Periodontitis misclassification bias and its impact on diabetes and CVD associations

Research Topic:

Background + Aims

Partial-mouth protocols (PMPs) are commonly used in population health studies to estimate periodontal disease prevalence, offering a cost-effective and time-efficient alternative to full-mouth protocols (FMPs). However, PMPs are known to underestimate periodontal disease prevalence, which may lead to biases in studies examining associations between periodontitis and systemic conditions like diabetes and cardiovascular disease (CVD).
Despite efforts to address these limitations through modified PMPs and mathematical corrections, the potential for misclassification bias has hindered their adoption in associational research. This bias arises due to systematic misclassification, which varies based on the case definition, disease severity, and whether periodontitis is the exposure or outcome.
While previous studies have focused on PMP biases in estimating prevalence, fewer have explored their impact on disease association studies.
This study aims to quantify the misclassification bias introduced by PMPs and evaluate how it affects the estimated associations between periodontitis, diabetes, and CVD.

Materials + Methods

This study analysed data from the National Health and Nutrition Examination Survey (NHANES) (2009–2014), a cross-sectional U.S. surveillance programme.
Participants were included if they were:
- Aged 30–79 years old
- Had ≥2 permanent teeth
- Had full-mouth and complete periodontal examination data
Periodontal status was classified using full-mouth protocols (FMPs) using clinical attachment loss (CAL) and probing pocket depth (PPD).
3 partial-mouth protocols (PMPs) were then applied to the data to allow for comparisons:
- Random Half-Mouth (RHM): measurement of either the maxillary right/mandibular left sides or the maxillary left/mandibular right sides was randomly assigned.
- Community Periodontal Index of Treatment Needs (CPITN): fixed set of ten teeth, including maxillary first and second molars, maxillary right central incisor, mandibular first and second molars, and mandibular left central incisor.
- Ramfjörd: included the maxillary right first molar, maxillary left central incisor, maxillary left first premolar, mandibular left first molar, mandibular right central incisor and mandibular right first premolar.
Periodontal status was classified using:
- Categorical definitions – none/mild, moderate, severe
- Continuous metrics – mean PD and CAL
Self-reported diabetes and cardiovascular disease (CVD) outcomes were dichotomised, and participants with missing data were excluded.
Covariates included age, gender, smoking, obesity, race/ethnicity, and socio-economic status, with additional adjustment for diabetes in CVD models.
Logistic regression models estimated associations between periodontitis and diabetes or CVD, incorporating NHANES sampling weights.
Bias was assessed using percent and absolute relative bias on the natural log scale, accounting for sampling variability in RHM through 10,000 iterations.
Stratified analyses examined the impact of tooth loss (≤15 vs. >15 teeth) on bias estimates.

Results

The study analysed data from 9,575 participants:
- Male: 49.2%
- Mean age of approximately 50 years
The prevalence of the following were:
- Severe periodontitis – 7.8%
- Moderate periodontitis – 29%
- Self-reported diabetes – 9.4%
- Self-reported cardiovascular disease (CVD) -5.5%
Severe periodontitis was associated with increased odds of diabetes (OR 1.04) and CVD (OR 1.22) when using full-mouth protocols (FMPs).
Moderate periodontitis showed stronger associations (OR 1.26 for diabetes; OR 1.30 for CVD).
Continuous measures also demonstrated a significant association between CAL and CVD with a 1 mm increase in CAL having an OR 1.10. Other continuous measures did not have any statistical significance on diabetes or CVD risk.
Periodontal misclassification
- The impact of PMPs on periodontitis exposure misclassification affected the direction and magnitude of bias across the protocols and outcomes studied.
- Severe periodontitis showed inconsistent bias in its association with diabetes under different PMPs:
  - The RHM and Ramfjörd protocols biased results towards the null, reversing the OR <1
  - CPITN biased results away from the null (OR = 1.14).
  - The extent and type of bias introduced by PMPs on diabetes and CVD:
    - Diabetes:
      - Relative bias of 177.7% – 204.1% demonstrates the difference between the observed estimate (from the PMP) and the true estimate (from the FMP), which is extremely high. This means the PMPR substantially over- or underestimates the association compared to the FMP.
      - Absolute bias of 0.07 – 0.09 quantifies the direct difference between the observed and true estimate, irrespective of the percentage. This range indicates significant distortion of the results.
    - CVD:
      - Relative bias ≤4.1% which is minimal and means the PMP-based estimates for severe periodontitis and CVD are close to the true estimates when using the FMP.
      - Absolute bias ≤0.01 is very small, indicating that the PMP introduces negligible distortion in these estimates.
    - Moderate periodontitis misclassification consistently biased diabetes associations towards the null, with relative bias ranging from 28.4% to 39.5%. For CVD, bias varied more, ranging from 8.9% to 46.7%, with the direction depending on the protocol (RHM and Ramfjörd towards the null; CPITN away from it). Absolute bias remained small (0.02–0.12).
    - Continuous measures like mean CAL minimised bias across outcomes but showed outcome-dependent directional trends. Diabetes results exhibited bias away from the null, while CVD results showed bias towards the null. Bias ranged from none to 48.6% (relative) and up to 0.02 (absolute). Mean probing depth (PD) introduced more significant bias for diabetes compared to CVD.
    - Stratifying by tooth count (≤15 vs. >15) increased bias, especially for those with fewer teeth. Patterns of bias by protocol and exposure severity changed post-stratification, with no PMP consistently minimizing bias. RHM, initially the least biased protocol, lost this consistency after stratification, and continuous measures showed amplified bias with inconsistent direction. Overall, misclassification effects were protocol-, outcome-, and exposure-dependent, underscoring the variability introduced by PMP use.
    - Continuous measures reduced bias compared to categorical definitions.

Limitations

The cross-sectional design precludes establishing temporality between periodontitis and systemic diseases, particularly for diabetes, where a bidirectional relationship is hypothesised.
Reliance on self-reported outcomes and covariates in NHANES introduced potential misclassification, such as the inability to differentiate between type I and type II diabetes. While outcome misclassification likely had minimal impact, residual confounding from self-reported covariates remains unpredictable.
Bias due to PMPs was influenced by periodontitis severity, the target population’s disease extent, and the case definition applied. PMP-related bias was outcome-dependent, possibly reflecting the ability of definitions to capture relevant phenotypes for diabetes versus cardiovascular disease (CVD), rather than protocol limitations. The study also revealed dependencies on the number of teeth available for evaluation.
Multiple classification errors, including PMP use and misclassification of causal exposures, may confound associations between periodontitis and systemic health.

Conclusion

This study underscores the impact of partial-mouth protocols (PMPs) on misclassification bias in perio-systemic disease associations.
Bias was outcome- and protocol-dependent, with minimal effects for cardiovascular disease but substantial for diabetes.
Accurate interpretation requires understanding PMP limitations, case definitions, and phenotypic differences, emphasising the need for robust methodologies in future research.

Read the full article Back to Research