The data used in this study were extracted from the CLHLS, a prospective, community-based cohort study with a median 5.4-year follow-up period (2011–2017), details of which are available elsewhere (19, 20). In brief, participants were enrolled from the sixth wave (2011) of CLHLS assessments that are performed in eight Chinese regions (including Yong Fu county, Ma Yang county, Chen Mai county, Rudong county, Xia Yi county, Zhong Xiang city, Shanshui city, and Lai Zhou city). The standards of Chinese longevity regions as follows: the existing centenarians in the region account for more than 7.0/100,000 of the total population; the average life expectancy of the regional population is 3 years higher than the national level, and the proportion of the older adults aged ≥80 years accounts for more than 1.4% of the total population. We obtained the baseline health examination and blood test data of all adults aged ≥65 years surveyed within the study period. After excluding cases with missing information on key variables, the final sample consisted of 1,785 older adults (Figure 1). Participant information was systematically collected during face-to-face interviews conducted by trained staff. The study was approved by the Research Ethics Committee of Peking University (IRB00001052-13074) and informed consent of all participants has been obtained.

Medical professionals collected blood samples from all eligible participants. Data on the blood-related variables of interest were obtained from the baseline survey (2011–2012). The blood samples were transported and stored at −80°C in Beijing. HB, total cholesterol (TC), C-reactive protein (CRP) levels, and serum creatinine (Scr) levels were measured using the HB cyanide test, enzymatic colorimetric test, immunoturbidimetric assay, and enzymatic method, respectively (21).

To assess whether the risk of mortality changed at different HB concentrations in older adults form longevity regions, we categorized HB concentrations into five categories (<11.0 g/dL, 11.0–11.9 g/dL, 12.0–12.9 g/dL, 13.0–13.9 g/dL, and ≥14.0 g/dL). An HB concentration of 13.0–13.9 g/dL was selected as the reference group for comparison based on previous studies (12).

Data on variables that may confound the relationship between HB concentrations and all-cause mortality were extracted from answers to the questionnaire and biochemistry tests (22). The covariates of interest were demographic characteristics (age, sex, residence, education level, marital status, and economic status); lifestyle factors (smoking status, drinking status, tea drinking status, food diversity score, and regular exercise uptake); disease and psychological conditions; history of chronic diseases (hypertension, diabetes, stroke, heart disease), depressive symptoms score; biochemical test findings [body mass index (BMI), white blood cell (WBC) count, the mean corpuscular volume (MCV) estimates, the levels of CRP and TC, platelet count, and eGFR]; findings from the assessment of the activities of daily living (ADL); and the Mini-Mental State Examination (MMSE) score. To reduce the potential impact of inferential bias, we used multiple imputation methods to correct for missing values of the relevant covariates (23).

The survival outcomes of the participant were confirmed in the seventh (2014) and eighth (2017) waves of the CLHLS. The date of death was confirmed by the participants' next-of-kin and local physician. We calculated the survival time from the interview date at baseline in 2011 to the date of death. “Lost to follow-up” status applied to individuals who could not be found or contacted. The data for the participants who were alive until the 2017 survey were censored at that date.

The variables were summarized to describe categorical variables as number (percentage), normally distributed variables as mean [standard deviation (SD)], and skew distribution variable as median (interquartile range). The Cox proportional hazards model was presented to evaluate hazard ratios (HRs) and 95% CI, as measures of the association of HB concentrations with all-cause mortality, adjusted for potential confounding factors. Model 1 was not adjusted for any covariates. Model 2 included adjustments for age (years), sex (male or female), residence (urban or rural), education level (years), marital status (married or unmarried), and economic status (rich, general, or poor). Model 3 included additional adjustments for smoking status (current smoker or non-smoker); drinking status (current drinker or non-drinker); tea drinking status (yes or no); food diversity score (cereals, vegetables, fruits, meat, fish, eggs, milk, beans, and grease, 1 point for each type of food consumed for each individual in 1 week); regular exercise uptake (yes or no); restricted ADL (bathing, getting out of bed, dressing, walking across a room, eating, and continence status, any of six activities not independently completed defined as restricted ADL); self-reported diagnosis of hypertension (yes or no), diabetes (yes or no), stroke (yes or no), or heart disease (yes or no); MMSE score (the measure of cognitive impairment, no cognitive impairment [25 ≤ MMSE score ≤ 30]); depressive symptom score (a five item scale, and a score from 0 to 4 assigned to each response) (24); Model 4, the final model, included additional adjustments for BMI (kg/m²); CRP levels (mg/L); eGFR (ml/min/1.73 m²; calculated based on Scr according to Chronic Kidney Disease Epidemiology Collaboration); WBC (10⁹/l); MCV estimates (fl); platelet count (10⁹/l), and TC levels (mmol/L).

