lower levels of frailty than those who did not; hence, the findings of this study together with current available literature lend support to the concept that social environment may be an especially important mechanism for understanding the trajectory patterns of frailty. In this study, the cohort effect on frailty has been adjusted for multiple covariates. Our findings reaffirm the importance of education, healthy lifestyle (regular exercise, not smoking), social support and social participation in slowing down the onset of frailty. Nevertheless, consistent with the results reported in the Women’s Health Initiative Observational Study, the Seniors-ENRICA cohort, and the Health and Retirement Study, where moderate drinkers were associated with a lower risk of frailty [29–31], our data demonstrated that regular alcohol consumption was associated with a lower risk of frailty, although it is well known that alcohol in excess is carcinogenic and detrimental to health [32]. The inverse association between alcohol intake and frailty may be due to the anti-inflammatory effect associated with low-to-moderate alcohol consumption [31]; however, it could also be due to the abstainer/quitter bias [33], as individuals in poor health, particularly older people, drink less than those individuals in good health. Several limitations of the current study require discussion. The earliest and latest cohorts did not capture a full age distribution, thus biasing the estimates for cohort trends. Individuals born between 1901 and 1923 were merged due to the comparatively small size of several age groups, thus limiting the number of birth cohorts being studied. Also, information on risk factors in early- and mid-life (e.g. changing dietary pattern) that may affect frailty at older ages was not available, limiting the ability to identify causes of increased frailty in late life. Another limitation is that study participation was voluntary which could result in selection bias. Compared to individuals who responded to follow-up assessments, those who did not respond to any follow-up assessment between 2001 and 2012 were older and had lower levels of education. Nevertheless, HLM was used to capture the heterogeneity of baseline data and the changes over time, which can yield reliable results despite the presence of missing data. In the HLM model, age is allowed to vary by cohort groups with the uses of random effects. By assuming independently and identically distributed residuals at the higher-level, the period effects as well as the cohort effects were distributed independently. We also followed the framework of the hierarchical age-period-cohort (HAPC) regression model from Yang and Land [34], adding a quadratic term of age to prevent the problem of under-identification among APC. Nevertheless, as demonstrated by Bell and Jones [35], the period or cohort trend could not be well-converted into a separate random effect if the independence assumption cannot be held. As shown in a simulation scenario in Bell and Jones’ paper, the cohort and age effects could be over-estimated if there is a priori linear trend of period effect. Finally, the design of the study is subject to survival bias as some people who were frail may not have survived to the age to be included in the study. This may have led to underestimations of the levels of frailty. Therefore, the study population might not completely represent the whole geriatric population in Hong Kong. The results of this study should be interpreted with cautions. The strengths of this study include the large sample size and the adjustment of multiple potential confounders. In addition, the FI obtained using the described methods were in line with age-specific/overall FI obtained in other studies, including the English Longitudinal Study of Ageing [16] and the Health and Retirement Survey [14]. The similarity of our estimates for the FI with those of independent studies underpins the validity of our calculations for FI at population level. In conclusion, our data demonstrate that more recent cohorts had higher levels of frailty than did earlier cohorts amongst both men and women. Therefore, frailty interventions, coupled with early detection, should be developed to combat the increasing rates of frailty in Hong Kong Chinese, which might have far-reaching benefits for individuals and society by preserving function into old age, increasing the number of years spent in good health at older ages, and decreasing health care costs. The inverse association between social participation and frailty observed in the present study also implies the need for encouraging the social participation of older people. At the same time, the surveillance of changes in frailty of older people will also be of fundamental importance to the planning of health care and the resources needed. Our results can also be used to predict future health trends and the public health burden of the older population. Key points More recent cohorts had higher levels of frailty than did earlier cohorts in both men and women. The increase in life expectancy has negative implications, as it is associated with concurrent increases in levels of frailty. Frailty interventions, coupled with early detection, should be developed to combat the increasing rates of frailty in Hong Kong. Supplementary Data Supplementary data mentioned in the text are available to subscribers in Age and Ageing online. Conflict of interest None. Funding None. Acknowledgements We thank the cohort members of the Elderly Health Centres. We thank the investigators, research associates and team members for the design, collection, collation, validation and management of the data used in this article. References 1 Crimmins EM, Beltran-Sanchez H. Mortality and morbidity trends: is there compression of morbidity? J Gerontol B Psychol Sci Soc Sci  2011; 66B: 75– 86. Google Scholar CrossRef Search ADS   2 GBD 2015 Disease and Injury Incidence and Prevalence Collaborators. Global, regional, and national incidence, prevalence, and years lived with disability for 310 diseases and injuries, 1990–2015: a systematic analysis for the Global Burden of Disease Study 2015. Lancet  2016; 388: 1545– 1602. CrossRef Search ADS PubMed  3 Yu R, Wong M, Chang B et al.  . Trends in activities of daily living disability in a large sample of community-dwelling Chinese older adults in Hong Kong: an age-period-cohort analysis. BMJ Open  2016; 6: e013259. Google Scholar CrossRef Search ADS PubMed  4 Zeng Y, Feng Q, Hesketh T et al.  . Survival, disabilities in activities of daily living, and physical and cognitive functioning among the oldest-old in China: a cohort study. Lancet  2017; 389: 1619– 29. Google Scholar CrossRef Search ADS PubMed  5 Freedman VA, Martin LG, Schoeni RF. Recent trends in disability and functioning among older adults in the United States: a systematic review. J Am Med Assoc  2003; 289: 3242– 2. Google Scholar CrossRef Search ADS   6 Seeman TE, Merkin SS, Crimmins EM et al.  . Disability trends among older Americans: National Health and Nutrition Examination Surveys, 1988–1994 and 1999–2004. Am J Public Health  2010; 100: 100– 7. Google Scholar CrossRef Search ADS PubMed  7 Lin SF, Beck AN, Finch BK et al.  . Trends in U.S. older adult disability: exploring age, period, and cohort effects. Am J Public Health  2012; 102: 2157– 63. Google Scholar CrossRef Search ADS PubMed  8 Fried LP, Tangen CM, Walston J et al.  . Frailty in older adults: evidence for a phenotype. J Gerontol A Biol Sci Med Sci  2001; 56: M146– 56. Google Scholar CrossRef Search ADS PubMed  9 Morley JE, Vellas B, van Kan GA et al.  . Frailty consensus: a call to action. J Am Med Dir Assoc  2013; 14: 392– 7. Google Scholar CrossRef Search ADS PubMed  10 Rockwood K, Song X, MacKnight C et al.  . A global clinical measure of fitness and frailty in elderly people. Can Med Assoc J  2005; 173: 489– 95. Google Scholar CrossRef Search ADS   11 Santos-Eggimann B, Cuenoud P, Spagnoli J et al.  . Prevalence of frailty in middle-aged and older community-dwelling Europeans living in 10 countries. J Gerontol A Biol Sci Med Sci  2009; 64: 675– 81. Google Scholar CrossRef Search ADS PubMed  12 Collard RM, Boter H, Schoevers RA et al.  . Prevalence of frailty in community-dwelling older persons: a systematic review. J Am Geriatr Soc  2012; 60: 1487– 92. Google Scholar CrossRef Search ADS PubMed  13 Woo J, Yu R, Wong M et al.  . Frailty screening in the community using the FRAIL scale. J Am Med Dir Assoc  2015; 16: 412– 9. Google Scholar CrossRef Search ADS PubMed  14 Yang Y, Lee LC. Dynamics and heterogeneity in the process of human frailty and aging: evidence from the U.S. older adult population. J Gerontol B Psychol Sci Soc Sci  2009; 65B: 246– 55. Google Scholar CrossRef Search ADS PubMed  15 Horder H, Skoog I, Johansson L et al.  . Secular trends in frailty: a comparative study of 75-year olds born in 1911–12 and 1930. Age Ageing  2015; 44: 817– 22. Google Scholar CrossRef Search ADS PubMed  16 Marshall A, Nazroo J, Tampubolon G et al.  . Cohort differences in the levels and trajectories of frailty among older people in England. J Epidemiol Community Health  2015; 69: 316– 21. Google Scholar CrossRef Search ADS PubMed  17 Schooling CM, Chan WM, Leung SL et al.  . Cohort profile: Hong Kong Department of Health Elderly Health Service Cohort. Int J Epidemiol  2016; 45: 64– 72. Google Scholar CrossRef Search ADS PubMed  18 Searle SD, Mitnitski A, Gahbauer EA et al.  . A standard procedure for creating a frailty index. BMC Geriatr  2008; 8: 24. Google Scholar CrossRef Search ADS PubMed  19 Kulminski A, Yashin A, Ukraintseva S et al.  . Accumulation of health disorders as a systemic measure of aging: findings from the NLTCS data. Mech Ageing Dev  2006; 127: 840– 8. Google Scholar CrossRef Search ADS PubMed  20 Raudenbush SW, Chan WS. Growth curve analysis in accelerated longitudinal designs. J Res Crime Delinq  1992; 29: 387– 411. Google Scholar CrossRef Search ADS   21 Jaacks LM, Gordon-Larsen P, Mayer-Davis EJ et al.  . Age, period and cohort effects on adult body mass index and overweight from 1991 to 2009 in China: the China Health and Nutrition Survey. Int J Epidemiol  2013; 42: 828– 37. Google Scholar CrossRef Search ADS PubMed  22 Yang YC, Land KC. Review of Age-Period-Cohort analysis: new models, methods, and empirical applications, by Yang YC and Land KC, J Am Stat Assoc, 109, 865: Comment by Fienberg, Hodges, and Luo and Replies REPLY. J Am Stat Assoc  2015; 110: 457– 7. Google Scholar CrossRef Search ADS   23 Mirowsky J, Ross CE. Education and self-rated health—cumulative advantage and its rising importance. Res Ageing  2008; 30: 93– 122. Google Scholar CrossRef Search ADS   24 Yang Y, Lee LC. Sex and race disparities in health: cohort variations in life course patterns. Soc Forces  2009; 87: 2093– 2124. Google Scholar CrossRef Search ADS   25 Stolz E, Mayerl H, Waxenegger A et al.  . Impact of socioeconomic position on frailty trajectories in 10 European countries: evidence from the Survey of Health, Ageing and Retirement in Europe (2004–2013). J Epidemiol Community Health  2017; 71: 73– 80. Google Scholar CrossRef Search ADS PubMed  26 Census and Statistics Department, HKSAR. The profile of the population in one-person households, 2011. 27 Census and Statistics Department, HKSAR. Thematic Report: Older Persons, 2011. 28 Peek MK, Howrey BT, Ternent RS et al.  . Social support, stressors, and frailty among older Mexican American adults. J Gerontol B Psychol Sci Soc Sci  2012; 67: 755– 64. Google Scholar CrossRef Search ADS PubMed  29 Woods NF, LaCroix AZ, Gray SL et al.  . Frailty: emergence and consequences in women aged 65 and older in the women’s health initiative observational study. J Am Geriatr Soc  2005; 53: 1321– 30. Google Scholar CrossRef Search ADS PubMed  30 Ortola R, Garcia-Esquinas E, Leon-Munoz LM et al.  . Patterns of alcohol consumption and risk of frailty in community-dwelling older adults. J Gerontol A Biol Sci Med Sci  2016; 71: 251– 8. Google Scholar CrossRef Search ADS PubMed  31 Shah M, Paulson D. C-reactive protein partially mediates the relationship between moderate alcohol use and frailty. Gerontologist  2016; 56: 318– 8. Google Scholar PubMed  32 Turati F, Galeone C, Rota M et al.  . Alcohol and liver cancer: a systematic review and meta-analysis of prospective studies. Ann Oncol  2014; 25: 1526– 35. Google Scholar CrossRef Search ADS PubMed  33 Rehm J, Irving H, Ye Y et al.  . Are lifetime abstainers the best control group in alcohol epidemiology? On the stability and validity of reported lifetime abstention. Am J Epidemiol  2008; 168: 866– 71. Google Scholar CrossRef Search ADS PubMed  34 Yang Y, Land KC. A mixed models approach to the age-period-cohort analysis of repeated cross-section surveys, with an application to data on trends in verbal test scores. Sociol Methodol  2006; 36: 75– 97. Google Scholar CrossRef Search ADS   35 Bell A, Jones K. Another’futile quest’? A simulation study of Yang and Land’s Hierarchical Age-Period-Cohort model. Demogr Res  2014; 30: 333. Google Scholar CrossRef Search ADS   © The Author 2017. Published by Oxford University Press on behalf of the British Geriatrics Society.All rights reserved. For permissions, please email: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Age and Ageing Oxford University Press

Trajectories of frailty among Chinese older people in Hong Kong between 2001 and 2012: an age-period-cohort analysis

Loading next page...
