Statistical Models for High-Risk Intestinal Metaplasia with DNA Methylation Profiling

Area of Science:

• Biostatistics
• Genomics
• Oncology

Background:

• Intestinal metaplasia (IM) is a precancerous condition requiring accurate risk prediction.
• Traditional logistic models have limitations in analyzing complex categorical data.
• DNA methylation data offers insights into cellular processes relevant to IM risk.

Purpose of the Study:

• To evaluate newly developed multinomial mixed-link models for IM risk prediction using DNA methylation data.
• To compare the performance of mixed-link models against traditional logistic models.
• To identify key predictors, including total stem cell divisions (TNSC) and gastric atrophy status, for IM risk.

Main Methods:

• Application of multinomial mixed-link models to DNA methylation data from an IM study.
• Utilizing ten-fold cross-validation to assess model performance via cross-entropy loss.
• Comparing selected mixed-link models (Models 1, 2, and 3) with traditional logistic models.

Main Results:

• The selected multinomial mixed-link model (Model 1) using TNSC significantly outperformed traditional logistic models (p < 10-4).
• TNSC was a highly significant predictor of IM risk (p < 10-6).
• Models incorporating gastric atrophy status (Models 2 and 3) further improved IM risk prediction accuracy compared to Model 1.

Conclusions:

• Multinomial mixed-link models provide a superior framework for IM risk prediction using DNA methylation data.
• Total stem cell divisions (TNSC) derived from DNA methylation is a critical biomarker for IM risk.
• Gastric atrophy status is an informative covariate that enhances the predictive power of IM risk models.