Bias analysis of confounding factors between diabetes and periodontitis

Research Topic:

Background + Aims

Periodontitis has been linked to various systemic diseases, including diabetes, cardiovascular disease, and rheumatoid arthritis.
The relationship between periodontitis and diabetes is particularly notable due to its bidirectional nature, with each condition potentially influencing the development and progression of the other through shared inflammatory processes. Periodontal treatment has also been shown to improve metabolic control in diabetic patients.
Despite extensive research, the role of unmeasured or residual confounding in these associations remains uncertain.
Observational studies suggest confounders like smoking or genetic factors may explain these links, underscoring the importance of systematic approaches, such as directed acyclic graphs (DAGs), to identify and control for confounders. Quantitative bias analysis (QBA) offers a novel method to simulate and evaluate the impact of unmeasured confounders.
This study uses QBA to assess the potential role of unmeasured confounding in the bidirectional associations between periodontitis

Materials + Methods

The Veterans Affairs (VA) Dental Longitudinal Study (DLS) is a subset of the Normative Aging Study, involving adult men from Boston enrolled in the 1960s.
Participants underwent comprehensive medical and dental examinations approximately every three years, with periodontal data collected from 1981 to 2009.
Periodontitis was assessed using CDC/AAP case definitions, dichotomized as moderate/severe or mild/none, depending on its role as an exposure or outcome.
Diabetes was identified from medical records, interviews, or death certificates.
Covariates included age, education level (as a proxy for socioeconomic status), smoking history (quantified using the Comprehensive Smoking Index), and BMI.
To evaluate the bidirectional associations between periodontitis and diabetes, directed acyclic graphs (DAGs) were constructed using DAGitty.
- One DAG modelled periodontitis as the exposure and diabetes as the outcome, while the other reversed these roles.
- Key covariates identified for adjustment included age, socioeconomic status (education level), smoking history (quantified using the Comprehensive Smoking Index, CSI), and BMI. Due to the homogeneous cohort of male, primarily non-Hispanic white participants, gender and race were not controlled.
Baseline education levels were dichotomised as college graduate or not.
Smoking data, collected via questionnaires, were used to compute the CSI, which accounts for the exponential decay of smoking’s impact on chronic diseases.
BMI was calculated using standardised height and weight measurements
Age, CSI, and BMI were analysed as continuous variables.
Statistical analyses employed crude and adjusted Cox proportional hazards models to examine associations between exposures and outcomes.
Models incorporated time-varying covariates and accounted for competing risks, such as diabetes-related death. Missing data were minimal (<40 cases) and handled via imputation. Participants were censored for extended follow-up gaps, loss to follow-up, or tooth loss affecting eligibility.
The study applied quantitative bias analysis (QBA) to simulate the potential impact of unmeasured confounders on observed associations.

Results

Periodontitis and incident diabetes:
- Participants with moderate or severe periodontitis were older, less educated, more likely to smoke, and more obese compared to those with mild or no periodontitis.
- Over a median follow-up of 14 years, the unadjusted hazard ratio (HR) for developing diabetes was 1.46 (95% CI: 0.79–2.71). After adjusting for baseline and time-varying covariates, the HRs were attenuated to 1.21 (95% CI: 0.64–2.30) and 1.33 (95% CI: 0.71–2.52), respectively, indicating potential confounding effects.
- Quantitative bias analysis (QBA) revealed that a simulated confounder strongly associated with both periodontitis and diabetes could nullify or amplify the observed association, suggesting sensitivity to unmeasured confounders.
Diabetes and incident severe periodontitis:
- Those with diabetes were older, more likely to smoke, and more obese compared to non-diabetics.
- Over a median follow-up of 8.5 years, the unadjusted HR for developing severe periodontitis was 1.45 (95% CI: 0.86–2.45). Adjustments for baseline and time-varying covariates slightly reduced the HRs to 1.35 (95% CI: 0.79–2.32) and 1.32 (95% CI: 0.77–2.26), respectively.
- QBA showed that confounders strongly associated with diabetes and periodontitis could attenuate or amplify the observed association but could not entirely explain it.
Both associations showed non-significant trends toward increased risk, with adjustments highlighting the role of confounding.
QBA demonstrated that unmeasured confounders could significantly influence results.

Limitations

The QBA simulated only a single time-fixed unmeasured confounder, which may not fully capture the complexities of time-varying or multiple confounders, leaving potential residual confounding. Misclassification of person-time for the simulated confounder could have occurred, as simulations were based on exposure status at the time of censoring or failure.
The competing event of death and prior tooth loss could introduce bias in estimating periodontitis status and confounder prevalence, though their effects appeared minimal.
The study’s reliance on baseline measures to determine diabetes and periodontitis may have resulted in misclassification or selection biases, such as undiagnosed diabetes at baseline or artificial censoring. These factors could affect the accuracy of hazard ratio estimates and QBA results.
The cohort’s restriction to males who were predominantly White and Non-Hispanic limits the generalisability of findings across genders and diverse populations.
Despite minimising bias by restricting eligibility to participants with at least eight teeth, tooth loss prior to study entry remains a potential confounder. Additional analyses controlling for baseline tooth count showed no significant impact on hazard ratios.
Random error and lack of precision in underlying effect estimates may influence the robustness of the QBA findings.

Conclusion

The findings suggest that the bidirectional associations between periodontitis and diabetes may not reflect true causality. Instead, some observed links might arise from an unmeasured common cause, such as a shared inflammatory pathway independently associated with both conditions.
QBA, as a straightforward tool, enhances the rigor of studies on periodontitis-systemic disease links. Given the limited understanding of these associations and their underlying mechanisms, addressing uncertainties caused by unknown confounding is essential. Quantifying such bias through QBA offers valuable insights for advancing research in this field.

Read the full article Back to Research