DOI: https://doi.org/10.21203/rs.3.rs-1852817/v1
Background: Frailty is associated with mortality in the elderly. We aimed to determine the appropriate time and frailty index (FI) threshold for frailty intervention in Chinese community-dwelling older adults.
Methods: In this prospective cohort study, we used data from the 2011 wave of the Chinese Longitudinal Healthy Longevity Study. Follow-up was performed for 7 years from baseline. Using the FI to evaluate frailty and define frailty status, we explored the best time point and FI score for frailty intervention, by comparing the relationships of FI and frailty status with mortality.
Results: From 2011 to 2018, 8642 participants were included and followed up. A total of 4458 participants died during the study period. After adjusting for variables such as age, sex, marital status, education level, and living conditions, the hazard ratio (HR) of mortality risk based on the FI at baseline was 37.484 (95% confidence interval [CI]: 30.217-46.498; P<0.001); female sex, living in the city, being married, and living with spouse were found to be protective factors, whereas ageing was a risk factor of frailty. The mortality risk was higher in pre-frail than in frail participants (HR: 3.588, 95% CI: 3.212-4.009, P<0.001). Piecewise linear regression analysis revealed an FI score threshold of 0.5. When the FI score was >0.5, the HR of mortality based on the FI was 15.758 (95% CI: 3.656-67.924; P<0.001); when the FI score was ≤0.5, the HR of mortality based on the FI was 48.944 (95% CI: 36.162-66.244; P<0.001).
Conclusion: The FI is a stronger predictor for mortality than the frailty status. The advancement of early interventions for mortality risk reduction is more beneficial in pre-frail than in frail patients, and an FI score of 0.5 was found to be the threshold for mortality prediction using the FI.
The rapid ageing of the global population has become a major trend in the global demographic structure owing to reductions in fertility and mortality rates [1, 2]. Frailty is becoming an increasingly obvious and common feature of the elderly with ageing; the decline of various physiological functions related to age increases, thereby increasing vulnerability to stressors. In addition to disease or disability, frailty is associated with a systemic impairment of physical and cognitive functions, including symptoms, diseases, and life-long deficits [3, 4]. People with frailty are more likely to experience a variety of negative health conditions, such as falls, fractures, hospitalization, need for nursing home placement, disability, poor quality of life, and dementia [5–9].
The frailty index (FI) is one of the most commonly used tools to measure frailty. The FI is evaluated based on the concept that frailty is a state caused by a life-long accumulation of health deficits; the higher the number of health deficits, the greater the tendency for frailty. These health deficits include symptoms, disease, disability, abnormal laboratory findings, and social characteristics [10–12]. The FI is predictive for adverse outcomes and is directly related to survival outcomes [13–15]. Moreover, compared with chronological age, the FI has a stronger correlation with mortality, especially within short intervals less than 4 years [16].
The FI has been shown to vary with time; thus, the FI evaluated using cross-sectional studies cannot accurately predict mortality risk [17, 18]. Therefore, it is necessary to perform mortality risk reassessment using dynamic FI changes [19, 20]. Moreover, frailty is not only associated with age but is also affected by risk factors such as impairment of activities of daily living, chronic diseases, depression, poor lifestyle habits, and geriatric syndromes [21, 22]. Effective prevention and treatment can reduce frailty occurrence in older adults [23]. Hence, mortality risk prediction and early intervention to treat debilitating conditions can prolong survival time, thereby alleviating the pressure on medical care [25].
We aimed to collect and evaluate longitudinal data at different time points, and to accurately determine the best time point for frailty intervention using a long follow-up duration. Our findings will potentially enhance decision-making regarding frailty intervention and the effective utilization of medical resources.
The Chinese Longitudinal Healthy Longevity Survey (CLHLS) is a nationwide longitudinal survey conducted in a randomly selected half of the counties and cities in 22 of the 31 provinces in China. All the participants provided written informed consent [24]. We used data from the 2011 wave of the CLHLS, which was followed up in 2014 and 2018. The medical ethics committee of Tongji University approved this study. Participants were excluded if more than 30% of FI variables were missing or died before the 2014 follow-up. Moreover, we excluded individuals who had 80% missing data on cognitive function and less than 30 variables for FI calculation.
Health deficits were evaluated using the FI. We selected 42 items on self-related health, physical function, psychological and cognitive function, comorbidity, and social deficits [25, 26]. Cognitive function was measured using the Mini-Mental State Examination (MMSE) scale [27]. Binary variables were encoded as 0 or 1. For ordered and continuous variables, encoding was based on the distribution. A score of 2 was assigned if the respondent had suffered from more than one serious disease in the past two years. The FI score was calculated as the ratio of health deficits present to the total number of deficits considered, with values ranging between 0 and 1. Higher scores indicated a higher degree of frailty; FI scores < 0.25 and ≥ 0.25 were considered to indicate non-frailty and frailty statuses, respectively [28, 29].