Furthermore, we performed restricted cubic spline curves based on multivariable Cox proportional hazards models using HB concentration as a continuous variable to examine linear or non-linear associations between HB concentrations and all-cause mortality in the fully adjusted model. We conducted stratified analyses to assess potential modification effects of the following factors: age, sex, education level, economic status, smoking, drinking, and tea drinking status, regular exercise uptake, and BMI. Anticipating the complex association between HB concentrations and all-cause mortality, we conducted three types of sensitivity analyses. First, we excluded the participants that died in the first year of the study to minimize the potential effect of reverse causation. Second, we excluded participants with a history of hypertension, diabetes, stroke, and heart disease to reduce spurious increases in the risk of mortality. Third, we excluded participants who had low BMI (<18.5 kg/m²) to eliminate the impact of malnutrition. A two-tailed p-value of <0.05 was considered indicative of a statistically significant finding. Statistical analysis was performed using R software version 3.5.0 (R Center for Statistical Computing, Vienna, Austria).

Of 1,785 participants form longevity regions, the mean age was 86.7 years (SD: 12.2), and 1,002 (56.1%) were women. Overall, the mean HB concentration was 12.2 g/dL (SD: 23.1). With the increasing HB concentrations, the proportion of female participants, restricted ADL, and mean age decreased. Conversely, participants with higher HB concentrations were more likely to be current smokers or drinkers; perform regular exercise; reside in rural environments; have better economic status; and have higher levels of eGFR, MCV, and platelet count (Table 1).

During a median of 5.4 years' follow-up (interquartile range: 4.4–6.1 years), we identified 999 deaths (men: 623, women: 376), accounting for 56.0% of all participants. Table 2 presents data of the association of stratified HB concentrations with all-cause mortality in older adults form longevity regions. Compared with the reference group (13.0–13.9 g/dL), groups with the lowest (<11.0 g/dL) and highest (≥14.0 g/dL) HB concentrations had fully adjusted HRs of 1.37 (95% CI: 1.10–1.70) and 0.76 (95% CI: 0.60–0.97), respectively, for all-cause mortality.

Similar to the results above, the restricted cubic spline curves analysis revealed no non-linear association of HB concentrations with all-cause mortality (p = 0.351 for non-linearity; Figure 2). Participants with HB concentrations lower than 12.3 g/dL had a significantly increased risk of all-cause mortality and those with HB concentrations of higher than 12.3 g/dL had a significantly lower risk of all-cause mortality. However, the HB concentration of 12.0–12.9 g/dL was not associated with all-cause mortality.

We further analyzed the association between HB concentrations and all-cause mortality by sex owing to the different HB reference criteria for men and women. A similar linear association was also observed between HB concentrations and all-cause mortality for both men and women (Figure 3). For men, the risk of all-cause mortality was significantly lower with the highest HB concentration of ≥14.0 g/dL in the fully adjusted model (HR: 0.69, 95% CI: 0.50–0.96); and for women, the risk of all-cause mortality was significantly higher with the lowest HB concentrations of <11.0 g/dL in the fully adjusted model (HR: 1.42, 95% CI: 1.11–1.82) (Table 3). Furthermore, there was no significant interaction effect between HB concentrations and all-cause mortality by age, education time, economic status, smoking status, drinking status, tea drinking status, regular exercise uptake, and BMI (Table 4). The sensitivity analyses revealed that the association between HB concentrations and all-cause mortality was largely similar to the primary results when excluding individuals who died in the first year; those with a history of hypertension, diabetes, stroke, and heart disease; and those who had low BMI (<18.5 kg/m²) based on the fully adjusted model (Table 5).