A study on healthcare expenditure in Italian regions via Symbolic Regression

Abstract

The study of the factors driving the dynamics of healthcare spending is of paramount importance to guide policymakers in the allocation of resources and to measure the effectiveness of the healthcare system under consideration. The identification of these drivers can be supported by the use of machine learning techniques, which enable the discovery of hidden patterns within vast amounts of data. In contrast to black-box methods, Symbolic Regression (SR) is an approach that allows for the identification of analytical models that explicitly capture the functional relationships within the data, thus enhancing interpretability. In this paper, we present the use of SR for identifying the drivers of healthcare expenditure in Italian regions. Given the dynamic and complex nature of this phenomenon, we generated several models based on distinct temporal windows, enabling us to analyze the drivers across different time horizons. In addition to identifying the main drivers based on variable frequency in the generated models, we also conducted a study on recurring substructures. The results show that SR was able to generate models with a good level of predictive accuracy for private healthcare expenditure, enabling a reliable analysis of its driving factors, but failed to do so for public healthcare expenditure.