Predicting distributional profiles of physical activity in the NHANES database using a Partially Linear Single-Index Fréchet Regression model

Object-oriented data analysis is a fascinating and developing field in modern statistical science with the potential to make significant and valuable contributions to biomedical applications. This statistical framework allows for the formalization of new methods to analyze complex data objects that capture more information than traditional clinical biomarkers. The paper applies the object-oriented framework to analyzing and predicting physical activity measured by accelerometers. As opposed to traditional summary metrics, we utilize a recently proposed representation of physical activity data as a distributional object, providing a more sophisticated and complete profile of individual energetic expenditure in all ranges of monitoring intensity. For the purpose of predicting these distributional objects, we propose a novel hybrid Frechet regression model and apply it to US population accelerometer data from NHANES 2011-2014. The semi-parametric character of the new model allows us to introduce non-linear effects for essential variables, such as age, that are known from a biological point of view to have nuanced effects on physical activity. At the same time, the inclusion of a global for linear term retains the advantage of interpretability for other variables, particularly categorical covariates such as ethnicity and sex. The results obtained in our analysis are helpful from a public health perspective and may lead to new strategies for optimizing physical activity interventions in specific American subpopulations.