Predictive Modeling of Medication Adherence Using Machine Learning Techniques

Abstract

This thesis aims to explore the correlation between socioeconomic factors and medication compliance in the sample of both German and Greek population based on machine learning technique. The kind of machine learning was unsupervised clustering and the algorithms utilized were K-means and hierarchical clustering.The study utilized two datasets: the German dataset, comprising 429 patients recruited from the neurology clinic at Jena University Hospital, and the Greek dataset, consisting of 81 individuals drawn from the general population. The German patients were slightly older (mean age 63.54 years) and the overall rate of adherence was higher than the Greek sample, most probably reflecting the structured clinical conditions approach of the study center. On the other hand, the Greek sample having a younger average age (average age = 54.9 years) had a broader span of adherence behaviors. The majority of factors that were positively associated with better adherence were age, caregiver involvement in the medication preparation, unemployment and lower levels of education. In the German dataset,older patients who relied on caregivers for medication management were more likely to struggle with adherence. On the other hand, in the Greek sample, older participants who managed their own medications tended to demonstrate better adherence, emphasizing the importance of self-management in maintaining medication routines. The K-means clustering algorithm identified distinct adherence patterns across various demographic groups, while hierarchical clustering provided a deeper understanding of the interplay between factors such as the caregiver interaction in medication preparation and socioeconomic status. The findings underscore the critical need for designing tailored healthcare interventions that comprehensively address both demographic and socioeconomic factors. Such interventions are essential to effectively enhance medication adherence, taking into account the diverse needs and challenges present in both clinical and general population settings. By acknowledging the unique influences of age, education, employment, and caregiving roles, these strategies can foster more equitable and sustainable adherence outcomes across varied demographic groups.