SW-4E - Use of Artificial Intelligence/Machine Learning Methods for Identification of Patients with Rare Diseases in Real-World Data

Background: Identifying patients with rare diseases is challenging due to low prevalence, underdiagnosis, unavailable diagnosis codes, and clinical heterogeneity. Artificial intelligence (AI) and machine learning (ML) approaches have been shown to improve and accelerate patient identification in real-world data, especially when key data elements are in unstructured formats. AI/ML approaches can enhance rare disease research and patient outcomes by enabling earlier diagnosis, early patient identification, risk stratification and study enrollment in observational and clinical studies, while also capturing critical information about patient’s health such as social determinants or laboratory results.

Objectives: In this symposium, we will discuss methodological and operational approaches for identifying patients with rare disease in real-world patient data using AI/ML methods.

Description: This session will:
Introduce the speakers and topics, providing rationale for the use of AI/ML methods to enhance patient identification and selection in rare disease. (Moderator: Julien Heidt, 5 min.)

Provide an in-depth discussion on the unique challenges of patient identification in rare diseases (e.g., phenotypic variability, delayed diagnosis, lack of standardized terminology, and small sample size) and their impact on study design, such as potential bias. (Laura Hester, 15 min.)

Discuss AI/ML methods for rare case patient identification and predicting outcomes in real-world clinical data: natural language processing (NLP) applied to unstructured electronic health record notes, large language models (LLMs), and predictive modeling. The interpretability of AI/ML models will also be discussed, including the strengths and limitations of each approach. (Jie Yang, 15 min.)

Discuss practical considerations for AI/ML methods through case studies, with highlight on methods and the importance of “expert-in-the-loop” workflows in producing research-ready insights more efficiently and robustly than traditional methods. (Christina Mack, 15 min.)

Examine regulatory and HTA considerations when using AI/ML in drug development, with emphasis on rare disease. Key considerations for data quality and model validation will be outlined. (Donna Rivera, 15 min.)

Conclude with an interactive discussion between the audience and panel, exploring opportunities for innovation. (Moderator: Julien Heidt, 20 min.)