Introduction
Population aging is a severe issue worldwide, and it is even more so in Taiwan than in most other countries. In 2018, 14% of Taiwan’s population was over the age of 65 years, according to the Taiwanese Ministry of the Interior (Ministry of the Interior, 2020). One of the most important issues for aging is frailty (Clegg et al., Reference Clegg, Young, Iliffe, Rikkert and Rockwood2013). Frailty is conceptually defined as having noticeable declines in physical conditions, which, subsequently, lead to increased possibilities of hospitalization, disability, and death (Fried et al., Reference Fried2001). The prevalence of frailty among older adults in a community setting was around 11%, but the number ranged widely from 4.0 to 59.1% (Collard et al., Reference Collard, Boter, Schoevers and Oude Voshaar2012). In late life, depression has been one of the leading mental illnesses for disability (Fiske et al., Reference Fiske, Wetherell and Gatz2009). The prevalence of major depression in older age ranges from 4.6% to 9.3%, and subthreshold depression is two to three times more prevalent than major depression (Rodda et al., Reference Rodda, Walker and Carter2011). In Taiwan, the prevalence of depression in adults aged 55 years or above is around 1.5% and 3.7% for major and minor depression, respectively (Wu et al., Reference Wu, Yu, Lee, Tseng, Chiu and Hsiung2017).
Mounting evidence points to a robust association between depression and frailty (Buigues et al., Reference Buigues, Padilla-Sánchez, Garrido, Navarro-Martínez, Ruiz-Ros and Cauli2015). The bidirectional associations between the two syndromes in cross-sectional studies have been summarized in several systematic reviews (Chu et al., Reference Chu, Chang, Ho and Lin2019; Mezuk et al., Reference Mezuk, Edwards, Lohman, Choi and Lapane2012; Soysal et al., Reference Soysal2017). However, relatively few prospective studies have explored the direction of effects between baseline depression and the development of frailty over time (Aprahamian et al., Reference Aprahamian2019; Lakey et al., Reference Lakey2012; Prina et al., Reference Prina2019; Sy et al., Reference Sy, McCulloch and Johansen2019; Woods et al., Reference Woods2005; Zhang et al., Reference Zhang2019). Part of the reason for the robust association could be that these two syndromes share common symptoms, even though some latent class analyses have suggested that depression and frailty are different entities (Lohman et al., Reference Lohman, Dumenci and Mezuk2016; Mezuk et al., Reference Mezuk, Lohman, Dumenci and Lapane2013).
Based on Fried’s definition, frailty includes three or more components out of five domains: unintentional weight loss, exhaustion, poor muscle energy, slow gait speed, and low physical activity. Meeting only one or two criteria was defined as pre-frailty, whereas meeting none of the above criteria was regarded as robust. There are some overlapping criteria for depression and frailty. For example, unintentional weight loss, exhaustion, and slowness in frailty criteria are analogous to appetite change, fatigue, and psychomotor retardation in depression syndrome. Thus, individuals with depression are more likely to meet frailty criteria. Because depression-related criteria can be attributed in part to the further development of frailty, the association between depression and incident frailty may have been overestimated. Despite the potential confounding effects of the two measures, most prospective studies did not adjust for baseline pre-frailty status in their risk assessment for incident frailty (Aprahamian et al., Reference Aprahamian2019; Lakey et al., Reference Lakey2012; Prina et al., Reference Prina2019; Woods et al., Reference Woods2005; Zhang et al., Reference Zhang2019). Only one study removed an overlapping item from baseline depressive symptoms to reduce the operational confounding effect (Sy et al., Reference Sy, McCulloch and Johansen2019).
Thus, in the present study, we investigated the associations between baseline depressive symptoms and incident frailty stratified by robust or pre-frailty status at baseline to address the issue of overlapping criteria. This study was conducted by examining the results of a large, nationwide, and longitudinal aging study of community-dwelling older people in Taiwan.
