A comparison of children's diet and movement behaviour patterns derived from three unsupervised multivariate methods

Area of Science:

• Behavioral science
• Multivariate statistics
• Public health research

Background:

• Behavioral patterns are typically identified using unsupervised multivariate methods like principal component analysis (PCA), latent profile analysis (LPA), and cluster analysis (CA).
• The comparability of patterns derived from these different statistical methods has not been previously investigated.
• This lack of investigation creates uncertainty regarding whether patterns identified across studies using distinct methodologies are directly comparable.

Purpose of the Study:

• To compare behavioral patterns derived from diet, physical activity, sedentary behavior, and sleep domains.
• To assess the congruence of patterns identified using PCA, LPA, and CA within a single dataset.
• To determine if different statistical approaches yield comparable behavioral typologies.

Main Methods:

• Utilized parent-report and accelerometry data from the HAPPY cohort study (n=432, child age 6-8y).
• Derived behavioral patterns using PCA, LPA, and CA on standardized variables for diet, physical activity, sedentary behavior, and sleep.
• Applied common criteria for pattern retention for each method.

Main Results:

• PCA yielded four patterns, while LPA and CA each produced three patterns.
• Despite variations in pattern number and characterization, all three methods identified distinct 'healthy,' 'unhealthy,' and 'mixed' behavioral patterns.
• Common themes across methods included high fruit/vegetable intake and outdoor play ('healthy'); poor diet and high sedentary behavior ('unhealthy'); and high physical activity, poor diet, and low sedentary time ('mixed').

Conclusions:

• A good degree of concordance in pattern characteristics was observed across PCA, LPA, and CA, despite differences in the number of patterns derived.
• Discrepancies in derived patterns can be attributed to the underlying statistical techniques of each method.
• Acknowledging methodological differences and ensuring thorough documentation are crucial for comparing behavioral patterns across studies using various analytical approaches.