The Situation
Paroxysmal nocturnal hemoglobinuria (PNH) is a rare and life-threatening condition that poses significant management challenges to the medical community. Traditionally, SOLIRIS® (eculizumab) has been the primary treatment. However, with the advent of ULTOMIRIS® (ravulizumab), which offers similar therapeutic benefits but with less frequent infusions, understanding the dynamics influencing treatment switching has become crucial. This knowledge can lead to improved patient outcomes and more refined treatment strategies.
The client’s goal was to identify the factors influencing the switch from Soliris to Ultomiris from both the patient and physician perspectives. EVERSANA was contracted to develop predictive models that analyzed extensive demographic, clinical and therapeutic data to construct robust frameworks that would anticipate trends and optimize treatment strategies for managing PNH.
The Solution
One of the main challenges was managing the high-dimensional healthcare data, which includes diverse patient profiles, treatment patterns and outcomes. Integrating vast datasets to identify significant predictors required advanced analytical techniques and meticulous modeling. Additionally, overcoming biases inherent in retrospective healthcare claims data and ensuring model applicability across different patient demographics posed significant hurdles.
EVERSANA’s approach involved developing three predictive models: a physician-focused model, a patient-focused model and a hybrid model incorporating both sets of features. These models analyzed data extracted from healthcare claims, including treatment histories, healthcare provider characteristics, patient demographics and clinical outcomes. We employed advanced machine learning algorithms and statistical techniques, such as gradient boosting and mutual information analysis, to enhance prediction accuracy regarding treatment-switching behavior.
The Results
The findings of this study underscore the significant roles both patient characteristics and physician practice behaviors play in PNH management decisions. Key insights revealed:
- Younger patients with frequent hospitalizations and those treated by experienced hematologists are more likely to switch treatments.
- Older, sicker patients; those with less response in their Soliris treatment; and those treated by more experienced physicians are more likely to switch treatments.
These models enhance the client’s ability to forecast treatment transitions, offering valuable insights that could lead to more personalized patient care and better overall treatment outcomes. The predictive models designed during this project not only demonstrate EVERSANA’s capability in handling complex datasets, but also validate our methodological approaches in deriving actionable insights from healthcare data. Our deep dive into PNH treatment dynamics provides a clear example of how predictive analytics can significantly enhance therapeutic decision-making, thus offering substantial value to healthcare providers, payers and patients alike.
Plots comparing feature importance between the patient-based model, the physician-based model and the hybrid patient-physician-based model.
