Sanofi logo

Sanofi: Transforming Rare Disease Benefit-Risk Assessment

Rare Diseases Immunology Submission Post-Marketing Bayesian BRA
52%

Faster regulatory submissions

85%

Signal detection accuracy improvement

6

Rare disease products covered

34M+

Real-world patient records integrated

Overview

Sanofi's Specialty Care Global Business Unit oversees one of the pharmaceutical industry's most diverse rare disease portfolios, spanning lysosomal storage disorders, rare hematology, rare endocrinology, and immunology. Key products include Dupixent (dupilumab) for atopic dermatitis and eosinophilic conditions, Cerdelga (eliglustat) for Gaucher disease type 1, Fabrazyme (agalsidase beta) for Fabry disease, Myozyme/Lumizyme (alglucosidase alfa) for Pompe disease, and Aubagio (teriflunomide) for relapsing multiple sclerosis.

In early 2024, Sanofi engaged ArcaScience to address a fundamental challenge inherent to rare disease pharmacovigilance: the limitations of traditional frequentist signal detection methods when applied to small, heterogeneous patient populations. The engagement covered benefit-risk assessment for six rare disease products, with particular focus on Dupixent's expanding indication set and Cerdelga's long-term post-marketing safety evaluation.

The Challenge

Rare disease pharmacovigilance is fundamentally different from safety monitoring in common therapeutic areas. With patient populations often numbering in the hundreds or low thousands, traditional disproportionality-based signal detection methods -- which rely on large denominators to achieve statistical significance -- frequently fail to identify genuine safety signals or produce excessive false positives.

Sanofi's rare disease safety team faced several compounding challenges. For Cerdelga, with a global treated population of approximately 1,500 patients with Gaucher disease type 1, standard PRR (Proportional Reporting Ratio) and ROR (Reporting Odds Ratio) analyses generated unstable estimates with wide confidence intervals, making it nearly impossible to distinguish true signals from statistical noise. Safety scientists were spending an average of 22 weeks per PBRER, with 40% of that time dedicated to manual reconciliation of conflicting signal detection outputs.

For Dupixent, the challenge was different but equally complex. As Sanofi pursued new indications in eosinophilic esophagitis, chronic rhinosinusitis with nasal polyps (CRSwNP), chronic obstructive pulmonary disease (COPD), and prurigo nodularis, the safety team needed to assess benefit-risk profiles across vastly different patient populations, disease pathophysiology, and co-morbidity profiles. Each indication required separate signal evaluation with different background incidence rates, comparator therapies, and benefit-risk frameworks -- but the EMA expected a unified, cross-indication safety assessment in each PBRER submission.

The challenge was compounded by a lack of reliable real-world evidence for rare disease populations. Registry data was fragmented across multiple disease-specific patient registries (ICGG Gaucher Registry, Fabry Registry, ADVANCE Rare Disease Database), each with different data collection standards, coding systems, and levels of completeness. Sanofi estimated it would take 2-3 additional FTEs exclusively dedicated to data harmonization to maintain the current submission cadence -- and the EMA's updated GVP Module VIII guidance on post-authorization safety studies was adding further requirements for real-world evidence integration.

The Solution

ArcaScience deployed a specialized rare disease configuration of its platform, built around Bayesian statistical models designed for small-sample safety assessment, combined with a real-world evidence integration layer purpose-built for rare disease registries and claims databases.

Bayesian Signal Detection for Small Populations

For Sanofi's rare disease portfolio, ArcaScience deployed Bayesian hierarchical models that borrow strength across related products and indications to improve signal detection sensitivity in small populations. For Cerdelga, instead of relying solely on the 1,500-patient treated population, the platform's models incorporated informative priors derived from the broader lysosomal storage disorder safety literature, the Gaucher disease natural history, and pharmacologically related compound classes. This approach reduced the minimum detectable signal threshold by 73% compared to traditional frequentist methods, while maintaining a false positive rate below 5%.

Real-World Evidence Integration

The Data Intelligence Engine was configured to ingest, harmonize, and continuously update data from the ICGG Gaucher Registry (6,200+ patients), Fabry Registry (4,800+ patients), ADVANCE Rare Disease Database, French BNDMR (Banque Nationale de Donnees Maladies Rares), Orphanet epidemiological data, and 34 million patient records from CPRD GOLD, Optum, and MarketScan claims databases for background incidence rate estimation. ArcaScience's rare disease data harmonization models mapped heterogeneous coding systems (ICD-10, ORPHA codes, HPO terms, MedDRA) to a unified ontology, enabling cross-registry signal evaluation for the first time across Sanofi's full rare disease portfolio.

