Study design and population
This study was performed within CLHLS, an ongoing prospective population-based cohort study conducted in half of the countries and cities in 23 provinces in China and designed to investigate the determinants of healthy aging among Chinese older adults. Further details about the study design have been previously described elsewhere42. In brief, CLHLS was established in 1998 and conducted follow-up surveys and further recruitments of new participants in 2000, 2002, 2005, 2008–2009, 2011–2012, 2014, and 2018. The CLHLS received approval from the Ethics Committee of Peking University (IRB00001052-13074). Written informed consent was obtained from all participants or their legal representatives during the face-to-face interview.
Of the total participants recruited from 8 waves of CLHLS, 37,456 participants completed the baseline and follow-up examinations. We further excluded participants with missing information on lifestyle score at baseline (n = 671), those aged less than 65 years old (n = 297), those with cognitive impairment at baseline (n = 9649), and those without available MMSE measurements at baseline or any follow-up surveys (n = 8028). Finally, 18,811 participants remained in the current analysis, and of those, 6301 participants with genotyping data were included for the joint association between healthy lifestyle and genetic risk analysis (Supplementary Fig. 6).
Lifestyle factors
Data on demographic characteristics, socioeconomic status, lifestyle factors, physical health, and psychological well-being were assessed by well-trained interviewers via the standardized questionnaire at baseline. We derived a healthy lifestyle score based on four modifiable lifestyle factors associated with cognitive function according to prior evidence30,38,39,43,44, including current non-smoking, never alcohol drinking, active physical activity, and healthy diet intake. Supplemental methods and Supplementary Table 6 provide additional details on the specific questions asked and the construction of healthy lifestyle scores.
Specifically, current non-smoking was referred to as never smoking or former smoking according to previous studies15. Alcohol drinking status was classified as non-drinker, current drinker, and former drinker. The category of never drinking was deemed a healthy lifestyle factor30. For physical activity, participants reported the frequency of participation in regular exercises, housework tasks, personal outdoor activities, gardening, rearing domestic animals/pets, reading, playing cards/mahjong, watching TV/listing to the radio, and attending social activities45. The frequency of “almost every day”, “occasionally”, and “rarely or never” was scored 2, 1, or 0, respectively. A total physical activity score was calculated as the sum of 9 activities, which is scored from 0 to 18 with a higher score indicating a greater level of physical activity. An ideal physical activity was defined as a physical activity score in the top 40% of cohort distribution46. Dietary intake was assessed by using a standardized food frequency questionnaire with acceptable reproducibility and validity7, including 9 commonly consumed food groups in the Chinese diet: fresh vegetables, fresh fruit, legumes, meat, eggs, fish and seafood, salty vegetables, tea, and garlic)7,47. Insufficient total daily protein intake and imbalance in protein synthesis and degradation were observed with aging among older adults, especially the oldest old. Thus, consumption of typical protein-rich food such as legumes, meat, eggs, and fish, which have prominent beneficial effects on mortality, was also considered as a healthy lifestyle in this study37,48. A total diet score was computed in the same way as physical activity. The diet intake was deemed ideal based on the top 40% of the cohort distribution in line with previous studies46. All component scores were summed to obtain the healthy lifestyle score ranging from 0 to 4, with a higher index indicating a healthier lifestyle, and were further categorized as “unfavorable” (less than three healthy lifestyle factors), “intermediate” (three healthy lifestyle factors), and “favorable” (four healthy lifestyle factors).
Covariates
Covariates were selected based on prior research and available cohort measures15,49. The Supplementary Information describes covariate details inquired by questionnaires, including age, sex, educational level, area of residence, marital status, occupation, source of income, self-reported health status, optimism status, and history of major chronic disease. Optimism status was evaluated with seven items (score range, 0–7, with higher scores denoting a higher level of optimism)50. Presence of major chronic diseases, including cardiovascular disease, diabetes mellitus, hypertension, respiratory disease, digestive system disease, or cancer, was self-reported.
Cognitive function
Cognitive function was assessed by an adapted Chinese version of the MMSE at baseline and during each follow-up survey according to the standard protocol29. The MMSE included 24 items regarding six cognitive dimensions: orientation, attention and calculation, visual construction, language, naming, and recall skills (Supplementary Table 7)29. A score of zero was given for incorrect and unknown answers, and one point was given for correct answers. All questions were of equal weight, culminating in a total possible score of 30. Cognitive score and its individual dimensions were transformed from raw score to Z score using the means and SDs at baseline15. A positive Z score indicates better cognitive function than the mean population score, and a negative score indicates a poor cognitive score. Higher cognitive score indicates better performance. Cognitive score was computed on each cycle for all study participants, and based on the cognitive score during the follow-up, we determined the rate of cognitive decline. Due to a high proportion of the study participants not having a formal education, cognitive impairment was defined as an overall cognitive score of <18 according to previous studies29.
