Study population
The Chinese Longitudinal Healthy Longevity Survey (CLHLS) is a prospective, nationally representative cohort study of older adults in mainland China, as described elsewhere in detail [20]. In brief, beginning in 1998, subsequent examinations of CLHLS were conducted every 2 to 3 years. To ensure representativeness, a multilevel cluster sampling method was used. In total, 631 cities or counties from 22 provinces were randomly selected, covering 85% of the Chinese population. Interviews were administered by qualified interviewers with a standardized questionnaire at the interviewees’ homes. For participants who were unable to answer questions, interviews were conducted with proxy respondents (e.g., spouses or close family members). However, questions regarding SWB were answered by the participants themselves. The CLHLS is a voluntary study, and participants were not provided with financial compensation or material rewards for their involvement. It was performed in accordance with the Declaration of Helsinki, and it was approved by the Duke University Healthy System’s Institutional Review Board, the National Bureau of Statistics of China, and the Ethical Committee of the Social Science Division of Peking University. Written informed consent was obtained from each participant.
The current study was based on six waves of the CLHLS from 2002 to 2018. The first wave (in 1998) of the CLHLS was excluded because participants aged 65–79 years were not surveyed in this wave. In total, 16,064 individuals were interviewed in 2002. Of these, 13,781 responded to the SWB items and were considered eligible for analysis. We excluded participants who had missing values for lifestyle factors (n = 382), demographic characteristics (n = 105) and physical comorbidities (n = 12). As the proportion of missing data among eligible participants was less than 5% of data, thus we did not perform imputation. Finally, 13,282 participants were included in further analyses. The baseline characteristics of participants included and excluded in our study were compared in supplementary Table S1. The participants excluded in our study tend to be older, be female, be widowed/separated/others, be illiteracy, and are less likely to have hypertension, diabetes, cancer and heart disease.
Subjective well-being
SWB was measured with a previously validated tool [12] as described in our earlier methodology [11]. The tool included eight items1) life satisfaction, 2) optimism, 3) conscientiousness, 4) anxiety, 5) loneliness, 6) personal control, 7) feeling of uselessness, and 8) happiness. The questions used to assess these items are provided in supplementary Table S2. Each item was coded as 1, 2, 3, 4, or 5, and negative items were reverse coded. The total SWB score was calculated by summing scores across all items, varying from 5 to 40, with a higher score indicating better SWB. Participants whose SWB scores were above the third quartile (≥ 31) were classified as having better SWB, while the others were regarded as having worse SWB, following previous studies [12, 21]. The Cronbach’s α for the internal consistency of the SWB scale was 0.68 and the confirmatory factor analysis by a previous study suggested this SWB scale has good validity among the older population [12].
Healthy lifestyles
Lifestyles of interest in the current study included dietary diversity, smoking status, alcohol consumption, exercise, and BMI. Dietary diversity data were collected at baseline based on a food frequency questionnaire including seven major food groups (fruits, vegetables, fish, eggs, meat, beans, and tea). Responses for each food group were recoded as “often or almost every day”, “occasionally”, or “rarely or never”. Participants who answered “often or almost every day” received one point, the others were scored as zero point. The total dietary diversity score was calculated via summing the points across all food groups, ranging from 0 to 7, with higher score indicating better dietary diversity. Individuals with a total dietary diversity score at or above the mean value were classified as having good food dietary diversity, following previous studies [9, 22, 23].
