Matching the Statistical Model to the Research Question for Dental Caries Indices with Many Zero Counts

Area of Science:

• Biostatistics
• Dental Public Health
• Epidemiology

Background:

• Traditional zero-inflated and hurdle models are commonly used for analyzing count data with excess zeros, such as dental caries indices.
• Marginalized zero-inflated count regression models have emerged as a novel alternative, directly estimating overall exposure or treatment effects.

Purpose of the Study:

• To discuss the interpretation and selection of marginalized zero-inflated count regression models for dental caries research.
• To compare marginalized models with traditional zero-inflated and hurdle models using real-world dental caries data.

Main Methods:

• Analysis of two datasets: fictional dmft counts and DMFS counts from a randomized clinical trial on toothpaste efficacy.
• Comparison of negative binomial hurdle, zero-inflated negative binomial, and marginalized zero-inflated negative binomial models.

Main Results:

• Treatment effect estimates varied depending on the specific incidence rate ratio (IRR) estimated by each model.
• IRR estimates from the randomized clinical trial were comparable across models, despite differing interpretations.
• High frequencies of zero counts in dmft and DMFS indices, reflecting a decline in caries experience, necessitate careful model selection.

Conclusions:

• The choice of statistical model class should align with the specific research question in dental caries studies.
• Marginalized zero-inflated models are recommended for directly assessing exposure effects on the marginal mean dental caries count, particularly when dealing with a high prevalence of zero counts.