A comparison of heuristic and model-based clustering methods for dietary pattern analysis.

Area of Science:

• Nutritional epidemiology
• Statistical modeling
• Computational biology

Background:

• Dietary pattern analysis is crucial for understanding health-disease relationships.
• Traditional methods like k-means and Ward's struggle with complex dietary data structures.
• Gaussian mixture models (GMM) present a more flexible clustering alternative.

Purpose of the Study:

• To compare the performance of Gaussian mixture models (GMM) against k-means and Ward's method for dietary pattern identification.
• To determine the most appropriate clustering approach for analyzing dietary data.
• To evaluate clustering methods using both simulated and real-world dietary datasets.

Main Methods:

• Application of GMM, k-means, and Ward's method to simulated datasets with known cluster structures.
• Evaluation of clustering performance based on accuracy in identifying true cluster memberships.
• Application of the same methods to a real-world dataset from the IDEFICS study involving 1791 children.

Main Results:

• GMM demonstrated superior performance in the simulation study, outperforming other methods in 72-100% of cases.
• K-means showed better performance than Ward's method in 64-100% of simulation scenarios.
• All methods identified three comparable dietary patterns in the IDEFICS study: 'non-processed', 'balanced', and 'junk food'.

Conclusions:

• GMM is recommended for dietary pattern analysis due to its flexibility with cluster shapes and orientations.
• K-means is a practical and effective alternative, yielding similar results to GMM on real dietary data.
• The choice of clustering method impacts the precision of dietary pattern identification.