Modeling Usual Nutrient Intake Distribution: A Comprehensive Comparative Study Using Hierarchical Models

Accurate estimation of the usual intake distribution is essential for assessing population-level nutritional risk and informing public health policy. Traditional approaches to dietary assessment often rely on short-term instruments, such as 24-h recalls, which are subject to significant within-individual variability.

This study comprehensively aims to compare several modeling approaches for usual intake distribution estimation and then to evaluate the impact of model selection on nutrient inadequacy prevalence estimation.

Dietary intake data were collected using repeated 24-h recalls (2 or 3 recalls) from a representative sample of individuals (n = 5800, Enquête Individuelle et Nationale sur les Consommations Alimentaires 3), in which weekends/holidays and weekdays were equalized with the survey's suggested weighting method. Proposing a model to capture the maximum variability in the data to accurately estimate usual intake has long been a methodological challenge. To address this issue, various statistical models were applied to estimate usual intake distribution, including crossed and nested random-effects models, as well as traditional regression-based approaches. After performance evaluation, descriptive statistics and prevalence of nutrient inadequacy were compared across models.

Although the descriptive statistics of usual intake distributions were broadly similar across models, the prevalence of nutrient inadequacy varied substantially depending on the model used. Hierarchical models, particularly those with nested random effects, demonstrated superior fit and lower error rates and yielded prevalence estimates that differed from those based on simpler classical models, as expected with observed data. These differences were most pronounced for nutrients with high within-individual variability.

The model choice can significantly affect the estimated prevalence of nutritional inadequacy. Hierarchical modeling approaches that account for within-individual variation provide more accurate and reliable estimates, underscoring the importance of rigorous statistical methodology in nutritional epidemiology. These findings have significant implications for the assessment of nutritional risk and the development of evidence-based public health interventions.