Data and Study Participants
We used data from the Chinese Longitudinal Healthy Longevity Survey (CLHLS), an ongoing prospective, longitudinal study with the largest sample of the oldest old in China. Half of the counties and cities in 22 of the 31 provinces in China (covering 85% of the population) were randomly selected through a multistage cluster sampling approach. A wide range of socio-demographic, lifestyle, and health measures were collected in the CLHLS. The baseline survey was conducted in 1998 and participants who were alive were re-interviewed in each follow-up survey (2000, 2002, 2005, 2008–2009, 2011–2012, 2014, and 2017–2018). In 2011–2012, an ancillary study, in which a blood test was added, was conducted in eight longevity areas: Laizhou City in Shandong Province, Xiayi County in Henan Province, Zhongxiang City in Hubei Province, Mayang County in Hunan Province, Yongfu County in Guangxi Autonomous Area, Sanshui District in Guangdong Province, Chengmai County in Hainan Province, and Rudong County in Jiangsu Province. The Research Ethics Committees of Peking University and Duke University granted approval for the Protection of Human Subjects for the CLHLS. All study participants gave informed consent. A more detailed description of the recruitment strategy and study design of the CLHLS has been published elsewhere [11, 31, 32].
A total of 2,439 persons contributed blood sample in the ancillary study (2011–2012). Participants were excluded from the analytic sample if they had (i) incomplete data on any biomarkers for constructing AL (n = 251), (ii) no follow-up data (time to death or censorship was undetermined; n = 552), (iii) had extreme values on the biomarkers (n = 109), or (iv) were less than 60 years old (n = 16). The final analytic sample consisted of 1,519 participants. We did not observe appreciable differences in age, ethnicity, marital status, smoking, or chronic conditions between the analytic sample and those excluded (n = 920; Table S1). Compared to the analytic sample, excluded persons had higher education level and higher prevalence of exercise.
Calculation of AL Score
Based on previous research [2, 9, 14, 23] and availability of data in the CLHLS, we selected nine biomarkers to construct AL: heart rate, systolic BP (SBP), and diastolic BP (DBP), body mass index (BMI), total cholesterol, high density lipoprotein (HDL) cholesterol, glucose, triglyceride, and C-reactive protein (CRP). BMI, heart rate, SBP, and DBP were collected from physical examinations. BMI was calculated as body weight (kilograms) divided by height (meters) squared. SBP and DBP were measured by a mercury sphygmomanometer with an appropriately sized cuff, taken in the seated position after 5 minutes of quiet rest under the supervision of trained research assistants. We used the average of two measurements for further analyses. Blood samples were used for assays of the level of the total cholesterol, HDL cholesterol, glucose, triglyceride, and CRP.
To be in line with previous studies [6, 8, 24, 25], we used the highest quartile for heart rate, SBP, DBP, glucose, and CRP and the lowest quartile for HDL cholesterol to define high-risk group (coded 1). Because BMI, total cholesterol, and triglyceride were inversely associated with mortality among older adults, especially the oldest old [19, 29, 30], we used the lowest quartile to define high-risk group for these three biomarkers. For participants who self-reported having been diagnosed with hypertension and heart disease, we classified their SBP, DBP, and glucose into the high-risk category. Similarly, we classified participants’ glucose into the high-risk group if they self-reported having been diagnosed with diabetes. The cut-points of all nine AL components by men and women were presented in Table 1. We constructed the AL score based on the count of biomarkers falling in the high-risk group, ranging from 0 (lowest) to 9 (highest). We then classified the AL score into three categories based on sample distribution: 0–1 (low burden), 2–4 (medium burden), and 5–9 (high burden).
The outcome was all-cause mortality. Vital status and date of death (for persons who died by the end of the study) was ascertained by the close family member or village doctor of the deceased participant during the follow-up survey in 2014 and 2017–2018. We calculated the survival time from the date of the baseline interview to the date of last interview (censored) or the death date.
Demographic and lifestyle characteristics were collected by interview, including age, sex, ethnicity, education, and marital status, smoking status, and physical exercise. We divided ethnicity into Han and others (minority groups). Years of education were dichotomized as any (one year or more) and no education. Marital status was dichotomized as married and others (widowed, not married, and divorced). Cigarette smoking was categorized as current, past, and never smoker. Information of exercise was collected using the question “Do you do exercise at present?” and dichotomized into yes or no. Chronic conditions were measured based on self-reported physician’s diagnosis, including hypertension, diabetes, heart disease, stroke, pulmonary disease (including bronchitis, emphysema, pneumonia and asthma), arthritis, and cancer.
All analyses were conducted separately for males and females. We first presented the relative frequency of the AL score using histograms and calculated mean AL score. Then, we described the baseline characteristics of study sample by AL burden (low, medium, and high) using means and SDs for continuous variables and counts and percentages for categorical variables. Characteristics were compared across the three AL categories using analyses of variance for continuous variables and chi-square tests for categorical variables.
We calculated the death rates across three AL categories (low, medium, and high burden). We used the Cox proportional hazards model to determine the unadjusted and adjusted associations between the AL and all-cause mortality. Age, sex, education, and marital status were included in the demographically adjusted models; smoking status and physical exercise were added in the fully adjusted models. Because only about 6% of females were current or previous smokers, smoking status was modelled as a binary variable (never vs. current or previous) for females. We modelled AL both continuously and in categories. All tests were two-sided with a significance level of P-value less than 0.05. We conducted all analyses using STATA version 16.0 (Stata Corp, College Station, TX).