Methods
Sample
This study collected data from a nationwide survey, the Healthy Aging Longitudinal Study in Taiwan (HALST). HALST recruited community dwellers aged 55 years and older from urban and rural areas in seven regions in Taiwan. The population represented diverse sociodemographic features from Taiwan’s northern, western, southern, and eastern regions. A systematic random sampling method ensured that the study covered a sufficient segment of the older adult population by stratifying age, gender, and education levels. Individuals with any of the following conditions were excluded: highly contagious infectious diseases, chronic terminal illness, dementia, severely impaired motor or cognitive function, impaired hearing, schizophrenia, organic mental disorders, or intelligence disability.
In the first wave of the HALST (Baseline; 2008–2013), 5,664 older adults were enrolled in the study. Details of the sample selection, inclusion, and exclusion were described in our previous report (Hsu et al., Reference Hsu2017). In this study, the second wave of the HALST, or the follow-up, was conducted between 2014 and 2020. A total of 2,220 subjects did not complete the follow-up assessment and thus were excluded from this analysis. Further 583 subjects were excluded due to missingness of frailty data. Another 144 out of 2,923 participants were excluded due to frailty at baseline. As a result, 2,717 participants were included in our analyses. The baseline characteristics of the participants included and excluded, or with versus without follow-up, are shown in Tables S1 and S2, respectively. The flowchart for selection of subjects is depicted in Figure 1.
Figure 1. Flowchart of sample inclusion and exclusion.
Informed consent was obtained from all participants. Data were collected through clinical evaluation, physical performance assessment, home interviews, and laboratory examinations. Interviewers were trained and supervised. The HALST study protocol was approved by the ethics committee of the National Health Research Institutes in Taiwan.
Measures of depressive symptoms
We used the 20-item Center for Epidemiological Studies Depression (CES-D) scale to measure depressive symptoms (Radloff, Reference Radloff1977). CES-D includes 20 items recorded on a scale of 0 to 3, with 3 being the highest. This CES-D has been validated and widely used with community-residing older adults (Fu et al., Reference Fu, Lee and Chen2003; Lyness et al., Reference Lyness, Noel, Cox, King, Conwell and Caine1997). CES-D scores range from 0 to 60; and the sensitivity and specificity of the CES-D scale using a cutoff point of 15 for the general population of Taiwan were 92% and 91%, respectively (Chien and Cheng, Reference Chien and Cheng1985). To be consistent with Radloff’s original research and international studies, as well as our previous publications (Hsu et al., Reference Hsu2016; Radloff, Reference Radloff1977; Wu et al., Reference Wu, Yu, Lee, Tseng, Chiu and Hsiung2017; Ying et al., Reference Ying, Yap, Gandhi and Liew2019), participants were classified into mutually exclusive groups with or without depressive symptoms. A CES-D score of 16 or more was regarded as positive for depressive symptoms.
Assessment of frailty
We used the Fried frailty phenotype to identify frailty status (Fried et al., Reference Fried2001); this includes five domains: unintentional weight loss, exhaustion, poor muscle energy, slow gait speed, and low physical activity. Meeting three or more out of five criteria was defined as frailty, while meeting only one or two criteria was defined as pre-frailty. Participants who matched no criteria were defined as robust.
Unintentional weight loss was determined by an involuntary weight loss of three kilograms over the preceding year. Exhaustion was defined according to the response to two statements from the CES-D: (1) “I felt that everything I did was an effort”; and (2) “I could not get going.” If either of the responses was “occasionally (3–4 days)” or “mostly (5–7 days)” during the past week, the subject would be determined as positive for exhaustion. Muscle energy was assessed by handgrip strength using a handheld dynamometer. Gait speed was evaluated by a time walking test (3 or 4 meters according to different study sites at usual pace, turn and return to the starting point). The above two measures were determined by the lowest 20% sex- and body mass index (BMI)-specific cutoff points for muscle energy and the lowest 20% of sex- and height-specific cutoff points for gait speed. Finally, physical activity was assessed by measuring either energy expenditure (Ainsworth et al., Reference Ainsworth2011; Wai et al., Reference Wai2008) or the response to the following two questions: (1) “Did you do regular exercise during the preceding year?”; and (2) “Did you do any strenuous activity during the preceding year?” Individuals were classified into the low physical activity group by having an energy expenditure of less than 685 kcal per week for males or 420 kcal per week for females, or by responding “no” to both those questions.