Cox proportional hazards regression and piecewise linear regression[30] were used to evaluate the relationship between the FI and mortality, and the Kaplan-Meier survival function curve was used to estimate the 7-year survival in relation to the FI and frailty status. The areas under the receiver-operating characteristic (ROC) curves (AUCs) of FI and frailty status were calculated to compare the effects of these parameters on death outcomes during the follow-up period. Statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC, USA), IBM SPSS Statistics version 20 (SPSS Inc., Chicago, IL, USA), and R statistical software version 4.2.0 (R Foundation for Statistical Computing, Vienna, Austria).
A total of 8642 older people participated in the baseline survey in 2011. Table 1 shows the participant characteristics and frailty status at baseline. Participants had a median age of 85.6 ±11.3 years, with a range of 50–114 years. At baseline, 2020 (23.4%), 2802 (32.4%), and 3820 (44.2%) participants were robust (FI score ≤0.1), pre-frail (0.1<FI score<0.25), and frail (FI score ≥0.25), respectively.
Table 1. Participant baseline frailty characteristics
|
Robust |
Pre-frailty |
Frailty |
P value |
||||
Age, n (%) |
|
|
|
|
|
|||
60–75 |
955 (11.1) |
145 (1.7) |
918 (10.6) |
<0.001 |
||||
76–85 |
682 (7.9) |
464 (5.4) |
1193 (13.8) |
|
||||
86-94 |
294 (3.4) |
852 (9.9) |
1074 (12.4) |
|
||||
95-114 |
89 (1.0) |
1341 (15.5) |
635 (7.3) |
|
||||
Sex, n (%) |
|
|
|
|
||||
Male |
1219 (14.1) |
865 (10.0) |
1821 (21.1) |
<0.001 |
||||
Female |
801 (9.3) |
1937 (22.4) |
1999 (23.1) |
|
||||
Residence, n (%) |
|
|
|
|
||||
City |
349 (4.0) |
526 (6.1) |
568 (6.6) |
0.001 |
||||
Town |
642 (7.4) |
839 (9.7) |
1203 (13.9) |
|
||||
Rural |
1029 (11.9) |
1437 (16.6) |
2049 (23.7) |
|
||||
Education level, n (%) |
|
|
|
|
||||
Illiterate |
748 (8.7) |
2074 (24.0) |
2218 (25.7) |
<0.001 |
||||
Primary |
871 (10.1) |
552 (6.4) |
1209 (14.0) |
|
||||
Middle |
351 (4.1) |
129 (1.5) |
340 (3.9) |
|
||||
Higher |
48 (0.6) |
37 (0.4) |
48 (0.6) |
|
||||
Marital status, n (%) |
|
|
|
|
||||
Single |
27 (0.3) |
18 (0.2) |
42 (0.5) |
<0.001 |
||||
Married |
1234 (14.3) |
552 (6.4) |
1548 (18.0) |
|
||||
Divorced or widowed |
753 (8.7) |
2220 (25.8) |
2219 (25.8) |
|
||||
Economic status, n (%) |
|
|
|
|
||||
Poor |
147 (1.7) |
568 (6.6) |
608 (7.1) |
<0.001 |
||||
Rich |
497 (5.8) |
387 (4.5) |
619 (7.2) |
|
||||
Middle |
1369 (16.0) |
1794 (21.0) |
2572 (30.0) |
|
||||
Total 8642 |
2020 (23.4) |
2802 (32.4) |
|
3820 (44.2) |
|
In addition, 4458 participants died during the study period, as observed in 2018. The AUC of FI at baseline was 0.768 (95% CI: 0.758-0.778, P<0.001), whereas the AUC of frailty status was 0.537 (95% CI: 0.524-0.549, P<0.001), thereby showing a weaker prediction with mortality (Figure 1).
The hazard ratio (HR) of mortality according to the FI at baseline was 37.484 (95% CI: 30.217-46.498), P<0.001). Female sex (HR: 0.624, 95% CI: 0.584-0.666, P<0.001), living in the city (HR: 0.864, 95% CI: 0.792-0.943, P=0.001), being married and living with spouse (HR: 0.797, 95% CI: 0.736-0.864, P<0.001) were found to be protective factors, whereas ageing (HR: 1.057, 95% CI: 1.053-1.061, P<0.001) was a risk factor for mortality (Table 2)..
