Early Detection of Rare Diseases Using Real-World Data

SITUATION

A Fortune 500 pharmaceutical company wanted to accelerate its R&D efforts in rare diseases. But with limited real-world data and underdiagnosed patient populations, feasibility assessments and early clinical investment decisions were difficult to justify.

CHALLENGE

Rare diseases are notoriously hard to diagnose. With 90% lacking FDA-approved treatments and minimal physician awareness, most patients are misdiagnosed or untreated for years. The client needed a model to identify at-risk patients early in their diagnostic journey long before a specialist might intervene.

APPROACH

Mu Sigma built a suite of AI models using EMR and insurance claims data to detect early signs of a specific rare disease:

• Predictive Modeling:Leveraged XGBoost and Random Forest to build an ensemble detection model
• Feature Engineering: Identified 4 new early-stage indicators using historical RWD
• Medical Marker Validation: Collaborated with physicians to validate clinical “red flags” for the condition
• Physician Mapping: Located doctors already interacting with potential patients to enable better targeting

IMPACT

• Increased estimated eligible patient pool from 0.5% to 7%, improving drug pricing and market viability
• Identified 95% of patients with the rare disease in test sets
• Discovered 4 new predictive indicators
• Validated key clinical symptoms, enabling better awareness campaigns and earlier interventions