Assessment of covariates
Demographic characters included age, gender, education level, and marital and residential (urban/rural) status. Individuals’ education levels were classified as none (illiterate), low (≤ elementary school), intermediate (high school), or high (college or higher). Lifestyle variables included BMI, current smoking, betel nut chewing, and alcohol consumption. The Mini-Mental State Examination (MMSE; Folstein et al., Reference Folstein, Folstein and McHugh1975) was used to assess cognitive function.
History of cardiovascular diseases (i.e. hypertension, diabetes mellitus, dyslipidemia, cerebrovascular disease, and heart disease) and non-cardiovascular diseases (i.e. asthma, chronic pulmonary disease, malignancy, peptic ulcer, chronic liver disease, chronic renal disease, parkinsonism, hip fracture, osteoporosis, anemia, gout, cataract, and spinal cord disorder) were identified by self-reported conditions. Social support networks were defined based on the number of social contacts with neighbors, relatives, or friends within the past 1 year. Life events during the past year included being a victim of a catastrophe or trauma, the death of an intimate friend or a family member, a change in economic or health status (e.g. unemployment, financial burden, hospitalization, and newly diagnosed cancer), migration, increased care burden on family, and relational problems (e.g. separation, divorce, conflict with a family member).
Statistical analysis
Participants were classified into mutually exclusive groups based on whether they showed depressive symptoms over the past year. Characteristics of individuals at baseline were reported as follows. Continuous variables were presented as mean ± standard error (SE), and classification variables were reported as percentages. The participants’ characteristics were compared groupwise as well as by the status of depressive symptoms using: (1) the Wilcoxon rank-sum test, a nonparametric analog of two sample t-tests, for continuous variables and (2) chi-square tests for categorical variables.
Cox proportional hazard regression models were used to estimate the risk of developing frailty (three or more criteria) with baseline depressive symptoms. Hazard ratios (HRs) and their 95% confidence intervals (CIs) adjusting for multiple covariates were estimated. In the survival analysis, age was used as timescale. Thus, the effect of age on incident frailty was directly taken into account. This approach allowed us to make individual inferences at specific ages, providing a more understandable interpretation than using follow-up years as the timescale (Lamarca et al., Reference Lamarca, Alonso, Gomez and Muñoz1998).
A subgroup analysis was conducted to explore the effects of robust or pre-frailty status on the association between depressive symptoms and incident frailty. Participants without depressive symptoms served as the reference group in overall and subgroup analyses.
All models were adjusted for gender, education level, partnership status, urbanicity, current smoking, current betel nut eating, current drinking, BMI, MMSE, number of cardiovascular diseases, number of non-cardiovascular diseases, social network, and life events. These covariates were potentially associated with both depressive symptoms and frailty (Cole and Dendukuri, Reference Cole and Dendukuri2003; Feng et al., Reference Feng2017; Hoare et al., Reference Hoare, Jacka and Berk2019; Payne et al., Reference Payne2018; Woo et al., Reference Woo, Zheng, Leung and Chan2015; Yaka et al., Reference Yaka, Keskinoglu, Ucku, Yener and Tunca2014). All statistical analyses were completed using SAS statistical software version 9.4 (SAS Institute Inc., Cary, NC, USA), with p-values of less than 0.05 considered statistically significant.