Body weight was recorded to the nearest kilogram with participants wearing light clothing. Height was estimated using knee height, measured as the vertical distance from the sole to the upper surface of the knee with both the knee and ankle flexed at a 90° angle. A validated equation was applied: height = 67.78 + 2.01 × knee height for men and height = 74.08 + 1.81 × knee height for women [24]. BMI was calculated by weight in kilograms divided by the squared height (m [2]), and was classified into four groups: underweight (BMI < 18.5 kg/m2), normal weight (18.5 kg/m2 ≤ BMI < 24.0 kg/m2), overweight (24.0 kg/m2 ≤ BMI < 28.0 kg/m2) and obese (BMI ≥ 28.0 kg/m2) [25]. The question “do you exercise regularly at present?” was used to assess the exercise, and the response was recorded as yes or no. Smoking status (never, current and former) and alcohol consumption (never, current and former) were self-reported at baseline. Individuals who had good dietary diversity, never smoked, never drank, exercised regularly, and had normal weight were categorized as healthy. For each lifestyle, participants received a score of 1 if the individuals met the criteria for health, or 0 if they did not. The total lifestyle score was calculated by summing the points across all lifestyle factors for a maximum score of 5, with higher scores indicating a healthier lifestyle [8, 26]. We classified the participants into the following three lifestyle groups: unhealthy (0–1), intermediate (2–3) and healthy (4–5), following previous studies [9].
Mortality
To ensure the completeness and accuracy of mortality data, interviews with close family members or village doctors were conducted in each follow-up survey in 2005, 2008, 2011, 2014, and 2018. The individuals who survived in 2018 wave or who were lost-to follow-up were defined as censored cases. Participants who could not be reached or contacted after more than three reasonable efforts were defined as ‘loss to follow-up’. Previous studies have suggested that data on mortality in CLHLS are of high quality and have been widely used [27, 28].
Covariates
Covariates were collected using a standardized questionnaire at baseline by qualified interviewers. Demographic characteristics included age, sex, educational level (illiteracy [no education], literacy [≥ 1 year education]), marital status (married, widowed/separated/singled/others), and place of residence (urban/city, rural). The urban/city and rural area were classified based on the definitions provided by the National Bureau of Statistic China. Physical comorbidities were defined based on a self-report of doctor’s diagnosis, including hypertension, diabetes, heart diseases, cerebrovascular diseases and cancer.
Statistical analysis
Baseline characteristics were presented across different levels of healthy lifestyles, and differences among groups were compared by analysis of variance for continuous variables and chi-square tests for categorical variables. Cox proportional hazard regression models were fitted to estimate the hazard ratios (HRs) and 95% confidence intervals (CIs) of all-cause mortality associated with SWB and healthy lifestyle scores. The proportional hazards assumption was assessed graphically, and no evidence of obvious departure from this assumption was detected. The person-years were calculated from baseline until the date of death, loss to follow-up, or the end of follow-up, whichever came first. Three models were fitted: model 1 was unadjusted to calculate the crude HR and 95% CI for the SWB and lifestyle score with total mortality; model 2 was adjusted by age and sex; model 3 was further adjusted for marital status, place of residence, educational level, and physical comorbidities (including hypertension, diabetes, heart diseases, cerebrovascular diseases, and stroke); and model 4 was mutually adjusted for SWB and healthy lifestyle score. In these analyses, the healthy lifestyle (with a lifestyle score of 4–5) and the better SWB group (SWB score ≥ 31) were regarded as the reference group.
Stratified analyses were further carried out to investigate the associations of SWB with all-cause mortality among elderly Chinese individuals in different lifestyle groups, and vice versa. To examine the joint associations, we classified the participants into six groups according to SWB (better, worse) and lifestyle score (healthy, intermediate, unhealthy) and estimated the HRs of all-cause mortality in different groups, in comparison with those with better SWB and healthy lifestyles.
Several sensitivity analyses were further conducted to test the robustness. We repeated the analyses for the combination of healthy lifestyles and SWB under the following scenarios. First, to reduce reverse causation, we excluded those who died within the first year of follow-up because those who died quickly may have pre-existing conditions influencing their lifestyles, SWB and mortality risk. Second, we excluded those aged 85 years and above because they are likely to have a higher risk of death than their younger counterparts. Finally, we excluded those with potentially life-threatening diseases, including heart disease, cancer and cerebrovascular disease because both SWB and lifestyles could be influenced by major chronic diseases. All analyses were carried out using SPSS, version 26.0 and R software version 3.6.1, and two-sided p values < 0.05 were considered to be significant.