Pompe disease is a rare, chronically debilitating metabolic disorder caused by deficiency of the enzyme acid alpha-glucosidase. As enzyme replacement therapy has extended patient survival, those living with the disease may experience new long-term manifestations that established clinical trial endpoints were not designed to capture. At the same time, many clinically meaningful endpoints do not map directly to codes available in routine healthcare records – leaving a gap between what patients experience and what the evidence base reflects, with consequences for disease monitoring, HTA submissions, and trial design.
This research, conducted by Volv Global in collaboration with Sanofi and presented at ISPOR 2026, addressed that gap using machine learning applied to the Komodo Health US administrative claims database, encompassing 3,549 confirmed Pompe disease patients – the largest machine learning study of the disease conducted to date.
What the research demonstrates
The study makes three sequenced methodological contributions, each building on the last:
Validating the patient cohort and confirming measurability in claims data
The prevalence of literature-based disease symptoms was compared between the Pompe patient cohort and a matched control population without Pompe disease. All literature-based symptoms were significantly more prevalent in the Pompe cohort than in controls – confirming both that the correct patients are represented in the claims data, and that these symptoms are reliably measurable within it. This foundational validation step underpins all subsequent analyses.
Mapping clinician-defined endpoints to real-world claims codes
Machine learning models were applied to map 46 of 67 pre-specified clinical endpoints to diagnosis, procedure, and treatment codes available in claims data, demonstrating that endpoints designed for clinical trials can be reliably tracked in routine healthcare data. All mapped endpoints were significantly more prevalent in the Pompe cohort than in controls.
Discovering novel disease features beyond pre-specified frameworks
An unsupervised, data-driven discovery approach identified novel cardiovascular, respiratory, and systemic disease features that were highly prevalent in the Pompe cohort but absent from any pre-specified endpoint framework. These findings were confirmed against the same control population. Treated patients showed higher prevalence of both mapped and newly discovered endpoints than untreated patients – reflecting longer survival and the emergence of new manifestations under therapy. The newly identified features offer clinically grounded candidates for endpoint design in future natural history studies and clinical trials.
Why this matters
For rare diseases where treatment has fundamentally changed the patient trajectory, evidence frameworks built around earlier disease understanding can fall out of step with clinical reality. This research demonstrates that machine learning can close that gap in a structured and reproducible way – confirming what is already known, and independently discovering what the data reveals. The methodology is applicable across rare and difficult-to-diagnose diseases where treatment advances have outpaced existing evidence frameworks, and provides a foundation for more rigorous endpoint selection, natural history evidence, and clinical trial design.
About the study
- Disease: Pompe disease (infantile and late onset)
- Data source: Komodo Health US administrative claims database
- Cohort: 3,549 confirmed Pompe disease patients
- Control population: Matched patients with mimic disease codes and no Pompe disease history in the preceding seven years
- Presented at: Poster Session 1, ISPOR Global 2026, Philadelphia, PA, 18 May 2026 (Poster RWD22)
- Collaboration: Volv Global and Sanofi
All findings are based on retrospective analysis of claims data. Prospective validation in an independent cohort has not been conducted.
Links:
- Volv Global at ISPOR 2026 – event page: https://volv.global/volv-global-at-ispor-2026-with-sanofi/
- ISPOR Global 2026: https://www.ispor.org/conferences-education/conferences/upcoming-conferences/ispor-2026
- Published abstract: https://www.ispor.org/heor-resources/presentations-database/presentation-cti/ispor-2026/poster-session-1-4/real-world-endpoint-mapping-and-identification-of-evolving-phenotypes-of-pompe-disease-using-machine-learning
- Volv Global: Value & Access, HEOR: https://volv.global/value-access-heor/
- Other posters: ATS 2025
- Other posters: ATS 2024
- Other posters: AHP (not ready yet)