Results
The characteristics of participants with or without depressive symptoms are shown in Table 1. Among the overall participants, the mean age was 67.5 ± 0.1 years, and 50.3% were women. Compared to participants without depressive symptoms, those who had depressive symptoms were more likely to be female; to have cardiovascular or non-cardiovascular diseases; and to have a poorer social network, lower MMSE score, lower BMI, and more life events. In addition, they were less likely to be partnered. Participants with depressive symptoms were also more likely to have a higher number of Fried’s criteria.
Table 1. Baseline characteristics of study participants
The Wilcoxon rank-sum test or χ2 test was conducted to detect the differences between the two groups (“depression” vs “no depression”).
a Urban areas: Taipei, Changhua, Kaohsiung, and Hualien. Rural areas: Miaoli, Yangmei, Shoufeng, and Chiayi.
b Pre-frailty is defined as the presence of one or two frailty characteristics.
Among the groups with depressive symptoms, 27.0% of participants (n = 27) developed frailty during the second-wave survey (Table 2). The crude HR of depressive symptoms for becoming frail was 2.9 (95% CI 1.9, 4.2) during an average of 5.9 years of follow-up. After adjusting for the covariates listed in Table 1, the HR was 2.6 (95% CI 1.6, 4.2). Subgroup analyses were conducted to account for the effects of baseline robust or pre-frailty status. Compared to participants without depressive symptoms, the adjusted HRs of frailty for those with depressive symptoms were 3.8 (95% CI 1. 6, 9.3) and 2.2 (95% CI 1.2, 4.1) among those with one and two characteristics of pre-frailty, respectively. Among the participants without any frailty characteristics, only one case with depressive symptoms developed frailty during the follow-up. Thus, due to the fairly limited sample size, these findings were not statistically significant.
Table 2. Associations between baseline depressive symptoms and incident frailty, stratified by the number of frailty characteristics at baseline
BMI, body mass index; MMSE, mini-mental status examination.
a The models were adjusted for gender, education, BMI, living in urban areas, partnership status, current smoking, current betel nut eating, current drinking, number of cardiovascular diseases, number of non-cardiovascular diseases, MMSE, social network, and life events.
Table 3 shows the effects of a baseline CES-D score of ≥16 on developing each of the frailty characteristics during the follow-up period. In the overall analysis, the adjusted HR for developing unintentional weight loss was 2.1 (95% CI 1.2, 3.9), and that for developing exhaustion was 4.8 (95% CI 2.5, 9.3). However, there was no significant association of depressive symptoms with the development of poor muscle strength, slowness, or low physical activity. In subgroup analysis, we found that depressive symptoms were still associated with developing exhaustion, except for those with only one frailty characteristic. In addition, depressive symptoms increased the risk of subsequent unintentional weight loss among those with one frailty characteristic. However, the incidences of other individual frailty characteristics were not associated with depressive symptoms in the subgroup analysis.
Table 3. Associations between baseline depressive symptoms and incident individual frailty characteristics, stratified by the number of frailty characteristics at baseline
a The models were adjusted for gender, education, BMI, living in urban areas, partnership status, current smoking, current betel nut eating, current drinking, number of cardiovascular diseases, number of non-cardiovascular diseases, MMSE, social network, and life events.
b We estimated the hazard ratios for those without a particular frailty characteristic at baseline, yet developing that particular frailty characteristic at follow-up. For example, those with slowness at baseline were excluded when estimating the hazard ratio of depressive symptoms for developing slowness.
Finally, Table 4 shows the factors associated with the incidence of frailty in multiple Cox regression modeling. Higher BMI, current smoking status, more non-cardiovascular diseases, and higher CES-D were associated with a higher risk of developing frailty.
Table 4. Multiple survival analysis for factors associated with incidence of frailty
BMI, body mass index; MMSE, mini-mental status examination.
Discussion
In this study, we investigated the association of depressive symptoms with incident frailty in a longitudinal study with an average 5.9-year follow-up among an Asian older population. Overall, the results showed that older adults who had depressive symptoms were more likely to develop frailty compared to their counterparts, with a greater than twofold risk. We also conducted subgroup analysis to account for the effects of baseline robust or pre-frailty status and found their patterns similar to those of the overall analysis. Among the five frailty characteristics, exhaustion and unintentional body weight loss were those most associated with depression.
