Introduction: The assessment of actual nutrition of the population, both at the individual and population level, strongly depends on the accuracy of data on the chemical composition of food products. Milk is an important component of a diet, and a precise estimation of micro- and macronutrients consumed with it is essential for public health assessment.
Objective: To develop an algorithm for obtaining statistically accurate values of average concentrations and variability of basic micro- and macronutrients in milk.
Materials and methods: To elaborate and check the algorithm, we used milk fat test results collected within the Federal Project on Public Health Strengthening by the laboratories of the Federal Service for Consumer Rights Protection and Human Wellbeing (Rospotrebnadzor) in the years 2020–2021.
Results: Due to numerous missing and outlying values of milk composition testing, an appropriate adjustment of the algorithm was necessary. The best separating ability was demonstrated by the approach of dividing types of milk into clusters based on their fat and calcium content. The three clusters obtained included milk with a 2.5 % fat content and the average calcium concentration of 1,144 mg/L, milk with a 3.2 % fat content and the average calcium concentration of 1,180 mg/L, and milk with both fat contents and the mean calcium level of 597 mg/L. The algorithm was validated by checking the completeness of data on the fatty acid composition and a low variability of values.
Conclusion: The developed algorithm has enabled us to obtain up-to-date information on the chemical composition of milk sold by food retailers in the Russian Federation. Low-calcium milk on the market is of special concern as its average consumption fails to satisfy human physiological needs. At the same time, the content of saturated fat was below 2.2 g/100 g in the cluster of milk types with the maximum fat content, thus raising no additional health concerns. Further studies should be aimed at determining the acceptable and correct stages of data preprocessing that maintain a balance between the obtained accuracy of values and their actual reproducibility.