Genotyping and genetic risk score calculation
According to the results of the CLHLS genome-wide association study on longevity, a replication study was carried out among 13,228 individuals using a well-designed and customized chip targeting 27,656 single nucleotide polymorphisms (SNPs) previously associated with longevity and its related traits51. Of these selected SNPs, 3966 SNPs were associated with common diseases of old age, i.e., Alzheimer’s disease, cardiovascular disease, type 2 diabetes mellitus, cancer, and immune-related diseases. Detailed information on SNPs selection and genotyping process used in the CLHLS study has been published previously51,52. The polygenic risk score in the present study was calculated based on the number of risk alleles of 34 genetic variants increasing the risk for Alzheimer’s disease as previously published51,53. Details regarding the selected SNPs are provided in Supplementary Table 8. Individual SNPs were coded as 0, 1, and 2 according to the number of risk alleles. The weight coefficient for each SNP was reported in previous studies51,52. The genetic risk score was formulated as the sum of the number of risk alleles at each locus multiplied by the respective weight coefficient. The genetic risk score was categorized into two groups according to median: low and high genetic risk groups.
Statistical analyses
Baseline characteristics of participants were described as mean values (standard deviation) for continuous variables and numbers (percentage) for categorical variables. The marginal mean values for cognitive score and its individual dimensions according to lifestyle categories were estimated by multiple linear regression models with adjustment for baseline covariates.
Linear mixed-effects models were used to test the association of a healthy lifestyle with longitudinal change in cognitive function and each cognitive domain separately. To account for within-individual associations between repeated measurements, all models included time since baseline as well as random intercepts and random slopes of time. Analysis was adjusted for age (years, continuous), sex (female/male), entry time, educational attainment (years of schooling completed <1 year, 1–6 years, or >6 years), area of residence (urban/rural), current marital status (in marriage, not in marriage), occupation (agriculture/forestry/husbandry/fishery, commercial, service, industrial worker/self-employer, professional/governmental/managerial personnel, or houseworker/never worked/other), source of income (independent, dependent), and baseline cognitive score (continuous). Same method was used to evaluate the association between genetic risk and the rate of cognitive decline. We further assessed the association of a healthy lifestyle with cognitive decline stratified by genetic risk, and the interaction between a healthy lifestyle and genetic risk was also evaluated. Additionally, the joint associations of healthy lifestyle and genetic risk with the rate of cognitive decline. Participants were classified into six categories according to lifestyle group (unfavorable, intermediate, and favorable) and genetic risk (low, high), with the combination of high genetic risk and unfavorable lifestyle group as the reference.
With time-on-study as timescale, person-years were calculated for each participant from baseline to the date of cognitive impairment, death, or end of follow-up, whichever came first. Cox proportional hazard models were used to estimate the associations of lifestyle and genetic factors with incident cognitive impairment, adjusted for age, sex, entry time, educational attainment, area of residence, current marital status, occupation, and source of income. The proportional hazards assumption was examined by the Schoenfeld residuals method, and no violation was identified (P > 0.05). We also assessed the influence of death as a competing risk for cognitive impairment via competing risk analyses.
To test the robustness of our results, we performed several sensitivity analyses. First, in addition to primary adjustment, we further controlled for self-reported health status (yes/no), optimism status (continuous), and history of chronic disease status (yes/no) in the multivariate mixed model. Because these factors were identified as risk factors for cognitive function15. Second, to explore the possibility of reverse causation due to impaired cognitive function, which might influence the accuracy of reported lifestyle behaviors, participants with baseline MMSE scores below the 10th percentile were excluded34. Additionally, changes in cognitive score were truncated at the 0.5th and 99.5th percentiles to minimize the influence of outliers. Third, to address the adverse effects of former smoking, we created a new healthy lifestyle score using “never smoking” as a healthy lifestyle factor. Fourth, due to the common occurrence of missing values for specific items of the MMSE test, such as visual construction, the analysis was restricted to participants who completed all the MMSE items41. Fifth, as lifestyle scores might change during follow-up, the latent class trajectory model was used to identify distinct lifestyle score trajectories, and then to assess the effect of lifestyle trajectory groups on cognitive decline. Finally, to account for the non-linear relationship, quadratic terms of time were included in the multivariate model15. Two-sided P < 0.05 was accepted as statistically significant, except for separate analyses for individual domains of cognition in which the Bonferroni correction was applied to account for multiple testing (P < 0.008 considered significant [=0.05/6]). All data analyses were performed in SAS V.9.4 (SAS Institute, Cary, NC, USA), and R V.4.3.1 (R Foundation for Statistical Computing, Vienna, Austria).
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.