Since conceptually there are overlapping criteria between depression and frailty, we conducted subgroup analysis to reduce potential confounding effects. The findings were consistent in the subgroup analyses, except for the robust subgroup, which had a very limited sample size. Thus, it is suggested that the overlapping constructs were only partially attributable to the associations. To overcome this overlapping construct issue, one study used a different strategy: removing overlapping symptoms from CES-D; it also observed results consistent with studies using the original CES-D (Sy et al., Reference Sy, McCulloch and Johansen2019). Potentially overlapping criteria included body weight loss, fatigue or exhaustion, and slowness or psychomotor retardation. Thus, excluding one criterion, as per Sy et al. (Reference Sy, McCulloch and Johansen2019), might not fully eliminate the confounding effects. In addition, after one specific criterion was excluded, the patients with remaining depressive symptoms continued to have more frailty characteristics than those without depressive symptoms.
Our findings with an Asian older population are consistent with the findings in other populations (Aprahamian et al., Reference Aprahamian2019; Hajek et al., Reference Hajek2016; Prina et al., Reference Prina2019; Woods et al., Reference Woods2005; Zhang et al., Reference Zhang2019). A large-scale study focusing on low- and middle-income Latin Americans with a higher prevalence of depression reported that depression was associated with a 59% increase in the adjusted risk of developing frailty with modified Fried’s measures (i.e. handgrip measure was not included in their study) (Prina et al., Reference Prina2019). Among the outpatient population in Brazil, depressive symptoms plus antidepressant use were associated with a 2.75-fold increased risk of frailty (Aprahamian et al., Reference Aprahamian2019). Using a six-item short-form of CES-D, researchers observed a twofold increase of hazard among US females (Woods et al., Reference Woods2005). In a separate analysis, researchers used the 15-item Geriatric Depression Scale (GDS) and found a 2.79-fold increased risk of frailty in a Chinese population aged between 70 and 84 years (Zhang et al., Reference Zhang2019).
Of note, we found that the adjusted HRs of depressive symptoms for developing frailty seemed higher among participants with one frailty characteristic (3.8; 95% CI 1. 6, 9.3) than those with two characteristics (2.2; 95% CI 1.2, 4.1). This finding might be because the incidences of participants with two frailty characteristics, regardless of depressive symptom status, were relatively high. However, the incidence differences of depressive symptoms for frailty were almost equal, about 1.6 per 100 person-years for both subgroups. Thus, the HR was attenuated among those with two frailty characteristics.
We found that those with depressive symptoms at baseline were more likely to have unintentional weight loss and exhaustion at follow-up. It should be noted that unintentional weight loss and exhaustion are similar to appetite change and fatigue, the two diagnostic criteria for depressive disorder. Thus, the link between depression and frailty might partially be due to these shared symptoms. In a similar vein, we also expected to see an association between depression and slowness, a frail symptom similar to psychomotor retardation of depression syndrome. To our surprise, however, the association was not significant, even though a prior study with a clinical sample showed that depressive symptoms significantly affected exhaustion and slow gait (Sy et al., Reference Sy, McCulloch and Johansen2019). Another study using a different frailty scale, the FRAIL questionnaire, found that the health subdomains, fatigue, weight loss, and the number of medical diseases led to a more than twofold increased hazard compared with the physical subdomains, resistance, and ambulation (Aprahamian et al., Reference Aprahamian2019). Thus, we concluded that the shared criteria partially contributed to the associations between depressive symptoms and frailty.
