Simulation study of misclassification bias in association studies employing partial-mouth protocols

Brenda Heaton, Raul I. Garcia, Thomas Dietrich

Research output: Contribution to journalArticlepeer-review

5 Citations (Scopus)
194 Downloads (Pure)

Abstract

Aim: To simulate the exposure misclassification bias potential in studies of perio-systemic disease associations due to the use of PMR protocols.

Methods: Using data from 640 participants in the Dental Longitudinal Study, we evaluated distributions of clinical periodontitis parameters to simulate hypothetical outcome probabilities using bootstrap sampling. Logistic regression models were fit using the hypothetical outcome as the dependent variable. Models were run for exposure classifications based on FMR and PMR protocols over 10,000 repetitions.

Results: The impact of periodontitis exposure misclassification was dependent on periodontitis severity. Percent relative bias for simulated ORs of size 1.5, 2 and 4 ranged from 0 to 30% for the effect of severe periodontitis. The magnitude and direction of the bias was dependent on the underlying distribution of the clinical parameters used in the simulation and the size of the association being estimated. Simulated effects of moderate periodontitis were consistently biased toward the null.


Conclusion: Exposure misclassification bias occurring through the use of PMR protocols may be dependent on the sensitivity of the classification system applied. Using the CDC-AAP case definition, bias in the estimated effects of severe disease were small, on average. Whereas, effects of moderate disease were underestimated to a larger degree.
Original languageEnglish
Pages (from-to)1034-1044
Number of pages11
JournalJournal of Clinical Periodontology
Volume45
Issue number9
Early online date4 Jul 2018
DOIs
Publication statusPublished - Sept 2018

Keywords

  • bias
  • misclassification
  • periodontal disease
  • periodontitis
  • simulation

Fingerprint

Dive into the research topics of 'Simulation study of misclassification bias in association studies employing partial-mouth protocols'. Together they form a unique fingerprint.

Cite this