Case Study: Improving an AATD Predictive Model Using EMR Data

The Situation

Alpha-1 antitrypsin deficiency (AATD) is a genetic condition with significant variability in clinical presentation and outcomes, making accurate diagnosis challenging yet crucial. EVERSANA was initially approached by a midsize pharmaceutical company to perform advanced predictive modeling to improve the identification of AATD among patients using electronic medical records (EMRs) and open claims. The original model, validated using initial EMR datasets, was designed to optimize the prediction and management of AATD by incorporating comprehensive diagnostic, medication and laboratory features.

The project aimed to enhance the diagnostic precision and reliability of the predictive model for AATD by validating and retraining it with a newly integrated EMR dataset. Key objectives included increasing predictive accuracy metrics, refining the patient definition to ensure a more precise diagnosis and extending the model’s applicability across different EMR systems without compromising performance.

The Solution

Initial analysis revealed a decrease in model performance when applied to a new EMR dataset, with predictive accuracy dropping notably. This was attributed to variations in the availability and distribution of clinical data across EMRs. Additionally, the complexities of integrating and normalizing data from various sources posed significant hurdles in maintaining consistency in model training and validation processes.

To address these challenges, the team implemented a comprehensive retraining of the analytic model. This involved revising the patient cohort definitions based on updated clinical data and diagnostics, employing advanced feature selection techniques such as mutual information to refine the predictive variables, and retraining the model with new data subsets to test and enhance its performance under varied clinical scenarios. Ongoing collaboration between the clinical and technical teams to get feedback and address the integration of data and model adjustments was essential to achieving the desired outcome.

The Results

The retrained model demonstrated a significant improvement, increasing the predictive accuracy from 78% to 85%. The enhanced model now effectively incorporates a wider range of clinical features, making it more representative and inclusive of diverse patient populations. Sensitivity and specificity analyses indicated high efficacy in diagnosing true positive and negative cases, thus validating the model’s utility in a clinical setting.

This predictive modeling capability not only promises to improve diagnostic accuracy for AATD, but also serves as a scalable blueprint for other genetic and chronic conditions. For potential clients, adopting this model can translate to better patient outcomes, optimized treatment pathways and, ultimately, a higher standard of personalized care.

This comprehensive project summary encapsulates EVERSANA’s capabilities in harnessing predictive modeling for enhancing disease diagnosis and management, showcasing a profound impact on healthcare delivery through innovation and technology integration.