Our findings of risk factors for frailty in the Asian population were generally consistent with previous literature conducted in other countries (Feng et al., Reference Feng2017). Several studies demonstrated that multiple chronic illnesses were associated with the incidence of frailty (Aranda et al., Reference Aranda, Ray, Snih, Ottenbacher and Markides2011; Fugate Woods et al., Reference Fugate Woods2005). One study showed that a higher allostatic load would dysregulate multiple physiological systems, which, in turn, accelerated the development of frailty (Gruenewald et al., Reference Gruenewald, Seeman, Karlamangla and Sarkisian2009). High BMI and smoking were both associated with multiple chronic conditions, thereby increasing the risk of frailty. Previous studies showed that living in a high-density neighborhood and also low to the middle socioeconomic neighborhoods were positively associated with frailty (Aranda et al., Reference Aranda, Ray, Snih, Ottenbacher and Markides2011; Myers et al., Reference Myers, Drory, Goldbourt and Gerber2014). However, in our study, no significant association was reported between living in an urban area and incident frailty. In Taiwan, the population density and socioeconomic status in neighborhoods are diverse in urban areas. As a result, the effects of living in urban areas might be mitigated.
There are several limitations to this study. First, the sample size was moderate. In a stratified analysis, only one participant who showed depressive symptoms but no frailty characteristics at baseline eventually developed frailty in the follow-up period. Thus, our findings could not be generalized to individuals who have depressive symptoms but do not show frailty symptoms at baseline. Second, the follow-up rate of the HALST study was not satisfactory. Around 38.1% of the participants at baseline were not enrolled in the second-wave follow-up. The follow-up rates were 53.8% for those with depressive symptoms and 61.3% for those without depressive symptoms. It appeared that people who showed depressive symptoms at baseline were more likely to quit participating in this study in the following waves due to multiple health and functional deteriorations associated with depressive symptoms (Tables S1–S2). Furthermore, hospitalized patients due to mental illness were not recruited into the HALST study. Consequently, the negative effects of depressive symptoms could be underestimated. Third, the present study used the CES-D scale to measure depressive symptoms rather than a clinical diagnosis. Thus, our findings may not be directly generalizable to the association between clinically diagnosed depression and frailty. Finally, we used the subscale of CES-D to measure exhaustion, which might overlap with depressive syndrome, even though results from stratified analysis have eliminated some influences from overlapping effects.
In conclusion, this study took baseline frailty characteristics into account and found an association between depressive symptoms and consequent incident frailty among an older Asian population. Our results suggest that the influences of depressive symptoms could accelerate the development of frailty across different racial and ethnic groups. Among these frailty domains, depressive symptoms were specifically associated with developing unintentional weight loss and exhaustion. For incident frailty prevention, future policies on geriatric public health should focus more on intervention and treatment for depressive symptoms among older adults to prevent subsequent frailty.
Conflict of interest
The authors have no conflicts of interest to declare.
Source of funding
This work was supported by the National Health Research Institutes (project no. PH-97˜110-SP-01, PH-101˜110-PP-04). NHRI had no role in the study design, data collection, analysis, interpretation, or writing of the manuscript.
Description of authors’ roles
Che-Chia Chang: manuscript writing; Chi-Shin Wu: study concept and design, data interpretation, and manuscript writing; Han-Yun Tseng: manuscript revising; Chun-Yi Lee: data analysis; I-Chien Wu, Chih-Cheng Hsu and Hsing-Yi Chang: data acquisition and study concept; Yen-Feng Chiu: study concept and design, data analysis and interpretation, and manuscript revising and preparation; Chao Agnes Hsiung: obtaining funding and data acquisition.
Acknowledgements
We thank all the HALST participants and the members of the HALST study group – in particular Mr. Rui-Ching Wu, for his dedicated efforts on data management. Our appreciations also go to Dr. Tsung-Jen Hsieh for his kind assistance on physical activity classification and to Mr. Mark Swofford (Scientific Editing Office, NHRI) for his timely assistance with the editing of this manuscript. We deeply appreciate all the editors’ and reviewers’ constructive and insightful comments, which have greatly improved the quality of the manuscript.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S1041610221000806
