Case Study: Finding APDS Patients Using Predictive Models

The Situation

Activated PI3K delta syndrome (APDS) is a rare immunodeficiency disorder caused by mutations in the PIK3CD or PIK3R1 gene that affect critical immune system functions. Patients with APDS suffer from severe recurrent infections, predominantly affecting the lungs, sinuses and ears, alongside symptoms such as lymphoproliferation and autoimmunity.

A midsize pharmaceutical company’s main objective was to enhance the early identification of patients with APDS using predictive machine learning (ML) modeling techniques applied to administrative claims databases. EVERSANA utilized an extensive medical claims dataset, targeting a deeper understanding and earlier diagnosis through predictive modeling, addressing a significant unmet need in APDS patient care.

The Solution

Finding APDS patients is complicated due to the absence of a specific ICD-10 code and the variability of symptoms among affected individuals, even though they share the same genetic mutation. The challenges extended to constructing a predictive model capable of accurately finding APDS phenotypes in administrative claims databases.

EVERSANA segmented the data into two age-based cohorts to refine analysis and model accuracy. Predictive modeling used ML algorithms to analyze patterns within the claims data, focusing on key indicators such as immune disorders, lymphadenitis and respiratory infections. The team also employed various data validation methods, such as comparing cohort data with the European Registry and patients with confirmed genetic testing results, ensuring robustness and accuracy. By integrating clinical and technical expertise, and comparisons with cohorts of patients with a diagnosis confirmed with genetic testing, the model could more effectively predict APDS phenotypes and potential treatment paths. These models were then applied to find in the database undiagnosed patients who are candidates for genetic testing to confirm the diagnosis.

The Results

The EVERSANA model achieved high precision and sensitivity, underscoring its effectiveness in finding potential APDS patients, who are otherwise invisible to the healthcare system. Ultimately, the project demonstrates the transformative potential of predictive analytics in healthcare, paving the way for broader applications in managing other complex diseases.