Association Between Periodontitis and All-cause and Cancer Mortality

Retrospective Elderly Community Cohort Study

Ping-Chen Chung; Ta-Chien Chan


BMC Oral Health. 2020;20(168) 

In This Article


Study Design and Population

This was a retrospective cohort study from January 1, 2005 to December 31, 2012, which used a dataset of health examinations for the elderly with age equal to or above 65 years old, performed by Taipei-contracted hospitals and supported by the Department of Health, Taipei City Government in Taiwan. Participation in the annual health examinations was voluntary for senior citizens. The study population received an interview, physician consultation and clinical examination from January 1, 2005 to December 31, 2008.

Those aged less than 65 years old (n = 853), or with a misrecorded examination date (n = 9) or missing data on periodontal status (n = 5257) at the first visit were excluded. Finally, we enrolled 82,548 study participants for further analyses. The total visits numbered 262,035 as of the end of the study after excluding 26,461 visits with missing data regarding periodontal status (n = 26,455) or misrecorded examination dates (n = 6) (Figure 1).

Figure 1.

The flow chart of the study sample enrollment and follow-up

Assessment and Definition of Periodontitis

In the oral examination, participants with periodontal status reported as "inapplicable" or "refused" were excluded. If participants' periodontal status as diagnosed by dentists showed "no obvious abnormalities" then these participants were classified as having healthy periodontium, while participants with "abnormal periodontal status" diagnosis and periodontal tissues described as "tooth mobility" or "periodontitis" by dentists were classified as having periodontitis.

Outcome Definition

The primary endpoint was the date of death, especially death from cancer, or the end of the follow-up period (December 31, 2012). The cause of death was recorded according to the International Classification of Diseases, Ninth Revision (ICD-9: 001–998) or Tenth Revision (ICD-10: A00-Z99).[25]

Measurement of Exposure and Potential Confounders

The baseline interview collected age, sex, education level (illiterate, 1–6 years of schooling, 7–14 years of schooling, or above 14 years of schooling), marital status (married and living together, vs. others), self-reported smoking status in the past 6 months (smoked every day, smoked some days or socially, or did not smoke), and self-reported intake of two fruits and three dishes of vegetables daily (yes, no). If the participant had a history of diabetes or took long-term medication for controlling diabetes, or the fasting blood glucose report revealed abnormality, then the participant was defined as diabetic. In each oral examination, periodontal status was recorded by dentists.

Statistical Analysis

The proportions of participants with different periodontal status at the baseline were calculated separately by demographic characteristics and health behaviors. Comparisons of baseline characteristics between subgroups according to the periodontal status were made using logistic regression in which the first category in each variable was regarded as the reference group. Kaplan-Meier curves with the log-rank test were employed to demonstrate the differences in survival curves in subgroups of different periodontal status at the baseline. At each time point, Kaplan-Meier survival data included the numbers at risk.

A Cox proportional hazards model[26] and Cox frailty model[27,28] were used for calculating the hazard ratios of all-cause mortality and all-cause cancer mortality under different periodontal status. A Cox proportional hazards model as a semi-parametric model is a common method for study of time-to-event data.[26] A Cox frailty model is a time-dependent model considering random effects of time. This approach can be used for repeated events for the same individual.[29] In our elderly cohort, a participant might be examined in several years, and the smoking and periodontal status might be different each time. The Cox frailty model is a suitable method of analysis. The coefficients estimated from the frailty models might differ from those of the general Cox model if there is a meaningful contribution of the random term. After deleting participants who had one or more missing covariates regarding education level (n = 12,592), marital status (n = 1347), and smoking status (n = 335) in the baseline data, the Cox proportional hazards model and the Cox frailty model estimated the hazard ratio for all-cause and all-cancer mortality and included age, sex, education level, marital status, smoking status and periodontal status. Due to the low number of each specific cancer to test the association, besides periodontal status, the Cox frailty models of deaths from cancer were only adjusted for age and sex. Hazard ratios and 95% confidence intervals of all-cause, all-cancer and specific cancer mortalities in subgroups are summarized in Table S1. All analyses were two-sided with alpha set at 0.05. We conducted all statistical analyses by using SAS (version 9.4) and RStudio (Version 1.0.153) with packages of survival,[30] and ggplot2.[31]