 
/lp/ou_press/trajectories-of-frailty-among-chinese-older-people-in-hong-kong-hROgf0a9Ta
Publisher
Oxford University Press
Copyright
© The Author 2017. Published by Oxford University Press on behalf of the British Geriatrics Society.All rights reserved. For permissions, please email: journals.permissions@oup.com
ISSN
0002-0729
eISSN
1468-2834
D.O.I.
10.1093/ageing/afx170
Publisher site
See Article on Publisher Site

Abstract

Abstract Background there is little evidence to suggest that older people today are living in better health than their predecessors did at the same age. Only a few studies have evaluated whether there are birth cohort effects on frailty, an indicator of health in older people, encompassing physical, functional and mental health dimensions. Objectives this study examined longitudinal trajectories of frailty among Chinese older people in Hong Kong. Methods this study utilised data from the 18 Elderly Health Centres of the Department of Health comprising a total of 417,949 observations from 94,550 community-dwelling Chinese people aged ≥65 years in one early birth cohort (1901–23) and four later birth cohorts (1924–29, 1930–35, 1936–41, 1942–47) collected between 2001 and 2012, to examine trajectories of the frailty index and how birth cohorts may have contributed to the trends using an age-period-cohort analysis. Results more recent cohorts had higher levels of frailty than did earlier cohorts at the same age, controlling for period, gender, marital status, educational levels, socioeconomic status, lifestyle and social factors. Older age, being female, widowhood, lower education and smoking were associated with higher levels of frailty. Conclusion more recent cohorts had higher levels of frailty than did earlier cohorts. Frailty interventions, coupled with early detection, should be developed to combat the increasing rates of frailty in Hong Kong Chinese. frailty, trajectories, cohort, age-period-cohort, Chinese, older people Introduction Life expectancy at age 60 has increased substantially in recent decades, which has been attributed to declining mortality rates at older ages [1]. However, there is little evidence to suggest that older people today are living in better health than their predecessors did at the same age. Consequently, the question of whether the extension in life expectancy is accompanied by longer/shorter healthy life expectancy has gained more importance, with far-reaching consequences to the individual, society and health care policies. Many studies have examined the trends in mortality and morbidity, with the aim of determining whether subsequent cohorts are healthier than those that have preceded them as well as the need for future care resources [1–4]. Although trends in disability according to the activities of daily living (ADL) of the older population are generally positive [5], some reports support the opposite trend. For example, findings from two American studies showed that ADL disabilities were worse in recent cohorts than in earlier cohorts [6, 7]. A Chinese cohort study also showed that the recent cohorts scored significantly worse for physical and cognitive functioning than did their earlier counterparts [4]. Compared with trends in mortality from specific causes of death, morbidity or levels of chronic disease indicators, trends in frailty (i.e. impaired capacity to withstand intrinsic and environmental stressors) may offer a more comprehensive perspective of ageing, as frailty develops as a consequence of age-related decline in multiple physiological systems, encompassing physical, functional and mental health dimensions [8, 9]. Depending on the used definition [8, 10], the prevalence of frailty at age 65 years and older varies from 4.0 to 59.1% [11–13]. Although frailty is considered an important indicator of an individual’s capacity for independent living and his/her risk of suffering from an adverse event, only a few studies have evaluated whether there are birth cohort effects on frailty, and with conflicting results [14–16]. Furthermore, these studies had limited waves of longitudinal data and/or did not adjust for the effects of survey year (i.e. the year in which the surveys were carried out) to reflect changes in sociocultural, economic and environmental factors that may affect the entire population simultaneously. Although frailty is common in the older population in Hong Kong [13]; the trends in frailty in relation to cohort differences, in conjunction with evolving demographics, socioeconomic status, lifestyle and social factors, have not been studied. A better understanding of the longitudinal trajectories of frailty is important in predicting an individual’s health status at different stages of demographic transitions in Chinese populations, which are ethnically and culturally distinct from Caucasians and have different health and social care systems. Using data from the 18 Elderly Health Centres (EHCs, hereafter) of the Department of Health of Hong Kong, a longitudinal sample following multiple Chinese birth cohorts with different entry years during the period from 2001 to 2012 was analysed to describe the cohort effect on frailty among Chinese older people and to assess the modification of this effect by age, period, gender, marital status, educational levels, socioeconomic status, lifestyle and social factors using an age-period-cohort analysis. Methods Study sample We used data from the 18 EHCs of the Department of Health, which have been collecting longitudinal health data from a large population-based cohort in Hong Kong since 1998. All residents of Hong Kong aged 65 years and older can voluntarily enrol. Enrollees in the cohort receive standard medical examinations at baseline and are reassessed in subsequent years. The details of this cohort have been described elsewhere [17]. In this study, we retrieved longitudinal data from the EHCs of individuals who enroled between 2001 and 2012 (n = 427,790), and excluded those who were institutionalised (n = 9,490, 2.2%), those with missing data on living arrangement (n = 6, 0.0%), and those with data with less than 27 of the 30 frailty deficits (n = 345, 0.1%). Of these observations, we further