Study design and participant
The Chinese Longitudinal Healthy Longevity Survey (CLHLS) is a nationwide survey with the largest sample of participants aged 80 years and above in the world. Till recently, the program randomly selected half of the cities and counties in 23 provinces of China. It began in 1998, with subsequent follow-up and recruitment of new participants in 2000, 2002, 2005, 2008, 2011, 2014 and followed up till 2018. A more detailed description of the CLHLS has been published elsewhere . The study design of CLHLS and the enrollment of participants are described in Supplementary Materials 1 and 2.
The Protection of Human Subjects for the CLHLS was approved by the biomedical ethics committee of Peking University and was conducted in accordance with the principles of the Declaration of Helsinki.
Definition of cognitive decline
Cognitive decline was evaluated using the Chinese version of Mini-Mental State Examination (MMSE), a widely used cognitive test . Cognitive decline was defined as a decline to a lower MMSE category between 2-3 year intervals from baseline . Specifically, baseline MMSE scores were classified into four categories: severe cognitive impairment (0-17), mild cognitive impairment (18-23), low normal cognitive function (24-27) and high normal cognitive function (28-30). When high normal cognitive function declined to 0-3 scores, 4-7 scores, 8-14 scores and below 14 scores, they were defined from high normal cognitive function decline to low normal cognitive function, to mild cognitive impairment, and to severe cognitive impairment, respectively. Similarly, to decline to different stages from high normal cognitive function, there were six types of cognitive decline. Participants in the same cognitive category at both time points and those who maintained the same MMSE scores and transitioned to higher MMSE scores were categorized as maintained cognitive function. Thus, there were four types of maintained cognitive functions. Detailed division are described in Supplementary Material 3. Also, cognitive decline was defined as continuous variable according to the rate of change in MMSE score, which was calculated as the difference between at the baseline cognitive function test and the second cognitive function test divided by follow-up times ((MMSE score at baseline−MMSE score at the second cognitive function test)/the interval between two follow-ups, years).
All-cause mortality and cause-specific mortality
The main outcome was all-cause mortality occurring during the follow-up survey from 1998 to 2018. Survival status was ascertained from family members or relatives of the oldest old during the follow-up survey in 2018 to assess whether the subjects completed the study, died and the date of death or could not be traced during follow-up. The oldest old who “lost to follow-up” was not be found and contacted. The oldest old who survived or lost to follow-up were defined as censored data. Cause-specific (CVD or non-CVD) mortality was ascertained by local doctors during the follow-up survey for deceased.
A standardized questionnaire was designed to collect data involving the following variables, demographic characteristics, socioeconomic status, lifestyle factors, leisure activities and health conditions: (1) demographic characteristics included sex (men or women) and age (as continuous variable). (2) socioeconomic status included residence (urban or rural), educational background (illiterate or not), current spouse status (have spouse or have no spouse ) marry status (in marriage or not), and living pattern (with family members or not), (3) lifestyle factors included regular exercise(yes/no), current smoke status (yes/no), current drink status (yes/no), dietary diversity (DD) (yes/no), (4) leisure activities were divided into 3 categories (never, sometimes, and often) included doing housework, reading, watching TV and listening to the radio, keeping pet and growing flowers. (5) health conditions included high BP (SBP > 140mmHg or DBP > 90mmHg, yes or no), disability in ADL (yes/not), hypertension (yes or no), and respiratory disease (yes or no). More details are described in Supplementary Material 4.
This study reported the hazard ratio (HR) and 95% confidence interval (CI) using Cox proportional hazards model. Stepwise regression was used to determine the independent risk factors of mortality and the important confounders identified by previous studies. Several models were developed: model 1 adjusted for demographic characteristics, model 2 adjusted for the variables in model 1 plus socioeconomic status and lifestyle factors, and model 3 adjusted for the variables in model 2 plus leisure activities, and model 4 adjusted for the variables in model 3 plus health conditions. To exclude the co-linearity, model 4 did not include hypertension. The maintenance of high normal cognitive function was defined as the reference. Kaplan-Meier analysis was used to draw the survival curves according to cognitive status, and the survival curves were compared by the log-rank test. This study tested the suitability of the proportional risk assumption using hypothesis tests based on Schoenfeld residuals and the proportional hazards assumption was not severely violated (Supplementary Material 5, Schoenfeld P =0.08). Follow-up time in years was used as the time axis since enrollment. Additionally, the missing data was less than 1.1% for covariates and mean value imputation methods were applied to correct for the missing covariate values.
Potentially modifiable risk factors were estimated by testing for interactions between cognitive decline-all-cause mortality association and potentially modifiable factors. These modifiable factors included age, sex, BP, hypertension and CVD mortality. Age and BP were the key risk factors and the main research targets in the association between cognitive decline and all-cause mortality and so, this study explored them by cross-stratifying with age-at-enrollment and BP strata (high BP, age 80-89 years (octogenarians), high BP, age ≥90 years (nonagenarians), non-high BP, 80-89 years, non-high BP, age ≥90 years). In addition, to differentiate from previous studies, we included the younger age group (oldest old range: 65-79) to verify our hypothesis in appendix materials.
In the further analysis, we conducted the following various analytical strategies to check the robustness of the primary results: (1) Excluding the oldest old whose cognitive scores increase to a higher MMSE category (2) Stratified analyses were performed, excluding comorbidities (hypertension, heart disease or respiratory disease). (3) Excluding mortality that occurred in the first 0.5, 1 and 1.5 year, due to the possibility that the drops in cognitive performance before mortality and/or disease condition in the last year of life might influence the results. (4) Additionally, to evaluate whether the associations differ for different follow-up times and reverse causation, we stratified across time strata by median (3 years) follow-up periods.
Data analysis was conducted using R version 3.3.4 with package of “survival”. All statistical tests were 2-sided, and statistically significant was judged by P-values < 0.05.