Информационный анализ моделей планирования медико-биологического эксперимента при контроле дискретных изменений уровней влияющего фактора

Abstract

В статье рассмотрены две базовые статистические модели для обозначения эффектов уровня влияющего фактора, а именно параметрическая модель и модель компонент дисперсий. Проведен сравнительный анализ эффективности двух моделей однофакторного медицинского эксперимента с односторонней классификацией результатов по получаемой информации, учитывающей априорную неопределенность модели. Получены аналитические выражения для определения информационного содержания рассматриваемых моделей исходя из параметров плана медицинского эксперимента. Проведен анализ рассмотренных статистических моделей в результате, которого определены эффекты влияния каждого из параметров на эффективность использования модели и приведены соответствующие рекомендации по их использованию.The paper considers the main approaches of modern evidence-based medicine and the role of the medical-biological experiment in it. It is established that one of the most used method in medical practice is the method of single-factor variance analysis (ANOVA). Two basic statistical models for designating effects of the level of the influencing factor (parametric model and the model of the variance components) are considered. Objective: A comparative analysis of the effectiveness of two models of a single-factor medical experiment with a one stage classification of the results from the received information, taking into account the a priori uncertainty of the model. Solution methods: The positions of the information theory of measurements and the model of representation of the experimental results, taking into account the influence of the factor, were used for information analysis. In the analysis, two variants of the results of the biomedical experiment were considered: the factor influence takes place or does not take place in the groups under consideration. Expressions for variance are obtained in case of confirmation of the main or alternative hypothesis. On the basis of the analysis, expressions were obtained to determine the information content of the models under consideration based on the parameters of the medical experiment plan. The analysis of expressions for estimating the amount of information contained allowed us to compare two basic statistical models for describing effects of the level of the influencing factor in the analysis of variance: parametric and random. Results: For all other things being equal, the random model has a higher information content, since a priori it takes into account the presence of probabilistic properties in the levels of the influencing factor. Positive dependence of the information content of models on the number of groups of measurements is obtained. The model of the dispersion components is more sensitive to a change in this parameter of the experimental plan than the parametric one. Conclusions: The obtained analytical expressions allow estimating the effect of the parameters of the experimental plan, namely the number of groups, the volume of samples, the level of significance, on the informational content of models, and use this parameter as the criterions of the choice of that one or another model.