Study design and participants
CLHLS is an ongoing longitudinal interview survey of over 40,000 old adults in 22 of 31 Chinese provinces. The included participants represent 985 million persons, about 85% of the national population [22]. The investigation began in 1998 and follow-up surveys were undertaken in 2000, 2002, 2005, 2008, 2011, and 2014, with approximately a 90% response rate in each wave. To maintain a large enough sample size, the CLHLS replaced the deceased respondents with new participants in the follow-up waves. To avoid the problem of small sub-sample sizes at the more advanced ages, the CLHLS interviewed nearly all centenarians [22]. Details of the CLHLS have been described elsewhere [22]. The study was approved by the Ethics Committee of Peking University and Duke University. Informed consents were obtained from all participants during the face-to-face interview.
Our inclusion criteria for participants were: aged 65 years or above; had data on the number of natural teeth and/or denture use; and data about the death time was available for at least one follow-up survey. We included all participants from the 1998 survey as all participants were newly recruited. For the follow-up surveys, the participants included the survivals from the previous waves and the new recruits. Only the new recruits needed to be included as the survivals from the previous waves had already been included. This strategy enabled us to analyze based on the largest sample size. In the end, we included 36,283 participants from the CLHLS, with 36,153 participants for the analysis of number of natural teeth and 36,230 for denture use (Figure 1 in the Additional file). There was a small difference in the participant numbers for the two analyses because some participants only reported one of the two exposure data.
Assessment of the exposures
Self-reported number of natural teeth and the use of dentures were collected with the following questions: 1) How many natural teeth (teeth that are naturally grown) do you still have? 2) Do you have false teeth? (False teeth referred to any type of non-natural teeth, including partial or complete, removable or implant-retained fixed dentures). Data were repeatedly collected in each wave. We grouped the number of remaining teeth into four categories (0, 1–9, 10–19, and ≥ 20). We evaluated the combined effects by grouping the participants into 8 categories according to teeth number and denture use. Though the education level of included participants were low and one third of the subjects were cognitively impaired, the validity of self-reported number of natural teeth and denture use may not be influenced as the interviewers could help older adults to confirm their responses to these questions.
Assessment of deaths
We ascertained the survival status through a face-to-face interview with a close family member for those interviewees who had died before the next wave [22]. The survival time was defined as the period from the date of the baseline visit to the date of death. The data for the participants who survived until the 2014 survey were censored at the time of the 2014 survey, and those who were lost to follow-up were censored at the time of the last survey.
Assessment of covariates
We selected covariates that may confound the relationship based on a review of literature [23, 24]. The selected covariates should have an impact on the overall mortality and been collected in the CLHLS. For example, we included fresh fruit intake and vegetable consumption because they may influenced by teeth healthy and have an impact on mortality [25]. Because the sample size of this study is large and statistical power is strong, we included as many covariates as possible to minimize the potential confounding effect. We obtained covariate information from the structured questionnaire [26]. The covariates for our analyses included sociodemographic characteristics, routine physical checkup (weigh, height, and blood pressure), lifestyle behaviors (smoking, alcohol drinking, physical activity, fresh fruit intake, and vegetable consumption), self-reported medical history, activities of daily living (ADL), cognitive function (evaluated by the Mini-Mental State Examination, MMSE) [27], and depressive symptoms.
Statistical analysis
We evaluated all-cause mortality according to the number of natural teeth and denture use with Kaplan-Meier survival plots. To explore the shape of the relationship, we used addictive Cox regression, taking number of natural teeth as a smoothed term. Penalized splines were used for smoothing. The degrees of freedom were determined according to Akaike Information Criterion and residual deviance. For better interpretation of effect, we also evaluated the hazard ratios (HRs) and 95% confidence intervals (CIs) with time-dependent Cox regression. We divided the subject’s time of observation into several intervals based on the time of follow-up surveys and date of death. The exposure values at the beginning of each interval were used for this interval.
Multivariable models were applied to adjust for confounding factors. We adjusted for age at baseline, sex, and residence (urban or rural, defined according to the official registration record for Chinese citizen) in the basic analysis model. Additionally, we adjusted for education, living arrangement, sufficient income for daily needs, smoking, alcohol drinking, frequent vegetable consumption, frequent fruit consumption, frequent physical activity, ADL, cognitive impairment (defined as MMSE < 24), body mass index, hypertension, self-reported diabetes mellitus, self-reported heart disease, self-reported cerebrovascular disease, self-reported respiratory disease in the fully-adjusted model. We coded the participants with missing covariate data to the reference group or median group when the missing rate was low (< 5%). When the rate of missing data was ≥5%, a separate missing response category was created. Based on the HRs and mortality rate, we calculated the age- and sex- specific number needed to treat (NNT) [28], which is conceptually easier to understand.
We conducted subgroup analyses by age, sex, residence, years of education, BMI, smoking status, and drinking status, in order to investigate potential effect-modifying effect. We also evaluated the association of teeth number with all-cause mortality by denture use and vice versa.
We conducted a number of sensitivity analyses: 1) additionally adjusting for depressive symptom, current marital status, time of recruitment; 2) using the baseline number of natural teeth and denture use as covariates; 3) excluding patients with a history of diabetes mellitus, heart disease, cerebrovascular disease, or respiratory diseases; 4) excluding participants with an observation time of < 3 years, > 12 years, or without tooth loss; 5) additionally adjusting for living location; and 6) considering the participants with unknown survival status censored at the median of follow-up (3 years) as previous study [29]; 6) excluding the participants who reported more number of natural teeth than the teeth number in the previous wave of survey. Analyses were completed using Stata version 12.0 (StataCorp LP, College Station, TX, USA) and R software version 3.4.1 (R Development Core Team, 2017).