Distance metrics optimized for clustering temporal dietary patterning among U.S. adults

Area of Science:

• Nutrition Science
• Data Science
• Public Health

Background:

• Understanding temporal dietary patterns (TDP) is crucial for assessing diet quality.
• Previous methods for TDP analysis have limitations in capturing time and proportion of energy intake.
• Modified dynamic time warping (MDTW) was developed to address these limitations.

Purpose of the Study:

• To explore dynamic time warping (DTW) distance metrics (UDTW, CDTW, MDTW) with spectral clustering for optimizing TDP.
• To determine the association between optimized TDP and dietary quality using the 2005 Healthy Eating Index (HEI-2005).
• To compare the effectiveness of different DTW metrics in identifying TDP related to dietary quality in U.S. adults.

Main Methods:

• Optimized proportional energy intake by time of day metrics to create TDP using MDTW, UDTW, and CDTW.
• Applied spectral clustering to group participants based on generated TDP.
• Utilized multiple linear regression to assess the association between TDP clusters and HEI-2005 scores, controlling for confounders.

Main Results:

• Four distinct TDP clusters were generated for each distance metric.
• MDTW exhibited TDP clusters with the strongest association to HEI-2005 scores compared to UDTW and CDTW.
• Visualization of MDTW-generated TDP clusters supported their stronger association with dietary quality.

Conclusions:

• MDTW, when combined with spectral clustering, is a valuable tool for dimension reduction in dietary data.
• This approach effectively uncovers temporal patterns in dietary intake.
• The findings highlight MDTW's utility in identifying TDP associated with improved dietary quality.