Table 2. Cox regression model analysis of the effect of the frailty index on mortality
|
B |
SE |
Wald |
df |
Sig. |
Exp(B) |
95% CI |
|
Lower |
Upper |
|||||||
Age |
.056 |
.002 |
977.972 |
1 |
.000 |
1.057 |
1.053 |
1.061 |
Sex |
-.472 |
.033 |
200.532 |
1 |
.000 |
.624 |
.584 |
.666 |
Residence |
-.146 |
.045 |
10.754 |
1 |
.001 |
.864 |
.792 |
.943 |
Marital status |
-.227 |
.041 |
30.716 |
1 |
.000 |
.797 |
.736 |
.864 |
FI_11 |
3.624 |
.110 |
1086.390 |
1 |
.000 |
37.484 |
30.217 |
46.498 |
Abbreviations: FI_11, frailty index in 2011; B, Regression coefficients; SE, standard error; df, degree of freedom.
We further classified frailty as non-frailty (FI <0.25) and frailty (FI ≥0.25), and analysed the HR for mortality in different states of frailty. The HR of mortality according to the FI was 2.209 (95% CI: 2.064-2.364, P<0.001) when the frailty status was dichotomized. The female sex, education level, being married, and living with spouse were found to be protective factors, whereas ageing was a risk factor of frailty. The HR for mortality was higher in pre-frail (HR: 3.588, 95% CI: 3.212-4.009, P<0.001) than in frail (HR: 1.820, 95% CI: 1.640-2.021, P<0.001) participants, when the frailty status was triaged as robust, pre-frailty, and frailty. The female sex, being married, and living with spouse were found to be protective factors, whereas ageing was a risk factor of frailty (Table 3).
Table 3. Cox regression model analysis of the effect of frailty status on mortality
|
B |
SE |
Wald |
Df |
Sig. |
Exp(B) |
95% CI |
|
Lower |
Upper |
|||||||
Non-frailty/Frailty |
||||||||
Age |
.061 |
.002 |
1148.558 |
1 |
.000 |
1.062 |
1.059 |
1.066 |
Sex |
-.464 |
.036 |
167.825 |
1 |
.000 |
.628 |
.586 |
.674 |
Education level |
-.065 |
.026 |
6.101 |
1 |
.014 |
.937 |
.890 |
.987 |
Marital status |
-.202 |
.041 |
24.172 |
1 |
.000 |
.817 |
.754 |
.885 |
Frailty |
.792 |
.035 |
525.386 |
1 |
.000 |
2.209 |
2.064 |
2.364 |
Robust/Pre-frailty/frailty |
||||||||
Age |
.058 |
.002 |
1077.900 |
1 |
.000 |
1.060 |
1.056 |
1.064 |
Sex |
-.449 |
.033 |
183.795 |
1 |
.000 |
.638 |
.598 |
.681 |
Marital status |
-.182 |
.041 |
19.724 |
1 |
.000 |
.834 |
.769 |
.903 |
Pre-Frailty |
1.278 |
.057 |
510.629 |
1 |
.000 |
3.588 |
3.212 |
4.009 |
Frailty |
.599 |
.053 |
126.720 |
1 |
.000 |
1.820 |
1.640 |
2.021 |
Abbreviations: B, Regression coefficients; SE, standard error; df, degree of freedom.
Due to the inconsistency of the different frailty status classifications, we reconsidered the FI as a continuous variable. We found that the curves of the FI at baseline and 7-year survival rate could be divided into two segments around an FI score of 0.5 (Figure 2), where the partial regression coefficients were 3.891 and 2.757, respectively. To further explore the effect of a unit increase in FI on the mortality risk, piecewise regression analysis was performed by segment within the FI score ranges of 0-0.5 and 0.5-1. When FI score was >0.5, the HR of mortality based on FI was 15.758 (95% CI: 3.656-67.924, P<0.001); however, when the FI score was ≤0.5, the HR was 48.944 (95% CI: 36.162-66.244, P<0.001). The female sex, living in the city, being married, and living with spouse were found to be protective factors, whereas ageing was a risk factor of frailty (Table 4).
Table 4. Piecewise Cox regression model analysis of the effect of frailty on mortality
|
B |
SE |
Wald |
df |
Sig. |
Exp(B) |
95% CI |
|
Lower |
Upper |
|||||||
FI_11 ≤0.5 |
|
|
|
|
|
|
|
|
Age |
.057 |
.002 |
916.001 |
1 |
.000 |
1.059 |
1.055 |
1.063 |
Sex |
-.491 |
.035 |
197.162 |
1 |
.000 |
.612 |
.571 |
.655 |
Residence |
-.141 |
.048 |
8.806 |
1 |
.003 |
.868 |
.791 |
.953 |
Marital status |
-.197 |
.043 |
21.259 |
1 |
.000 |
.821 |
.755 |
.893 |
FI |
3.891 |
.154 |
634.752 |
1 |
.000 |
48.944 |
36.162 |
66.244 |
FI_11 >0.5 |
|
|
|
|
|
|
|
|
Age |
.028 |
.006 |
21.437 |
1 |
.000 |
1.028 |
1.016 |
1.041 |
Sex |
-.242 |
.115 |
4.454 |
1 |
.035 |
.785 |
.627 |
.983 |
Marital status |
-.382 |
.146 |
6.789 |
1 |
.009 |
.683 |
.512 |
.910 |
FI |
2.757 |
.745 |
13.681 |
1 |
.000 |
15.758 |
3.656 |
67.924 |
Abbreviations: FI_11, frailty index in 2011; B, Regression coefficients; SE, standard error; df, degree of freedom.
