Adherence to a healthy lifestyle counteracts the negative effects of risk factors on all-cause mortality in the oldest-old

Characteristics of study population

The 17660 participants (mean age 92.7 years) included 10758 women and 6902 men. During the follow-up, 11094 (62.8%) in total, and 4298 (62.3%) men and 6796 (63.2%) women died. Table 1 shows the characteristics of study population by gender. Compared with males, females were older, more likely not to be married, to be illiterate and underweight, and to live in a rural residence, to have a lower economic level, a lower systolic and diastolic blood pressure (BP), a higher heart rate, a physical disability, impaired cognitive function, an unhealthy diet, and less comorbidity and physical exercise, and tended to be non-smoker and non-drinker.

Risk factors, healthy lifestyle and mortality

We found that rural residence, not in marriage, lower economic situation, physical disability, impaired cognitive function, and comorbidity were independently associated with higher mortality. On the other hand, never smoking, never drinking, physical exercise, ideal diet, and normal weight were associated with lower mortality (Figure 1). In assessing the adverse effect of the five unhealthy lifestyles on mortality, we found that former and current smoking, former drinking, never exercising, and underweight were associated with increased mortality, while there was no significant association between current drinking and overweight with mortality (Supplementary Table 1). Moreover, systolic BP and diastolic BP were not associated with mortality, neither in males nor in females (P > 0.05) (Supplementary Table 2). When taking into account the effect of adverse health conditions on mortality, we excluded the oldest-old with impaired cognition, disability and comorbidity; nonetheless, we found similar independent risk factors related to mortality for all other adverse health conditions considered (Supplementary Table 3).

Joint effect of risk factors and healthy lifestyle on mortality

We generated a composite “weighted risk score” of mortality by gender, ethnicity, residence, marital status, economic situation, physical activity, cognitive health, and comorbidity together, ranging from 0 to 39. In contrast, a “weighted protection score” against mortality was composed of healthy factors such as never smoking, never drinking, physical exercise, ideal diet, and normal weight, with the range being 0–9.

Cox models analyses showed that our risk score correlated positively with risk of mortality (HR: 1.065, 95% CI: 1.061–1.068) while our protection score was inversely correlated with mortality (HR: 0.94, 95% CI: 0.93–0.95) (Table 2). We divided the risk and protection scores of all participants into three groups according to their tertiles. Compared with the lowest tertile, the HR of mortality for the oldest-old in the high and middle risk score groups was 2.38 (2.27–2.50) and 1.46 (1.38–1.53), respectively. On the other hand, the HR of mortality for people in the high and middle protection score groups was 0.74 (0.70–0.77) and 0.91 (0.87–0.96), respectively.

In assessing the joint effect of risk factors and healthy lifestyles on mortality, we found that participants in the combined highest risk score and lowest protection score profile had a nearly threefold risk of mortality compared with the lowest risk and the highest protecting score profile. In the highest risk score profile, the risk of mortality obviously increased with decreases in protection score (high score: HR=2.97, 2.76–3.20; middle: HR=2.82, 2.62–3.03; low score: HR=2.56, 2.38–2.75) (Figure 2, Supplementary Table 4). The trend of increasing risk of mortality with decreasing protection scores was consistent with median and low risk score profiles. Furthermore, the association of risk score profile with mortality differed when stratified by protection score profile (P for interaction = 0.018).

The harmful effect of risk factors on mortality was counteracted by healthy lifestyle

