Estimation of Habitual Intake of Infrequently Consumed Nutrients Using Mixture Distribution Method

Smitha Joseph^1,2Santu Ghosh³Sumathi Swaminathan¹Tinku Thomas^3*

• ¹St. John's Research Institute, Bangalore, India
• ²Center for doctoral studies, MAHE, Manipal, India
• ³St. John’s Medical College Hospital, Bangalore, India

Background: The habitual intake of infrequently consumed nutrients typically exhibits a highly skewed distribution, primarily driven by the reported consumption and nonconsumption of nutrients in repeated 24-hr dietary recalls. Current methods for estimating this distribution are often computationally intense. Methods: A mixture distribution method (MDM) was proposed to estimate habitual intake distribution of infrequently consumed nutrients, wherein the frequency of consumption of a nutrient was modelled using a beta-binomial distribution and the amount consumed using a gamma distribution. The habitual intake using this method was compared to the Iowa State University Foods method (ISUF) using sample data consisting of four non-consecutive 24-hr diet recalls collected from 120 children aged 6 to 59 months in Bihar, India. To assess the impact of zero inflation on the estimation of habitual intake, nutrient intakes were simulated with varying percentages of positive intakes and habitual intakes were calculated using both methods. Results: The median (IQR) habitual intakes estimated from MDM method and ISUF method were 0.47 mg (0.29, 0.65), and 0.46 mg (0.29, 0.62) for vitamin B6, 0.38 mcg (0.14, 0.68) and 0.40 mcg (0.18, 0.69) for vitamin B12, respectively. Similarly, comparable results were found for other nutrients such as vitamin B3, B5, B12, A and iodine. The simulated data showed that the estimated habitual intake by MDM increased with the proportion of positive intakes considering the higher probability of consumption. When the proportion of positive intakes was below 60%, the estimates using MDM which considers the probability of consumption were higher than the arithmetic mean calculated from fifteen recalls. Discussion: The proposed MDM method offers a computationally simpler approach to estimate habitual intake distribution by modelling both the probability distribution of nonconsumption and the distribution of positive intakes. The procedure is easily implemented using standard statistical software and estimates habitual intake for infrequently consumed nutrients from multiple 24-hr dietary recalls.