cleaned the data with logical errors. Although it has been suggested that a minimum of 30 deficits is required to define frailty [18], a previous analysis shows no difference between using the full set of 32 deficits and a reduced set of 26 deficits when it comes to define frailty [19]. The study has been approved by the ethics committee of the Department of Health of the Government of the Hong Kong Special Administrative Region. Variable definitions Frailty index The multiple-deficits approach was used to construct the Frailty Index (FI), which consisted of 30 components covering chronic disease history and medication use, physical and cognitive functioning, psychological wellbeing, and geriatric symptoms, with a score of 1 representing a deficit for each component (Supplementary Table). The FI was calculated as the proportion of the number of deficits for an individual to the maximum total number of deficits. The deficits included in the FI satisfied the criteria as previously described [18]. All deficits of FI were associated with the health status of the individual, had less than 5% missing values, presented in at least 1% of the study population, and the prevalence of each deficit generally increased with age. In addition, the FI has a strong association with mortality. However, the risks for mortality vary across birth cohorts (data not shown). It is possible that the follow-up time may not be long enough to complete survival, particularly among the recent cohorts. Age and cohort Age was calculated by subtracting the survey year from the birth year. It was transformed by subtracting the value from the grand median and further being divided by 10 for mathematical convenience. The birth cohorts were categorised into the years 1901–23 (reference group), 1924–29, 1930–35, 1936–41 and 1942–47. Covariates Gender and the following time-varying covariates were included in the analysis: marital status, educational attainments, employment, housing, recipient of social assistance, smoking, regular alcohol consumption (drink at least once per week), regular exercise (exercise at least once per week) and participation in social activities, including volunteering and participating in leisure activities. Statistical analysis Descriptive statistics (mean (SD) or %) were used for summarising the primary FI outcome and other explanatory variable stratified by birth cohort. The repeated measurement data were analysed by a two-level hierarchical linear model (HLM) [14, 19]. In HLM, random effects of individual and time are able to capture the heterogeneity of baseline data and the changes in FI over time. It is also a modelling technique used to handle missing data and unequal time spaces which are typical in longitudinal observation studies [20]. In our study, the model is able to depict the change in FI with age for each individual (level 1) and the heterogeneity due to differences in gender and birth cohort amongst our study population (level 2). The model is specified as follows:   Level1Model:FIit=β0i+β1iAgeit+β2iAgeit2+∑jαjiZj,it+εit (1)  Level2Model for intercept:β0i=γ00+γ01Genderi+γ02Cohorti+u0i (2)  Level2Model for growth rate:β1i=γ10+γ11Genderi+γ12Cohorti+u1i (3) The level 1 model (equation 1) describes the change of FI of an individual i at time t ( FIit) given the population mean ( β0i), linear ( β1i) and quadratic effects ( β2i) of age, after controlling for the effects ( αji) of other demographics variables (i.e., marital status and education level), socioeconomic status, lifestyle and social factors that vary over time ( Zj,it, j = 1, 2,…). εit is the random error within individual i over time. The level 2 model for intercept describes the differences between individuals and is modelled as a function of gender, birth cohort, and their interaction. The model for the growth rate describes the linear rate of change with age from the variables of gender and cohort. The coefficients γ00−γ02andγ10−γ12 are the effects on mean FI and linear growth rate with age in FI, respectively. The age-by-cohort interaction is adopted as a proxy of period effect [21, 22]. The gender-by-cohort interaction was not included in the model due to insignificant effects (result not shown). The Model 1 of HLM was fitted using age, gender and cohorts only. On top of these variables, the Model 2 of HLM was fitted using other demographics variables (i.e., marital status and education level), socioeconomic status, lifestyle and social factors. The goodness of fit of the models were assessed by the 95% credible interval of residual errors, mean absolute error (MAE and mean absolute percentage error (MAPE). The HLM models were fitted using the HPMIXED procedure in SAS 9.4 software. Results The total number of observations was 417,949, which included 94,550 participants (female 64.3%). Table 1 shows the characteristics of the study population by birth cohort. The mean age of cohort at the time of study was 72 years and the mean FI was 0.14. The mean age of the study population decreased from 82 to 67 years across successive cohorts, indicating the presence of confounding by age on trends in frailty and providing justification for subsequent APC analyses. Table 1. Descriptive statistics for the study samples Variablesa  Birth cohort  All (n = 94,550)  1901–23 (n = 9,783)  1924–29 (n = 20,561)  1930–35 (n = 34,646)  1936–41 (n = 20,316)  1942–47 (n = 9,244)  Frailty index (mean ± SD)  0.14 ± 0.07  0.18 ± 0.07  0.16 ± 0.07  0.14 ± 0.07  0.13 ± 0.07  0.13 ± 0.07  Demographics   Age, years (mean ± SD)  72 ± 5.0  82 ± 3.4  75 ± 2.6  70 ± 3.2  69 ± 2.6  67 ± 1.2   Female (%)  64.3  63.7  62.5  62.4  65.8  72.5   Marital status (%)    Married  64.4  41.8  55.7  68.7  72.3  74.4    Never married  2.7  3.7  3.0  2.4  2.5  3.0    Widowed  30.1  52.9  39.2  26.2  21.9  18.2    Separated/divorced/others  2.7  1.6  2.1  2.6  3.2  4.4   Educational levels (%)    No education  35.4  47.2  45.6  38.5  23.3  15.1    Primary  38.3  36.1  37.7  38.9  37.9  41.2    Secondary  19.3  12.5  13.4  17.1  26.3  32.9    Post-secondary  6.9  4.3  3.4  5.5  12.5  10.8  Socioeconomic status   Employment (%)    Unemployed/retired/others  95.2  98.5  96.8  94.1  94.7  93.4    Working