Cross-Indication Benefit-Risk Framework for Dupixent

For Dupixent, ArcaScience built a multi-indication benefit-risk framework using structured BRAT methodology with indication-specific value trees and shared safety endpoints. The platform's MCDA models applied indication-specific preference weights derived from published patient preference studies and clinician conjoint analyses, while maintaining a unified adverse event evaluation across all six indications. This enabled Sanofi to produce a single, coherent cross-indication benefit-risk narrative for the PBRER while preserving the granularity required for indication-specific risk management plan updates.

Automated Regulatory Outputs

The platform generated PBRERs in ICH E2C(R2) format, Risk Management Plan (RMP) updates per EMA GVP Module V, and post-authorization safety study (PASS) protocols per GVP Module VIII. For new indication submissions, ArcaScience generated integrated benefit-risk summaries for CTD Module 2.5, effects tables per FDA structured benefit-risk framework, and Bayesian interval estimates for rare adverse events compliant with ICH E2E guidance on pharmacovigilance planning.

Platform Modules Used

Data Intelligence Engine Decision Intelligence Bayesian BRA Models Regulatory Outputs

Implementation Timeline

16 weeks

Products Covered

Dupixent (dupilumab)

Cerdelga (eliglustat)

Fabrazyme (agalsidase beta)

Myozyme/Lumizyme

+ 2 additional rare disease products

Regulatory Deliverables

PBRERs, RMP updates, PASS protocols, CTD Module 2.5, Bayesian BRA reports

Results & Impact

52%

Faster Regulatory Submissions

Average PBRER completion time for the rare disease portfolio dropped from 22 weeks to 10.5 weeks. For Dupixent's new indication submissions, the integrated benefit-risk summary was completed in 6 weeks instead of the 14 weeks previously required. This acceleration enabled Sanofi to meet aggressive regulatory filing timelines for the eosinophilic esophagitis and COPD indications without adding headcount to the safety team.

85%

Signal Detection Accuracy Improvement

ArcaScience's Bayesian signal detection models achieved an 85% improvement in sensitivity for clinically relevant signals in small rare disease populations, compared to Sanofi's previous frequentist disproportionality methods. For Cerdelga, the platform identified a cardiac conduction signal (PR interval prolongation) that had been masked in traditional analyses by the drug's known mechanism of action as a glucosylceramide synthase inhibitor. The false positive rate simultaneously decreased from 18% to 4.2%.

34M+

Real-World Records Integrated

ArcaScience harmonized and integrated data from 34 million real-world patient records across 7 rare disease registries and 3 claims databases, establishing background incidence rates for 240+ adverse events across Sanofi's rare disease therapeutic areas. This real-world evidence base was cited in 4 EMA PBRER responses as a "significant methodological improvement" in safety signal contextualization for rare disease products.

6

Products Unified on Single Platform

All six rare disease products are now managed on a single ArcaScience platform instance, with unified data governance, consistent signal detection methodology, and standardized benefit-risk frameworks. This eliminated the previous fragmented approach where each product team used different statistical methods and reporting templates, enabling Sanofi to demonstrate methodological consistency across the portfolio in regulatory interactions.

"In rare diseases, the absence of evidence is not evidence of absence -- but traditional pharmacovigilance methods treat it that way. ArcaScience's Bayesian approach fundamentally solved a problem we had been struggling with for years. For Cerdelga, we went from signal detection outputs that our safety physicians couldn't act on -- because the confidence intervals were so wide they were meaningless -- to precise, clinically actionable intelligence. The cardiac conduction signal they identified would have taken us another 2-3 PBRER cycles to detect through our legacy methods. For Dupixent, the cross-indication framework gave us something we never had before: a unified benefit-risk narrative that the EMA explicitly praised."

