Factors that influence data quality in caries experience detection: a multilevel modeling approach

Area of Science:

• Dentistry
• Biostatistics
• Epidemiology

Background:

• Caries experience detection is prone to misclassification, impacting diagnostic accuracy.
• Calibration exercises assess and improve dental raters' scoring behavior using contingency tables.
• Current dental studies often fail to express the uncertainty in sensitivity and specificity estimates.

Purpose of the Study:

• To apply a Bayesian logistic multilevel model for estimating sensitivity and specificity in caries detection.
• To identify factors influencing the true scoring of caries experience.
• To account for the hierarchical data structure in caries assessment.

Main Methods:

• Utilized a Bayesian logistic multilevel model to estimate sensitivity and specificity.
• Analyzed data from calibration exercises involving dental raters and a benchmark scorer.
• Incorporated hierarchical data structure (surfaces nested in teeth within the mouth) into the model.

Main Results:

• The Bayesian multilevel model effectively estimates sensitivity and specificity while accounting for data hierarchy and uncertainty.
• Dentition type (e.g., primary vs. permanent) was found to significantly affect caries detection quality.
• Tooth type and surface type also influence the accuracy of caries experience scoring.

Conclusions:

• Bayesian multilevel modeling provides a robust framework for analyzing caries detection data, addressing uncertainty and hierarchical structures.
• Factors such as dentition and tooth/surface type are critical determinants of accurate caries diagnosis.
• Improved methods for assessing rater performance are essential for reliable epidemiological studies of dental caries.