full-time  2.7  0.9  1.8  3.4  3.0  3.2    Working part-time  2.1  0.7  1.5  2.5  2.3  3.4   Housing (%)    Public and subsidised  43.9  41.4  40.6  41.6  48.4  52.5    Private  54.1  55.6  56.6  56.3  50.5  46.6    Temporary/others  2.1  3.0  2.9  2.1  1.1  0.9   Recipient of social assistance (%)  12.1  24.0  15.5  10.5  8.1  6.3  Lifestyle   Smoking (%)    Never-smoker  73.1  64.4  67.8  72.9  78.6  83.4    Ex-smoker  19.7  28.4  24.1  19.0  15.8  11.5    Non-daily smoker  0.5  0.1  0.2  0.3  1.0  1.6    Daily smoker  6.7  7.2  8.0  7.8  4.6  3.6   Regular drinker (%)  3.8  2.9  3.8  4.4  4.1  2.3   Regular exercise (%)  90.4  89.0  90.4  89.7  91.9  91.7  Social support/social factor   Participation in social activities (%)  69.6  65.4  63.7  61.7  81.6  90.6  Variablesa  Birth cohort  All (n = 94,550)  1901–23 (n = 9,783)  1924–29 (n = 20,561)  1930–35 (n = 34,646)  1936–41 (n = 20,316)  1942–47 (n = 9,244)  Frailty index (mean ± SD)  0.14 ± 0.07  0.18 ± 0.07  0.16 ± 0.07  0.14 ± 0.07  0.13 ± 0.07  0.13 ± 0.07  Demographics   Age, years (mean ± SD)  72 ± 5.0  82 ± 3.4  75 ± 2.6  70 ± 3.2  69 ± 2.6  67 ± 1.2   Female (%)  64.3  63.7  62.5  62.4  65.8  72.5   Marital status (%)    Married  64.4  41.8  55.7  68.7  72.3  74.4    Never married  2.7  3.7  3.0  2.4  2.5  3.0    Widowed  30.1  52.9  39.2  26.2  21.9  18.2    Separated/divorced/others  2.7  1.6  2.1  2.6  3.2  4.4   Educational levels (%)    No education  35.4  47.2  45.6  38.5  23.3  15.1    Primary  38.3  36.1  37.7  38.9  37.9  41.2    Secondary  19.3  12.5  13.4  17.1  26.3  32.9    Post-secondary  6.9  4.3  3.4  5.5  12.5  10.8  Socioeconomic status   Employment (%)    Unemployed/retired/others  95.2  98.5  96.8  94.1  94.7  93.4    Working full-time  2.7  0.9  1.8  3.4  3.0  3.2    Working part-time  2.1  0.7  1.5  2.5  2.3  3.4   Housing (%)    Public and subsidised  43.9  41.4  40.6  41.6  48.4  52.5    Private  54.1  55.6  56.6  56.3  50.5  46.6    Temporary/others  2.1  3.0  2.9  2.1  1.1  0.9   Recipient of social assistance (%)  12.1  24.0  15.5  10.5  8.1  6.3  Lifestyle   Smoking (%)    Never-smoker  73.1  64.4  67.8  72.9  78.6  83.4    Ex-smoker  19.7  28.4  24.1  19.0  15.8  11.5    Non-daily smoker  0.5  0.1  0.2  0.3  1.0  1.6    Daily smoker  6.7  7.2  8.0  7.8  4.6  3.6   Regular drinker (%)  3.8  2.9  3.8  4.4  4.1  2.3   Regular exercise (%)  90.4  89.0  90.4  89.7  91.9  91.7  Social support/social factor   Participation in social activities (%)  69.6  65.4  63.7  61.7  81.6  90.6  aBaseline records were used for time-varying covariates. Table 2 shows the model estimates for all cohorts. Quadratic age trajectories of the FI were found, indicating accelerated increases in the accumulation of health deficit with age. Controlling for the effect of age, positive cohort effects on FI were observed. An interaction between birth cohort and age was observed, suggesting the existence of a cohort effect independent of age and period effects (Model 2). The HLM shows acceptable goodness of fit in which the 95% credible interval, MAE and MAPE of the full model are (−0.05 to 0.06), 0.026 and 22%, respectively. Table 2. Parameter estimates (95% confidence intervals) from HLM   Model 1  Model 2  Demographics   Age  0.0460 (0.0429, 0.0490)  0.0449 (0.0418, 0.0481)   Age2  0.0130 (0.0119, 0.0141)  0.0127 (0.0116, 0.0138)   Female  0.0234 (0.0225, 0.0242)  0.0234 (0.0224, 0.0244)   Age*female  −0.0012 (−0.0023, 0.0000)  −0.0011 (−0.0024, −0.0001)   Cohorts (Reference: 1901–1923)    1924–29  0.0162 (0.0140, 0.0184)  0.0164 (0.0142, 0.0186)    1930–1935  0.0243 (0.0221, 0.0265)  0.0257 (0.0235, 0.0279)    1936–41  0.0293 (0.0271, 0.0315)  0.0321 (0.0298, 0.0344)    1942–1947  0.0502 (0.0448, 0.0556)  0.0533 (0.0479, 0.0587)   Age*cohorts    Age*1924–29  0.0071 (0.0044, 0.0097)  0.0082 (0.0055, 0.0109)    Age*1930–1935  0.0136 (0.0104, 0.0169)  0.0148 (0.0115, 0.0182)    Age*1936–1941  0.0268 (0.0228, 0.0308)  0.0277 (0.0237, 0.0318)    Age*1942–47  0.0490 (0.0411, 0.0569)  0.0495 (0.0415, 0.0574)   Marital status (reference: married)    Never married    −0.0005 (−0.0027, 0.0017)    Widowed, separated/divorced, or others    0.0015 (0.0009, 0.0022)   Educational levels (reference: no education)    Primary    −0.0012 (−0.0018, −0.0006)    Secondary or above    −0.0029 (−0.0038, −0.0021)  Socioeconomic status   Working full-time or part-time    −0.0051 (−0.0062, −0.0039)   Private housing    −0.0005 (−0.0010, 0.0001)   Recipient of social assistance    0.0082 (0.0074, 0.0091)  Lifestyle   Smoker    0.0036 (0.0027, 0.0045)   Regular drinker    −0.0022 (−0.0033, −0.0010)   Regular exercise    −0.0047 (−0.0054, −0.0040)  Social support/social factor   Participation in social activities    −0.0023 (−0.0027, −0.0019)    Model 1  Model 2  Demographics   Age  0.0460 (0.0429, 0.0490)  0.0449 (0.0418, 0.0481)   Age2  0.0130 (0.0119, 0.0141)  0.0127 (0.0116, 0.0138)   Female  0.0234 (0.0225, 0.0242)  0.0234 (0.0224, 0.0244)   Age*female  −0.0012 (−0.0023, 0.0000)  −0.0011 (−0.0024, −0.0001)   Cohorts (Reference: 1901–1923)    1924–29  0.0162 (0.0140, 0.0184)  0.0164 (0.0142, 0.0186)    1930–1935  0.0243 (0.0221, 0.0265)  0.0257 (0.0235, 0.0279)    1936–41  0.0293 (0.0271, 0.0315)  0.0321 (0.0298, 0.0344)    1942–1947  0.0502 (0.0448, 0.0556)  0.0533 (0.0479, 0.0587)   Age*cohorts    Age*1924–29  0.0071 (0.0044, 0.0097)  0.0082 (0.0055, 0.0109)    Age*1930–1935  0.0136 (0.0104, 0.0169)  0.0148 (0.0115, 0.0182)    Age*1936–1941  0.0268 (0.0228, 0.0308)  0.0277 (0.0237, 0.0318)    Age*1942–47  0.0490 (0.0411, 0.0569)  0.0495 (0.0415, 0.0574)   Marital status (reference: married)    Never married    −0.0005 (−0.0027, 0.0017)    Widowed, separated/divorced, or others    0.0015 (0.0009, 0.0022)   Educational levels (reference: no education)    Primary    −0.0012 (−0.0018, −0.0006)    Secondary or above    −0.0029 (−0.0038, −0.0021)  Socioeconomic status   Working full-time or part-time    −0.0051 (−0.0062, −0.0039)   Private housing    −0.0005 (−0.0010, 0.0001)   Recipient of social assistance    0.0082 (0.0074, 0.0091)  Lifestyle   Smoker    0.0036 (0.0027, 0.0045)   Regular drinker    −0.0022 (−0.0033, −0.0010)   Regular exercise    −0.0047 (−0.0054, −0.0040)  Social