Identifying temporal eating patterns: a comparison of latent class analysis and dynamic time warping-based cluster...

Area of Science:

• Nutritional Science
• Data Science
• Public Health

Background:

• Temporal eating patterns (TEPs) influence diet quality and obesity, but inconsistent findings arise from varied analytical methods.
• Direct comparisons of these analytical approaches for TEPs are scarce.

Purpose of the Study:

• To compare latent class analysis (LCA) and modified dynamic time warping (MDTW)-based clustering for deriving TEPs.
• To examine the associations of TEPs derived by these methods with diet quality and obesity.

Main Methods:

• Cross-sectional study of 672 Australian adults (18-65 years) using 1-7 day food diaries via the "FoodNow" app.
• LCA utilized hourly eating occasion (EO) presence/absence; MDTW-based clustering used hourly energy intake (EI).
• Methods compared via pattern visualization, membership overlap, kappa statistics, R², and AUC.

Main Results:

• Both LCA and MDTW identified three distinct TEPs: "conventional", "later eating", and "earlier, spaced eating".
• Membership overlap was 56.2%-73.1% (κ = 0.38), indicating fair agreement.
• Conventional TEPs showed higher diet quality; neither method significantly associated TEPs with BMI. LCA explained slightly more diet quality variance (6% vs 4%).

Conclusions:

• LCA and MDTW-based clustering yield comparable yet non-interchangeable TEPs.
• LCA appears more suitable for diet quality research.
• MDTW-based clustering may be advantageous for analyzing multidimensional dietary and health data.