Kaplan-Meier survival analysis showed that the median survival time in the joint groups of low risk and high protection, low risk and low protection, high risk and high protection, high risk and low protection, was 4.8, 4.2, 2.8 and 2.5 years, respectively (P for log-rank < 0.001), suggesting that high protection scores may prolong survival in the oldest-old with high risk scores (Figure 3A). Figure 3B shows the offsetting effect of adherence to a healthy lifestyle on the association between risk factors and the risk of mortality. In the low risk profile, high protection scores can further reduce the risk of mortality (P < 0.0001). Similarly, mortality was strongly attenuated by high protection scores among the high-risk profile (P = 0.0003). High (middle and highest) protection scores counteracted the negative effect of high (middle and highest) risk scores on mortality by 23%. The proportion of incident death cases that was attributable to a high protection score profile was 0.26 (95% CI 0.24-0.27) among participants with a low risk profile, and 0.47 (95% CI 0.43-0.52) among those with a high risk factor profile. These findings suggest that death among the oldest-old with a high risk profile might be preventable in 26% of cases by having a healthy lifestyle (and thus a high protection score), implying not smoking, not drinking, ideal diet, exercising, and normal weight. Moreover, for the oldest-old with high risk profiles, having a high-protection score can prevent death in 47% of cases.

Strengths and limitations

Our study has several strengths. It is a longitudinal study based on a very old population of China, and participants were recruited from a large general population, which allows for some level of generalization of our results. Furthermore, we tried to include multiple types of potential influence factors to counterbalance each other, allowing to examine the effect of independent risk factors and the benefits of healthy lifestyle practices on mortality more precisely. Despite the strengths of the study, there are several limitations. Firstly, although we identified many independent risk factors of all-cause mortality in the oldest-old in terms of demographic characteristics, health conditions, cardiovascular factors, and lifestyle, many other factors are not fully taken into consideration in this study. Secondly, people who survive to age 80 could already represent a group of mostly healthy individuals; indeed, the number of participants with comorbidity ≥ 2 was only 3.6% of our cohort. Therefore, we calculated a composite comorbidity score for self-reported major disease of the oldest-old. Thirdly, due to limited information, we did not include quantified alcohol intake or exercise intensity; rather, we divided participants into three groups in terms of drinking current drinkers, former drinkers, and never drinkers, depending on the time of alcohol consumption through their lifetime.

Study population

This study draws on data from the oldest-old participants (aged over 80 years) from the 2005 and 2014 waves of the Chinese Longitudinal Healthy Longevity Surveys (CLHLS). The CLHLS is a nationwide survey done in a randomly-selected half of the counties and cities in 22 of the 31 provinces in China. A multistage cluster sampling approach was used in this prospective, longitudinal, community-based study. The CLHLS attempted to interview all centenarians who voluntarily agreed to participate in the study in the sampled counties and cities. Details of the sampling procedure and descriptions of CLHLS are available elsewhere [7, 15, 22]. All participants were interviewed about demographic characteristics, medical history, lifestyle, and health behaviors, and physical examinations with a standardized questionnaire and relevant instruments after obtaining informed consent from them. For the current study, we used data over a 10-year period with four assessments. The participants were initially recruited in 2005 (N=15,638), 2008 (N=9,479) and 2011 (N=1,360), and followed up until 2014. Excluding individuals under 80 or above 105 years of age (N=7,630), inaccurate date of death (N=140), and missing data (N=1,047), a total of 17,660 participants (aged 80-105 years) were eligible for the study.

Data collection

Data on demographic characteristics (age, gender, residence, marital status, education level, economic situation), health conditions (cognitive function, physical disability, comorbidity), cardiovascular factors (systolic BP diastolic BP, heart rate), and lifestyle (smoking, drinking, diet, physical exercise, BMI) were collected through structural interviews and physical measurement. Given the high illiteracy rate, auditory and visual impairments, and other inconveniences (especially among centenarians), some of the oldest-old agreed to participate in the study through proxy assistance by a close family member. No proxy was used for objective questions such as assessment of cognitive function and physical disability.

Cognitive function was assessed by the Mini-Mental State Examination (MMSE) [31], which has been widely applied in epidemiological studies. We classified participants in different cognitive function categories depending on their MMSE scores (which ranged from 0 to 30 points), as follows: no cognitive impairment (25-30), mild cognitive impairment (18-24), moderate cognitive impairment (10-17), and severe cognitive impairment (0-9) [32]. Physical disability was identified by Activities of Daily Living (ADL), which was measured by the participant’s self-reported results in six self-care tasks consisting of bathing, dressing, eating, indoor transferring, toileting, and continence based on the Katz index scale [33]. Physical disability was defined as a need for assistance or a difficulty in one or more of the six activities listed above [34]. Furthermore, a healthy diet was determined by the consumption of fruit, vegetables, and fish and abstinence from meats. We defined a healthy diet as adherence to at least two of the healthy food items listed above [29]. According to the self-reported chronic disease status of participants, we computed a comorbidity score taking into account instances of diabetes, heart disease, stroke, asthma, and cancer (categorized as ≥1 and =0), diagnose by a specialist doctor.