support/social factor   Participation in social activities    −0.0023 (−0.0027, −0.0019)  To compare cohort differences at the same age, we compared predicted age trajectories of the FI across birth cohorts (Figure 1). We found that more recent cohorts had higher levels of frailty than did earlier cohorts at the same age, controlling for demographics, socioeconomic status, lifestyle and social factors. We also found that, on average, being female was associated with 0.02 units increase in FI, after adjusting for age and cohort effects (Table 2, Model 1). At the same age, recent cohorts were at a higher risk of having an increased FI that did earlier cohorts in both men and women (Supplementary Figure). Figure 1. View largeDownload slide Age trajectories of the frailty index (FI) with 95% confidence intervals of the mean FI (red lines). Figure 1. View largeDownload slide Age trajectories of the frailty index (FI) with 95% confidence intervals of the mean FI (red lines). Table 2 (Model 2) also shows the associations of demographics, socioeconomic status, lifestyle and social factors with frailty. Individuals who were widowed had a 0.002 unit higher FI on average. Completing one’s secondary education or higher decreased the FI by 0.003. Working full-time/part-time similarly decreased the FI by 0.005. Recipients of social assistance had a 0.008 unit higher FI on average. Regular drinking and regular exercise were associated with a lower risk of frailty. Smokers including ex-smokers, daily and non-daily smokers, had an increased FI of 0.004 units. Participation in social activities was associated with a lower risk of frailty. The inclusion of these predictors did not attenuate the cohort effect. Discussion Using a large longitudinal dataset of community-dwelling Chinese people aged ≥65 years, we found that more recent cohorts had higher levels of frailty than did earlier cohorts, after adjusting for age, period, gender, marital status, educational levels, socioeconomic status, lifestyle and social factors. This finding indicates that the increase in life expectancy has negative implications, as it is associated with concurrent increases in levels of frailty. This may lead to greater costs for medical care, social services and long-term care. The results from the current study are consistent with a previous British analysis, which also reported higher levels of frailty in recent cohorts compared with earlier cohorts (for cohorts aged over 70 in 2002) [16]. An US-based study examining cohort-specific trajectories of frailty also reported higher levels of frailty in recent cohorts compared with earlier cohorts [14]. Similar findings were also observed in the US for cohort-specific trajectories of self-reported illness [23, 24]. In contrast to these findings, a Sweden-based study examining cohort-specific trajectories of frailty reported higher levels of frailty in earlier cohorts compared with recent cohorts [15]. The discrepancy of the findings may reflect different social, cultural and health policies at different stages of demographic transitions [25]. The use of different types of data (repeated cross-sectional or longitudinal) and definitions of frailty across studies can also influence results. In particular, some previous studies did not control for period effects, which may significantly cofound the cohort effects. Although the mechanisms that cause frailty to increase amongst more recent birth cohorts in our study population are not well understood, one possibility is that frail individuals in more recent cohorts had a better chance of surviving into old age than their predecessors, most likely due to improvements in living conditions and medical treatments over their lifetimes. These improvements imply that older people in more recent cohorts are expected to live longer, but the improved survival rates of these people may be associated with an increase in chronic diseases and impaired physical and cognitive functioning (as observed in a Chinese cohort) [4], which, in turn, increase the risk of frailty. Longitudinal trajectories of lifestyle and social factors due to demographic changes may also partly explain the differences in frailty risk across successive birth cohorts. For example, the economic reforms and high-tech revolution that have taken place over the past few decades in Hong Kong may have exposed individuals in recent cohorts to sedentary occupations (non-agriculture-/non-manufacturing-related jobs) that require less daily energy expenditure than earlier cohorts. This has been accompanied by a decline in functional capacity and health, which might have promoted frailty. In addition, the establishment and the mobility of nuclear families, as reflected by the trends in declining household size and the rising number of older people living alone [26, 27], might have had adverse impacts on the social networks/connections of older people, which may accelerate frailty. Findings from a Mexican–American study showed that social support was protective against worsening in frailty among individuals with moderate frailty [28]. Analyses of EHCs’ data also demonstrated that older people who had participated in social activities had lower levels of frailty than those who did not; hence, the findings of this study together with current available literature lend support to the concept that social environment may be an especially important mechanism for understanding the trajectory patterns of frailty. In this study, the cohort effect on frailty has been adjusted for multiple covariates. Our findings reaffirm the importance of education, healthy lifestyle (regular exercise, not smoking), social support and social participation in slowing down the onset of frailty. Nevertheless, consistent with the results reported in the Women’s Health Initiative Observational Study, the Seniors-ENRICA cohort, and the Health and Retirement Study, where moderate drinkers were associated with a lower risk of frailty [29–31], our data demonstrated that regular alcohol consumption was associated with a lower risk of frailty, although it is well known that alcohol in excess is carcinogenic and detrimental to health [32]. The inverse association