EHR Data Analysis for Predicting Medication Non-Adherence

Context

A client looking to establish a Real-World Data (RWD) team approached Rang Technologies for expertise in Electronic Health Records (EHR) data analysis. The objective was to analyze a large EHR database containing patient demographics, diagnoses, medications, lab results, and event data to identify factors contributing to medication non-adherence in patients with chronic conditions. By pinpointing key demographic, clinical, and lifestyle variables influencing missed doses, healthcare providers could proactively intervene and enhance treatment outcomes.

Resolution

Our team eagerly took on this challenge, leveraging supervised learning algorithms such as logistic regression to classify patients as adherent or non-adherent based on their clinical data.

The approach involved: 

• Extracting and preparing key variables like age, socioeconomic status, medication type, past medication history, and recent lab results.
• Developing a predictive model using logistic regression, decision trees, and random forests to assess the likelihood of non-adherence.
• Optimizing the model through feature selection and training to enhance accuracy.

Result

The model’s performance was evaluated using key metrics such as accuracy, precision, recall, and AUC-ROC, ensuring its reliability in identifying high-risk patients. Early identification of non-adherent patients enabled proactive interventions, helping lower medical expenses while enhancing patient care and treatment results. Additionally, data pattern analysis provided valuable insights into the root causes of non-adherence, informing future research and policy decisions in medication management.