Exploring Medication Effectiveness Using Machine Learning: A Multiclass Treatment Outcome Analysis
Abstract
Understanding how medication influences patient treatment outcomes is critical for improving healthcare decision-making. According to the Centers for Disease Control and Prevention (CDC, 2024), medicines are safe when used as prescribed; however, adverse drug events occur when medications cause harm, including side effects or medication errors. Medication errors harm at least 1.5 million people annually and cost approximately $3.5 billion each year in hospital-related expenses alone (Institute of Medicine, 2007). While clinicians rely on medical expertise, a significant gap exists in using automated data-driven approaches to reveal subtle medication patterns that are not immediately apparent in traditional analysis. This study examines whether medication-related features, including drug type, treatment duration, dosage, and reported side effects, can classify patient improvement levels using machine learning techniques. We hypothesized that medication-level data would significantly predict patient improvement outcomes. The dataset was cleaned, encoded, and split into training and testing sets, and the improvement score was categorized into three levels: Low, Medium, and High. Multiple supervised learning models, including Random Forest, Decision Tree, Logistic Regression, Support Vector Machine (SVM), Gradient Boosting, and AdaBoost—were trained and evaluated using accuracy and F1-weighted metrics. Results indicate moderate predictive performance, with the best-performing model achieving 48.33% accuracy. Although predictive accuracy was modest, findings suggest that medication-level data contain meaningful patterns associated with treatment improvement. The study highlights the complexity of healthcare outcomes and suggests that incorporating richer clinical and patient-specific variables may enhance predictive performance. Overall, this research demonstrates the potential of machine learning as a supportive tool for healthcare analytics and informed treatment decision-making.
Start Time
15-4-2026 1:30 PM
End Time
15-4-2026 4:30 PM
Room Number
Culp Ballroom 316
Presentation Type
Poster
Student Type
Graduate and Professional Degree Students, Residents, Fellows
Faculty Mentor
Ghaith Husari
Exploring Medication Effectiveness Using Machine Learning: A Multiclass Treatment Outcome Analysis
Culp Ballroom 316
Understanding how medication influences patient treatment outcomes is critical for improving healthcare decision-making. According to the Centers for Disease Control and Prevention (CDC, 2024), medicines are safe when used as prescribed; however, adverse drug events occur when medications cause harm, including side effects or medication errors. Medication errors harm at least 1.5 million people annually and cost approximately $3.5 billion each year in hospital-related expenses alone (Institute of Medicine, 2007). While clinicians rely on medical expertise, a significant gap exists in using automated data-driven approaches to reveal subtle medication patterns that are not immediately apparent in traditional analysis. This study examines whether medication-related features, including drug type, treatment duration, dosage, and reported side effects, can classify patient improvement levels using machine learning techniques. We hypothesized that medication-level data would significantly predict patient improvement outcomes. The dataset was cleaned, encoded, and split into training and testing sets, and the improvement score was categorized into three levels: Low, Medium, and High. Multiple supervised learning models, including Random Forest, Decision Tree, Logistic Regression, Support Vector Machine (SVM), Gradient Boosting, and AdaBoost—were trained and evaluated using accuracy and F1-weighted metrics. Results indicate moderate predictive performance, with the best-performing model achieving 48.33% accuracy. Although predictive accuracy was modest, findings suggest that medication-level data contain meaningful patterns associated with treatment improvement. The study highlights the complexity of healthcare outcomes and suggests that incorporating richer clinical and patient-specific variables may enhance predictive performance. Overall, this research demonstrates the potential of machine learning as a supportive tool for healthcare analytics and informed treatment decision-making.