between alcohol intake and frailty may be due to the anti-inflammatory effect associated with low-to-moderate alcohol consumption [31]; however, it could also be due to the abstainer/quitter bias [33], as individuals in poor health, particularly older people, drink less than those individuals in good health. Several limitations of the current study require discussion. The earliest and latest cohorts did not capture a full age distribution, thus biasing the estimates for cohort trends. Individuals born between 1901 and 1923 were merged due to the comparatively small size of several age groups, thus limiting the number of birth cohorts being studied. Also, information on risk factors in early- and mid-life (e.g. changing dietary pattern) that may affect frailty at older ages was not available, limiting the ability to identify causes of increased frailty in late life. Another limitation is that study participation was voluntary which could result in selection bias. Compared to individuals who responded to follow-up assessments, those who did not respond to any follow-up assessment between 2001 and 2012 were older and had lower levels of education. Nevertheless, HLM was used to capture the heterogeneity of baseline data and the changes over time, which can yield reliable results despite the presence of missing data. In the HLM model, age is allowed to vary by cohort groups with the uses of random effects. By assuming independently and identically distributed residuals at the higher-level, the period effects as well as the cohort effects were distributed independently. We also followed the framework of the hierarchical age-period-cohort (HAPC) regression model from Yang and Land [34], adding a quadratic term of age to prevent the problem of under-identification among APC. Nevertheless, as demonstrated by Bell and Jones [35], the period or cohort trend could not be well-converted into a separate random effect if the independence assumption cannot be held. As shown in a simulation scenario in Bell and Jones’ paper, the cohort and age effects could be over-estimated if there is a priori linear trend of period effect. Finally, the design of the study is subject to survival bias as some people who were frail may not have survived to the age to be included in the study. This may have led to underestimations of the levels of frailty. Therefore, the study population might not completely represent the whole geriatric population in Hong Kong. The results of this study should be interpreted with cautions. The strengths of this study include the large sample size and the adjustment of multiple potential confounders. In addition, the FI obtained using the described methods were in line with age-specific/overall FI obtained in other studies, including the English Longitudinal Study of Ageing [16] and the Health and Retirement Survey [14]. The similarity of our estimates for the FI with those of independent studies underpins the validity of our calculations for FI at population level. In conclusion, our data demonstrate that more recent cohorts had higher levels of frailty than did earlier cohorts amongst both men and women. Therefore, frailty interventions, coupled with early detection, should be developed to combat the increasing rates of frailty in Hong Kong Chinese, which might have far-reaching benefits for individuals and society by preserving function into old age, increasing the number of years spent in good health at older ages, and decreasing health care costs. The inverse association between social participation and frailty observed in the present study also implies the need for encouraging the social participation of older people. At the same time, the surveillance of changes in frailty of older people will also be of fundamental importance to the planning of health care and the resources needed. Our results can also be used to predict future health trends and the public health burden of the older population. Key points More recent cohorts had higher levels of frailty than did earlier cohorts in both men and women. The increase in life expectancy has negative implications, as it is associated with concurrent increases in levels of frailty. Frailty interventions, coupled with early detection, should be developed to combat the increasing rates of frailty in Hong Kong. Supplementary Data Supplementary data mentioned in the text are available to subscribers in Age and Ageing online. Conflict of interest None. Funding None. Acknowledgements We thank the cohort members of the Elderly Health Centres. We thank the investigators, research associates and team members for the design, collection, collation, validation and management of the data used in this article. References 1 Crimmins EM, Beltran-Sanchez H. Mortality and morbidity trends: is there compression of morbidity? J Gerontol B Psychol Sci Soc Sci  2011; 66B: 75– 86. Google Scholar CrossRef Search ADS   2 GBD 2015 Disease and Injury Incidence and Prevalence Collaborators. Global, regional, and national incidence, prevalence, and years lived with disability for 310 diseases and injuries, 1990–2015: a systematic analysis for the Global Burden of Disease Study 2015. Lancet  2016; 388: 1545– 1602. CrossRef Search ADS PubMed  3 Yu R, Wong M, Chang B et al.  . Trends in activities of daily living disability in a large sample of community-dwelling Chinese older adults in Hong Kong: an age-period-cohort analysis. BMJ Open  2016; 6: e013259. Google Scholar CrossRef Search ADS PubMed  4 Zeng Y, Feng Q, Hesketh T et al.  . Survival, disabilities in activities of daily living, and physical and cognitive functioning among the oldest-old in China: a cohort study. Lancet  2017; 389: 1619– 29. Google Scholar CrossRef Search ADS PubMed  5 Freedman VA, Martin LG, Schoeni RF. Recent trends in disability and functioning among older adults in the United States: a systematic review. J Am Med Assoc  2003; 289: 3242– 2. Google Scholar CrossRef Search ADS   6 Seeman TE, Merkin SS, Crimmins EM et al.  . Disability trends among older Americans: National Health and Nutrition Examination Surveys, 1988–1994 and 1999–2004. Am J Public Health  2010; 100: 100– 7. Google Scholar CrossRef Search ADS PubMed  7 Lin SF, Beck AN, Finch BK et al.  . Trends in U.S. older adult disability: exploring age, period, and cohort effects. Am J Public Health  2012; 102: 2157– 63. Google Scholar CrossRef Search ADS PubMed  8 Fried LP, Tangen CM, Walston J et al.  . Frailty in older adults: evidence for a phenotype. J Gerontol A Biol Sci Med Sci  2001; 56: M146– 56. Google Scholar CrossRef Search ADS PubMed  9 Morley JE, Vellas B, van Kan GA et al.  . Frailty consensus: a call to action. J Am Med Dir Assoc  2013; 14: 392– 7. Google Scholar CrossRef Search ADS PubMed  10 Rockwood K, Song X, MacKnight C et al.  . A global clinical measure of fitness and frailty in elderly people. Can Med Assoc J  2005; 173: 489– 95. Google Scholar CrossRef Search ADS   11 Santos-Eggimann B, Cuenoud P, Spagnoli J et al.  . Prevalence of frailty in middle-aged and older community-dwelling Europeans living in 10 countries. J Gerontol A Biol Sci Med Sci  2009; 64: 675– 81. Google Scholar CrossRef Search ADS PubMed  12 Collard RM, Boter H, Schoevers RA et al.  . Prevalence of frailty in community-dwelling older persons: a systematic review. J Am Geriatr Soc  2012; 60: 1487– 92. Google Scholar CrossRef Search ADS PubMed  13 Woo J, Yu R, Wong M et al.  . Frailty screening in the community using the FRAIL scale. J Am Med Dir Assoc  2015; 16: 412– 9. Google Scholar CrossRef Search ADS PubMed  14 Yang Y, Lee LC. Dynamics and heterogeneity in the process of human frailty and aging: evidence from the U.S. older adult population. J Gerontol B Psychol Sci Soc Sci  2009; 65B: 246– 55. Google Scholar CrossRef Search ADS PubMed  15 Horder H, Skoog I, Johansson L et al.  . Secular trends in frailty: a comparative study of 75-year olds born in 1911–12 and 1930. Age Ageing  2015; 44: 817– 22. Google Scholar CrossRef Search ADS PubMed  16 Marshall A, Nazroo J, Tampubolon G et al.  . Cohort differences in the levels and trajectories of frailty among older people in England. J Epidemiol Community Health  2015; 69: 316– 21. Google Scholar CrossRef Search ADS PubMed  17 Schooling CM, Chan WM, Leung SL et al.  . Cohort profile: Hong Kong Department of Health Elderly Health Service Cohort. Int J Epidemiol  2016; 45: 64– 72. Google Scholar CrossRef Search ADS PubMed  18 Searle SD, Mitnitski A, Gahbauer EA et al.  . A standard procedure for creating a frailty index. BMC Geriatr  2008; 8: 24. Google Scholar CrossRef Search ADS PubMed  19 Kulminski A, Yashin A, Ukraintseva S et al.  . Accumulation of health disorders as a systemic measure of aging: findings from the NLTCS data. Mech Ageing Dev  2006; 127: 840– 8. Google Scholar CrossRef Search ADS PubMed  20 Raudenbush SW, Chan WS. Growth curve analysis in accelerated longitudinal designs. J Res Crime Delinq  1992; 29: 387– 411. Google Scholar CrossRef Search ADS   21 Jaacks LM, Gordon-Larsen P, Mayer-Davis EJ et al.  . Age, period and cohort effects on adult body mass index and overweight from 1991 to 2009 in China: the China Health and Nutrition Survey. Int J Epidemiol  2013; 42: 828– 37. Google Scholar CrossRef Search ADS PubMed  22 Yang YC, Land KC. Review of Age-Period-Cohort analysis: new models, methods, and empirical applications, by Yang YC and Land KC, J Am Stat Assoc, 109, 865: Comment by Fienberg, Hodges, and Luo and Replies REPLY. J Am Stat Assoc  2015; 110: 457– 7. Google Scholar CrossRef Search ADS   23 Mirowsky J, Ross CE. Education and self-rated health—cumulative advantage and its rising importance. Res Ageing  2008; 30: 93– 122. Google Scholar CrossRef Search ADS   24 Yang Y, Lee LC. Sex and race disparities in health: cohort variations in life course patterns. Soc Forces  2009; 87: 2093– 2124. Google Scholar CrossRef Search ADS   25 Stolz E, Mayerl H, Waxenegger A et al.  . Impact of socioeconomic position on frailty trajectories in 10 European countries: evidence from the Survey of Health, Ageing and Retirement in Europe (2004–2013). J Epidemiol Community Health  2017; 71: 73– 80. Google Scholar CrossRef Search ADS PubMed  26 Census and Statistics Department, HKSAR. The profile of the population in one-person households, 2011. 27 Census and Statistics Department, HKSAR. Thematic Report: Older Persons, 2011. 28 Peek MK, Howrey BT, Ternent RS et al.  . Social support, stressors, and frailty among older Mexican American adults. J Gerontol B Psychol Sci Soc Sci  2012; 67: 755– 64. Google Scholar CrossRef Search ADS PubMed  29 Woods NF, LaCroix AZ, Gray SL et al.  . Frailty: emergence and consequences in women aged 65 and older in the women’s health initiative observational study. J Am Geriatr Soc  2005; 53: 1321– 30. Google Scholar CrossRef Search ADS PubMed  30 Ortola R, Garcia-Esquinas E, Leon-Munoz LM et al.  . Patterns of alcohol consumption and risk of frailty in community-dwelling older adults. J Gerontol A Biol Sci Med Sci  2016; 71: 251– 8. Google Scholar CrossRef Search ADS PubMed  31 Shah M, Paulson D. C-reactive protein partially mediates the relationship between moderate alcohol use and frailty. Gerontologist  2016; 56: 318– 8. Google Scholar PubMed  32 Turati F, Galeone C, Rota M et al.  . Alcohol and liver cancer: a systematic review and meta-analysis of prospective studies. Ann Oncol  2014; 25: 1526– 35. Google Scholar CrossRef Search ADS PubMed  33 Rehm J, Irving H, Ye Y et al.  . Are lifetime abstainers the best control group in alcohol epidemiology? On the stability and validity of reported lifetime abstention. Am J Epidemiol  2008; 168: 866– 71. Google Scholar CrossRef Search ADS PubMed  34 Yang Y, Land KC. A mixed models approach to the age-period-cohort analysis of repeated cross-section surveys, with an application to data on trends in verbal test scores. Sociol Methodol  2006; 36: 75– 97. Google Scholar CrossRef Search ADS   35 Bell A, Jones K. Another’futile quest’? A simulation study of Yang and Land’s Hierarchical Age-Period-Cohort model. Demogr Res  2014; 30: 333. Google Scholar CrossRef Search ADS   © The Author 2017. Published by Oxford University Press on behalf of the British Geriatrics Society.All rights reserved. For permissions, please email: journals.permissions@oup.com

Journal

Age and AgeingOxford University Press

Published: Mar 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off