Identification of death

Mortality status was ascertained during the follow-up survey in 2014, assessing whether subjects died and the date of death, completed the study, or were lost to follow-up. Information about death was ascertained and affirmed by a close family member or village doctor.

Statistical analyses

We summarized the participants’ baseline characteristics using descriptive statistics, reporting the mean and standard deviation (SD) of normal distribution or median and interquartile ranges of non-normal distribution for continuous variables, and proportions for categorical variables. We compared the baseline characteristics by all-cause mortality using chi-square test for categorical or student t test for continuous variables.

We calculated the follow-up time from the date of enrollment to the date of death, or loss to follow-up, or the end of the follow-up period, whichever came first. A “lost to follow-up” status was designated to those who could not be found and contacted. Participants who survived or were lost to follow-up were censored in 2014. The Cox proportional hazards regression model was used to calculate the hazard ratios (HRs) and 95% confidence intervals (CIs) of mortality in relation to risk factors and healthy lifestyle. Schoenfeld residuals were used to assess whether proportionality assumptions were satisfied, the results of which suggested that the assumptions were not violated. Age, gender, residence, ethnicity, education level, economic situation, marital status, systolic BP, diastolic BP, heart rate, physical disability, cognitive function, comorbidity, smoking status, drinking, exercising, diet, and BMI were adjusted for each other. We also did sensitivity analyses for which the oldest-old with impaired cognitive function, disability, and comorbidity were excluded to avoid the adverse causality of worsening health conditions on mortality.

According to the effects of each factor on all-cause mortality, we classified gender, ethnicity, residence, poor economic situation, unmarried status, physical disability, cognitive function, and comorbidity as independent risk factors for mortality, while never smoking, never drinking, physical exercise, ideal diet, and normal weight were classified as healthy lifestyle practices that decreased the risk of mortality. Then we created two composite scores to investigate the joint effect of independent risk factors and healthy lifestyle practices on mortality. The composite score was calculated in three steps [35, 36], as follows: 1) run a multivariable Cox regression model that includes all potential risk factors and healthy lifestyle practices to estimate the effect of each factor independent of potential confounders; 2) calculate the individual risk point for each response category of each variable by dividing the respective regression coefficient (β coefficient) with a single constant, which represents the regression coefficient for a one year increase in age in relation to risk of mortality; 3) round the risk points to the nearest integers and calculate the composite score by summing them up (Supplementary Table 5). Furthermore, independent risk factors and healthy lifestyle scores were treated as both continuous variables and categorical variables (tertiles) in the Cox regression model to test the dose-response relationship of risk/lifestyle factors with mortality. We also tested the statistical interactions between risk factors and healthy lifestyle scores in relation to mortality in the Cox model.

To further investigate whether and to what extent healthy lifestyle practices can counteract the harmful effect of the independent risk factors on mortality [28], we created a variable with four categories, which combines the tertiles of risk factor score (high level: top tertile vs low level: middle/bottom tertiles) with the tertiles of the healthy lifestyle score (high level: middle/top tertiles vs low level: bottom tertiles). This combined variable divided participants into four groups according to their risk and healthy lifestyle profiles: 1) low risk & high protection scores; 2) low risk & low protection scores; 3) high risk & high protection scores; (4) high risk & low protection scores. To quantify the contribution of joint risk factors and healthy lifestyle to mortality, we calculated the population attributable fraction (PAF), which is the estimated proportional reduction in mortality that would occur if risk factors and unhealthy lifestyle practices were prevented. We used STATA 15.0 for all statistical analyses. All statistical tests were two sided, and we considered P<0.05 to determine statistical significance.