Effect of day-to-day variation in dietary intake on dietary patterns identification through factor analysis

Short-term dietary assessments, such as 24-h recalls, are affected by day-to-day variation, influencing the stability and identification of dietary patterns through factor analysis. We evaluated the effect of within-person random error on replicating these patterns.

Twenty non-consecutive 24-h recalls were completed by 302 adults over about three months. Individual usual intakes were defined as the average of 20 days. Factor analyses identified: (i) usual dietary patterns (20dDP), based on usual individual intakes; and (ii) n-day (n = 2, 3, or 4) dietary patterns (ndDP), based on all possible recall combinations. Factor mirroring and interchangeability ensured alignment for comparison between 20dDP and ndDP factor loadings distributions.

Three usual dietary patterns were identified: Prudent, Traditional, and Western, explaining 25% of the variance. Mean loading factors from 190 two-day combinations were weaker than the usual ones, with substantial attenuation in the third factor. Foods in the usual first and second factors were consistently identified in most 2-day combinations. In contrast, foods of the usual third pattern were identified largely by chance, depending on the sampled 2-day combination, and more foods not belonging to this factor were incorrectly assigned. This indicates that while two-day data capture major patterns, subsequent, less explanatory factors are highly susceptible to random error. Results were similar for 3 and 4-day combinations.

Usual dietary patterns can be identified using at least 2 non-consecutive collection days; however, the first factor is most stable, while subsequent factors show greater susceptibility to day-to-day variation, potentially compromising replication of usual dietary patterns.