Previous studies have investigated the relationship between the FI and mortality and predicted the mortality risk based on the static and dynamic FI [20,26]. However, the relationship between the frailty status and mortality risk has not been studied [31]. To examine the relationship of the FI and frailty status with survival time, we used Kaplan-Meier survival curves to determine whether the FI was more strongly associated with mortality than the frailty status, by calculating the AUCs. Previous studies have reported a correlation between the FI and the short-term mortality; furthermore, our findings demonstrated that the FI can be used to predict the 7-year survival rate [21].
Impairment in activities of daily living, chronic diseases, depression, poor lifestyle habits, and geriatric syndromes are known risk factors for frailty [32]. Similarly, our study revealed the following predictive factors of frailty: female sex, living in a city, being married, and living with a spouse. This is probably because the marital status and living conditions of older adults are related to their mental health and access to medical resources [33]. Previous research has shown a relationship between frailty and type of death; hence, we used survival analysis to evaluate the association between the FI and mortality. Our findings provide evidence that clinicians should perform frailty interventions to reduce preventable suffering before death; moreover, these interventions should be performed based on the known risk factors associated with the FI [22].
We further explored the relationship between the frailty status and mortality risk at baseline (2011) and during follow-up (2014 and 2018), with the aim of establishing suitable frailty interventions [29]. When examining the frailty-related mortality risk, we adjusted for demographic (gender and age), and sociological (education level, marital status, and living conditions) factors. When the frailty status was divided into non-frailty and frailty, education level was found to be a protective factor for frailty, besides the female sex, being married, and living with spouse, whereas ageing was a risk factor; this finding was probably because more education increases health literacy. Furthermore, we found that the HR for mortality was higher in pre-frail than in frail individuals, which provides evidence for the possibility of early intervention in pre-frail older adults.
Frailty, defined by phenotype or FI, was found to be significantly associated with an increased risk of all-cause mortality in community-dwelling Chinese older adults in previous studies[34,35]. Previous studies showed slightly different results of the relative mortality risk for different frailty levels owing to a lack of a unified frailty classification standard and inconsistencies in frailty status classification [36]. In the present study, we stratified the FI by grade rather than frailty categorization, to perform a more precise risk prediction, and to confirm whether 0.5 was the FI threshold. The mortality risk increased with age, and the female sex and being married were found to be protective factors of frailty, which was consistent with previous study findings [37]. Living in the city was found to be a protective factor of frailty when the FI score was less than 0.5, indicating that lifespan may be prolonged by exposure to advanced medications in the early state of frailty [38]. When the FI was greater than 0.5, the effect of frailty on mortality was relatively small because patients with the highest number of health deficits had the highest all-cause mortality rates [26]. A score of 0.5 was the risk threshold when the IF score was close to it, and the risk of death increased significantly with frailty under a score of 0.5.
In addition to increasing mortality risk, frailty is a predictor of negative health outcomes and all-cause mortality. The advancement of early interventions for mortality risk reduction is more beneficial in pre-frail than in frail patients. An FI score of approximately 0.5 constitutes an adequate intervention point, which provides suggestions for clinical practice [11].
AUC, area under the curve; CI, confidence interval; FI, frailty index; HR, hazard ratio; MMSE, mini-mental status examination; ROC, receiver-operating characteristic; B, Regression coefficients; SE, standard error; df, degree of freedom
Ethics approval and consent to participate
The Medical Ethics Committee of Tongji University, Shanghai, People’s Republic of China, approved the present study (approval number: 2022tjdxsy041). Written informed consent was obtained from all participants prior to the study.
Consent for publication
Not applicable.
Availability of data and materials
The CLHLS analyzed during our study are available in the Peking University Open Research Data, [https://opendata.pku.edu.cn/].
Competing interests
The authors declare no conflicts of interest.
Funding
National Key R&D Program of China (No. 2020YFC2008703)
Authors’ contribution
QC and RZ were involved in the research design and funding support. XZ drafted the manuscript and performed the computational analysis. JH and QC acted as corresponding authors, responsible for reviewing the content of the article. All authors read and approved the final manuscript.
Acknowledgement
We would like to thank Editage (www.editage.cn) for English language editing.
Author information
1. School of Medicine, Tongji University, Shanghai 200092, China
2. Department of Health Statistics, Navy Medical University, Shanghai 200433, China