A New Approach to Estimate Habitual Intake of Nutrients with Skewed Distribution.

Abstract

Background: Habitual nutrient intake in a population is typically estimated using multiple 24-h recalls. Existing methods to estimate the habitual intake of skewed distributions are computationally intense.

Objectives: To propose a new method to estimate habitual nutrient intake with skewed intake distributions.

Methods: The proposed gamma regression method for estimating habitual intake distribution is compared with National Research Council and Iowa State University (ISU) methods using sample data of 4 nonconsecutive diet recalls collected from 120 children aged 0.5-5 y in Bihar, India. The gamma regression and ISU methods were considered comparable when the estimated difference in the habitual nutrient intake and its 95% confidence interval included the 8.2% equivalence limit for that nutrient. The impact of skewness on the habitual intake estimation was compared by simulating data with varying degrees of skewness.

Results: The median (quartile 1, quartile 3) intakes estimated from gamma regression, ISU, and National Research Council methods were respectively 896 kcal (757, 1043 kcal), 895 kcal (752, 1054 kcal), and 893 kcal (748,1045 kcal) for energy, 22.6 g (19.5, 28.9 g), 22.6 g (19.5, 29.6 g), and 22.7 g (19.5, 29.5 g) for protein, 5.8 mg (3.3, 7.7 mg), 6.1 mg (3.3, 8.3 mg), and 6.1 mg (3.3, 8.2 mg) for iron, and 107 mcg retinol activity equivalents (RAE) (75, 134 mcg RAE), 114 mcg RAE (80, 143 mcg RAE), and 113 mcg RAE (79, 143 mcg RAE) for vitamin A. The estimates of percent bias of the gamma regression method were within the 8.2% equivalence limit, 0.32% (-0.03%, 0.67%) for energy, 0.28% (-0.14%, 0.70%) for protein, 4.36% (1.51%, 7.21%) for iron, and 3.53% (0.74%, 6.33%) for vitamin A. Simulation demonstrated that gamma regression method is comparable to ISU method for different levels of skewness.

Conclusions: The simpler gamma regression method offers a viable alternative to address skewness in intake distributions by providing equivalent estimates without the complexity of the 2-step transformation of the ISU method.