Dr. Marc-Antoine Girard

Global Head of Pharmacovigilance, Specialty Care

Sanofi

Technical Details

Data Sources

  • FAERS (FDA): Spontaneous adverse event reports for all 6 products, including MedDRA-coded preferred terms and standardised queries (SMQs)
  • EudraVigilance (EMA): European ICSRs with E2B(R3) structured data for cross-source signal corroboration
  • ICGG Gaucher Registry: 6,200+ patients with Gaucher disease types 1, 2, and 3, with longitudinal treatment and outcome data (1991-present)
  • Fabry Registry: 4,800+ patients with Fabry disease receiving enzyme replacement therapy, including cardiac, renal, and neurological outcomes
  • ADVANCE Rare Disease Database: Multi-disease rare disease natural history data used for background incidence rate estimation
  • French BNDMR: Banque Nationale de Donnees Maladies Rares -- 48,000+ rare disease patients from French reference centers
  • CPRD GOLD, Optum & MarketScan: 34 million real-world patient records for background incidence rates and comparator safety profiling
  • PubMed, Embase & Cochrane: 2,900+ published studies on lysosomal storage disorders, immunology safety, and rare disease pharmacoepidemiology
  • Sanofi Internal Safety Database: Proprietary clinical trial and post-marketing data for all 6 products

AI Models Applied

  • Bayesian Hierarchical Signal Detection: Multi-level Bayesian models borrowing strength across related products and therapeutic classes, with informative priors derived from disease natural history data and pharmacologically similar compounds
  • Bayesian Interval Estimation: Credible interval calculation for rare adverse events in small populations, replacing unreliable frequentist confidence intervals with Bayesian posterior distributions
  • Rare Disease Data Harmonization Models: NLP and ontology mapping models converting heterogeneous coding systems (ICD-10, ORPHA, HPO, MedDRA) to unified rare disease ontology for cross-registry analysis
  • Cross-Indication MCDA Framework: Multi-criteria decision analysis with indication-specific value trees, shared safety endpoints, and sensitivity analysis across 6 Dupixent indications
  • Real-World Evidence Calibration: Bayesian models calibrating registry data completeness and ascertainment bias to produce adjusted background incidence rate estimates
  • Literature Surveillance Models: NLP-based continuous scanning of 200+ rare disease and immunology journals for emerging safety evidence and natural history updates

Validation Methodology

Validation for rare disease BRA required specialized approaches due to small sample sizes:

  • Prior Sensitivity Analysis: All Bayesian models were validated using prior sensitivity analysis, testing how results changed under non-informative, weakly informative, and strongly informative priors to ensure robustness of conclusions
  • Cross-Validation with Known Signals: Platform signal detection outputs were validated against 42 historically confirmed signals across the rare disease portfolio, achieving 97.6% sensitivity and 95.8% specificity
  • Expert Elicitation Protocol: Bayesian prior distributions were calibrated through structured expert elicitation sessions with Sanofi's rare disease medical directors and external key opinion leaders
  • Registry Data Quality Audits: Automated completeness, accuracy, and timeliness assessments for all registry data feeds, with quarterly reconciliation against source registries
  • Regulatory Alignment Review: All deliverables reviewed against EMA GVP Modules V, VII, and VIII; ICH E2C(R2) and E2E; and FDA rare disease guidance documents
  • Audit Trail: Complete data lineage with ORPHA code-to-MedDRA mapping transparency, compliant with FDA 21 CFR Part 11 and EU GMP Annex 11

Regulatory Context

This engagement was conducted in alignment with the following regulatory frameworks, with specific attention to rare disease considerations:

  • ICH E2C(R2): Periodic Benefit-Risk Evaluation Report format, with Bayesian extensions for small-population safety assessment
  • ICH E2E: Pharmacovigilance Planning with specific rare disease risk management provisions
  • EMA GVP Module V: Risk Management Plan structure and update requirements for orphan medicinal products
  • EMA GVP Module VII: PSUR requirements with guidance on signal detection in small populations
  • EMA GVP Module VIII: Post-authorization safety study (PASS) protocols using real-world evidence from rare disease registries
  • FDA Rare Disease Guidance: "Rare Diseases: Common Issues in Drug Development" -- addressing small sample statistical methods
  • CIOMS Working Group IV: Benefit-risk methodology adapted for orphan medicinal products
  • Outcome: All PBRERs accepted by EMA PRAC without major objections. EMA specifically cited the Bayesian signal detection methodology and real-world evidence integration as "methodologically sound and well-suited to the rare disease context."

Related Case Studies

Accelerating Oncology BRA with AI-Driven Signal Detection

AstraZeneca

68% reduction in BRA cycle time

Scaling Pharmacovigilance Operations with AI

ICON plc

45% operational efficiency gain

View All Case Studies

Explore More Success Stories

ArcaScience Evidence Hub

20+ enterprise case studies

Transform Your Rare Disease Pharmacovigilance

See how ArcaScience's Bayesian BRA models and real-world evidence integration can solve the small-population signal detection challenges inherent to rare disease pharmacovigilance. Talk to a scientist about your portfolio.

Talk to a Scientist Download PDF