| Abstract [eng] |
Digital therapy is a new way to treat and prevent chronic diseases. Digital therapy combines physician recommendations, software, data analysis, and artificial intelligence to provide personalized support and fit individual needs. For example, diabetes is a chronic disease that can be prevented or controlled by physical activity, a proper meal, and continuous monitoring of various other parameters. Specialists usually provide general meal recommendations, and it will be difficult for a person to change certain foods without the necessary knowledge about nutrients. The development and implementation of mathematical models of digital therapy is relevant today. Literature usually provides a single method for meal planning. Comparing meal plans according to the methods used and constrains applied is rarely analyzed. Dietary plans do not adapt to an individual person's physical activity on a daily basis and do not take into account all the products a person wants to consume when he or she wants to switch from one food to another for some reason. The developed model compiles and periodically revises the individual meal plan for a human according to the list of foods of his choice and the physical activity data received from the sensors and based on the healthy meal and doctor's recommendations. This model will be used as a digital therapy prototype in the company dHealthIQ [1]. Based on the American Diabetes Association recommendations, the model can suggest different meal plans based on the type of carbohydrate content, the type of caloric content, and the different number of restrictions. Linear programming, weighted goal programming methods, genetic algorithm, and particle swarm algorithm were applied. To adjust the meal plan, the duration of physical activity is predicted using time series prediction methods. The most accurate method is selected for every period of forecast and assumptions are checked. Weighted goal programming is suitable for meal plans models with 3 and 6 constraints, and the genetic algorithm is suitable for models with 12 constraints in terms of accuracy, time, and memory cost. High carbohydrate type meal plans are the most accurate, and low carbohydrate meal plans have larger deviations. The meal plans include all the foods of human choice and are distributed appropriately: fruits, vegetables, and meat take the most quantity and fats, grains, sweets take the least. Predicting the duration of physical activity helps to make more appropriate meal plans for a human. Papers on the related topic of the final work were presented at the conferences "Mathematics and Natural Sciences: Theory and Application" in 2021 and 2022 (articles accepted for publication), also report is included in the 63rd conference program of the Lithuanian Mathematical Society and will be read in 2022 June 16. |
