Robust Clustering with Subpopulation-specific Deviations.

Area of Science:

• Epidemiology
• Nutritional Science
• Statistical Modeling

Background:

• Maternal diet is crucial for pregnancy outcomes and birth defect prevention.
• Dietary pattern analysis is complex due to diet's multi-dimensional nature.
• Traditional clustering methods struggle with large, diverse populations, leading to less interpretable results.

Purpose of the Study:

• To develop and apply a novel statistical method for characterizing maternal dietary patterns.
• To address the challenges of dimension reduction in large, heterogeneous populations for dietary pattern analysis.
• To identify pre-pregnancy dietary patterns within the National Birth Defects Prevention Study (NBDPS) while considering regional variations.

Main Methods:

• Proposed Robust Profile Clustering, a two-level clustering approach (global and local).
• Utilized an overfitted finite mixture model for global clustering.
• Employed a beta-Bernoulli process to account for regional subpopulation differences.
• Applied the method to analyze dietary data from the National Birth Defects Prevention Study (NBDPS).

Main Results:

• Successfully derived pre-pregnancy dietary patterns for women in the NBDPS.
• The Robust Profile Clustering method effectively handled data complexities and regional variability.
• Demonstrated improved interpretability of dietary patterns compared to traditional methods.

Conclusions:

• Robust Profile Clustering is a viable method for analyzing complex dietary data in large epidemiological studies.
• Accounting for regional variations enhances the understanding of maternal dietary patterns and their potential role in birth defects.
• This approach offers a more nuanced characterization of diet for public health research.