Periodontitis Misclassification from Partial-Mouth Measurements: Effect on Associations with Diabetes

Objectives: To quantify misclassification bias in associations between periodontitis and diabetes due to partial-mouth measurements.
Methods: Data from 9,586 adults aged 30-79 years with ≥2 teeth in NHANES 2009-2014 were used. Periodontitis was defined using the CDC/AAP 2007 categorical definition and mean clinical attachment loss (CAL) as a continuous measure. Estimates of periodontitis were derived from the full-mouth protocol and three partial-mouth protocols (PMPs), including Ramfjörd teeth, Community Periodontal Index for Treatment Needs teeth, and random half-mouth teeth. Effects of exposure and outcome misclassification of periodontitis were evaluated in relation to diabetes. Diabetes was ascertained by self-report. Adjusted multinomial, binary logistic, and linear regression models were used to estimate diabetes-periodontitis and periodontitis-diabetes associations. Percent relative bias (%RB) was calculated by comparing odds ratios/beta estimates obtained from PMPs to the full-mouth. To evaluate the role of dentition, analyses were stratified by number of teeth present (2-≤15, >15) and restricted by number of teeth available for observation (≥8, ≥10, ≥20, and 28). Analyses accounted for the complex survey design.
Results: When severe periodontitis was evaluated categorically, %RB was minimal, ranging from 3.0% to 7.0%, regardless of whether periodontitis was an exposure or an outcome. The extent of exposure and outcome misclassification of moderate periodontitis ranged from 8.5% to 14.4%, and was consistently biased toward the null. Stratified analyses showed increased %RB (range:14.3%-55.4%) for individuals with ≤15 teeth and greater variation across PMPs, whereas among those with >15 teeth, %RB did not exceed 14.3%. Bias due to exposure misclassification when using mean CAL was negligible, regardless of number of teeth or PMP.
Conclusions: Extent and direction of bias from PMPs may depend on the definition of periodontitis (categorical vs continuous), number of teeth present and PMP type. True effects of outcome misclassification may not have been observable due to small effect sizes.