IADR Abstract Archives
The Impact of Unmeasured Confounding on Periodontitis’ Associations with Diabetes
Objectives: To investigate the potential effect of unmeasured confounding on the hypothesized bidirectional associations between periodontitis and diabetes.
Methods: Data from the VA Dental Longitudinal Study were used to ascertain periodontitis status using the CDC/AAP case definition. Diabetic status was based on self-report. We estimated the association of periodontitis with diabetes incidence (periodontitis-diabetes) among participants who were free of diabetes at baseline, had ≥2 periodontal exams over the observation period (1981-2009), and ≥8 teeth at baseline (N=672). We similarly estimated the association of diabetes with onset of severe periodontitis (diabetes-periodontitis) among those who were free of severe periodontitis at baseline (N=521). Adjusted Cox proportional hazards models were used to calculate corrected hazard ratios (HR) after simulating confounders. Simulated confounders were repeated over 10,000 bootstrapped samples to obtain the 2.5th-97.5th percentile interval of corrected hazard ratios.
Results: The uncorrected periodontitis-diabetes HR was 1.19 (95% CI:0.63-2.26). A simulated confounder that had moderate or strong positive associations with diabetes (OR=1.3-2), and was positively associated with periodontitis, weakened the association between severe/moderate periodontitis and diabetes (HR between 1 and 1.1) or totally explained it (HR=1). The uncorrected diabetes-severe periodontitis HR was 1.34 (95% CI:0.78-2.30). The lower bound of the simulation interval for the corrected HR, never reached the null value (HRs ≥1.03), reflecting control of unmeasured confounding in the most extreme circumstances e.g., strongly associated with diabetes (OR=1.8-2.0) and highly prevalent among those with periodontitis.
Conclusions: When diabetes was the exposure, the presence of one source of unmeasured confounding affected the association with severe periodontitis less than when diabetes was the outcome. No matter how strong the confounder, it never totally explained the association of diabetes on development of periodontitis, increasing the plausibility of a causal diabetes-periodontitis association. Whereas, when diabetes was the outcome, weak unmeasured confounding could provide a plausible explanation for the observed effect of periodontitis.
Methods: Data from the VA Dental Longitudinal Study were used to ascertain periodontitis status using the CDC/AAP case definition. Diabetic status was based on self-report. We estimated the association of periodontitis with diabetes incidence (periodontitis-diabetes) among participants who were free of diabetes at baseline, had ≥2 periodontal exams over the observation period (1981-2009), and ≥8 teeth at baseline (N=672). We similarly estimated the association of diabetes with onset of severe periodontitis (diabetes-periodontitis) among those who were free of severe periodontitis at baseline (N=521). Adjusted Cox proportional hazards models were used to calculate corrected hazard ratios (HR) after simulating confounders. Simulated confounders were repeated over 10,000 bootstrapped samples to obtain the 2.5th-97.5th percentile interval of corrected hazard ratios.
Results: The uncorrected periodontitis-diabetes HR was 1.19 (95% CI:0.63-2.26). A simulated confounder that had moderate or strong positive associations with diabetes (OR=1.3-2), and was positively associated with periodontitis, weakened the association between severe/moderate periodontitis and diabetes (HR between 1 and 1.1) or totally explained it (HR=1). The uncorrected diabetes-severe periodontitis HR was 1.34 (95% CI:0.78-2.30). The lower bound of the simulation interval for the corrected HR, never reached the null value (HRs ≥1.03), reflecting control of unmeasured confounding in the most extreme circumstances e.g., strongly associated with diabetes (OR=1.8-2.0) and highly prevalent among those with periodontitis.
Conclusions: When diabetes was the exposure, the presence of one source of unmeasured confounding affected the association with severe periodontitis less than when diabetes was the outcome. No matter how strong the confounder, it never totally explained the association of diabetes on development of periodontitis, increasing the plausibility of a causal diabetes-periodontitis association. Whereas, when diabetes was the outcome, weak unmeasured confounding could provide a plausible explanation for the observed effect of periodontitis.
