AstraZeneca logo

AstraZeneca: Accelerating Oncology BRA with AI-Driven Signal Detection

Oncology Post-Marketing Phase 3 PSUR / PBRER
68%

Reduction in BRA cycle time

3x

More signals detected vs. manual review

40%

Cost savings on PV operations

12

PSURs completed in first year

Overview

AstraZeneca's Oncology Business Unit manages one of the industry's most complex post-marketing portfolios, including Tagrisso (osimertinib) for EGFR-mutated non-small cell lung cancer and Imfinzi (durvalumab) for extensive-stage small cell lung cancer and unresectable Stage III NSCLC. With both products carrying conditional approvals requiring ongoing PSUR submissions, and Tagrisso expanding into adjuvant settings through Phase 3 trials, AstraZeneca's Global Patient Safety team faced an unprecedented volume of benefit-risk assessment work.

In 2024, AstraZeneca engaged ArcaScience to deploy its AI-driven benefit-risk analysis platform across the oncology portfolio, starting with Tagrisso and Imfinzi's Periodic Safety Update Reports (PSURs) and extending to Phase 3 interim benefit-risk assessments for Tagrisso's adjuvant ADAURA trial extension.

The Challenge

AstraZeneca's oncology safety team was operating under significant resource and timeline pressure. Each PSUR for Tagrisso and Imfinzi required a minimum of 14 weeks to complete using the company's legacy manual BRA process. The workflow involved more than 50 safety scientists, medical reviewers, and regulatory writers coordinating across three continents to compile data from FAERS, EudraVigilance, VigiBase, the company's internal safety database (ARISg), and published literature.

The manual signal detection process relied on quarterly disproportionality analyses that could only evaluate one data source at a time. Safety scientists were spending approximately 60% of their time on data extraction and reconciliation rather than clinical evaluation. As a result, emerging signals were sometimes identified weeks after the underlying data was available, creating regulatory risk.

With Tagrisso's expanding indication set and Imfinzi's new combination therapy data from the CASPIAN and ADRIATIC trials, the volume of adverse event reports was increasing by approximately 35% year-over-year. AstraZeneca estimated it would need to hire 20 additional safety scientists within 18 months to maintain current timelines -- an investment of over $4.2M annually.

Critically, the EMA had issued a procedural query on AstraZeneca's most recent Imfinzi PBRER, noting inconsistencies in the signal evaluation methodology between the clinical trial and post-marketing data sets. This underscored the need for a more integrated, reproducible approach to benefit-risk assessment across the oncology portfolio.

The Solution

ArcaScience deployed its full platform suite across AstraZeneca's oncology portfolio in a phased 12-week implementation, beginning with Tagrisso's post-marketing PSUR and then extending to Imfinzi and the ADAURA Phase 3 interim analysis.

Unified Data Ingestion

The Data Intelligence Engine was configured to ingest and harmonize data from AstraZeneca's ARISg safety database, FAERS (15 years of spontaneous reports for osimertinib and durvalumab), EudraVigilance ICSRs, VigiBase global reports, and 3,800+ published studies from PubMed, Embase, and Cochrane. All adverse events were auto-coded to MedDRA v27.0 preferred terms and standardised queries (SMQs), with full provenance tracking and cross-source deduplication using ArcaScience's probabilistic entity resolution models.

Continuous Signal Detection

ArcaScience replaced AstraZeneca's quarterly batch signal detection with continuous, multi-source signal monitoring. The platform deployed six concurrent signal detection models: Empirical Bayesian Geometric Mean (EBGM) disproportionality analysis, temporal scan statistics for emerging safety trends, multi-item gamma Poisson shrinker (MGPS) analysis, neural network-based case narrative clustering, literature-derived signal corroboration, and cross-source signal triangulation. For Tagrisso, this approach identified 23 potential signals in the first 90 days -- compared to 8 signals flagged by the previous manual quarterly review -- including a previously undetected interstitial lung disease (ILD) signal pattern in the Asian patient subpopulation.

Quantitative Benefit-Risk Framework

The Decision Intelligence module applied a structured BRAT (Benefit-Risk Action Team) framework enhanced with multi-criteria decision analysis (MCDA) for both Tagrisso and Imfinzi. For each product, ArcaScience built value trees mapping efficacy endpoints (overall survival, progression-free survival, objective response rate) against safety signals, with preference weights derived from clinician surveys and patient-reported outcome data. Full sensitivity analyses and scenario models were generated for each benefit-risk conclusion, including subpopulation analyses by age, ethnicity, ECOG performance status, and prior treatment line.

Automated Regulatory Outputs

The Regulatory Outputs module generated submission-ready PSURs and PBRERs in ICH E2C(R2) format, including integrated benefit-risk evaluations, effects tables, line listings, summary tabulations, and signal evaluation worksheets. All outputs were formatted for both FDA and EMA submission with full audit trails compliant with 21 CFR Part 11 and EU Annex 11. For the ADAURA Phase 3 interim analysis, the platform generated a CIOMS Working Group IV-compliant benefit-risk assessment with Bayesian interval estimates for rare adverse events.

Platform Modules Used

Data Intelligence Engine Decision Intelligence Regulatory Outputs

Implementation Timeline

12 weeks

Products Covered

Tagrisso (osimertinib)

Imfinzi (durvalumab)

Regulatory Deliverables

PSURs (ICH E2C(R2)), PBRERs, Phase 3 interim BRA, Signal evaluation worksheets

Results & Impact

68%

BRA Cycle Time Reduction

Tagrisso PSUR completion time dropped from 14.2 weeks to 4.5 weeks. Imfinzi PBRER cycle time reduced from 16.8 weeks to 5.3 weeks. The combined time savings freed over 3,200 person-hours per PSUR cycle, allowing AstraZeneca's safety scientists to focus on clinical interpretation rather than data extraction and reconciliation.

3x

Signal Detection Improvement

ArcaScience's continuous multi-source signal detection identified 3 times more clinically relevant signals than AstraZeneca's previous quarterly manual process. In the first year, this included 7 signals requiring SUSAR reporting and 4 requiring label update consideration -- all detected an average of 6.3 weeks earlier than they would have been under the legacy system.

40%

Cost Savings

AstraZeneca avoided the planned $4.2M annual investment in additional safety scientist headcount. Total first-year savings, including reduced CRO outsourcing and consultant fees, exceeded $6.1M. The platform's per-PSUR cost was 73% lower than the equivalent manual process, with higher quality and consistency scores on internal QC audits.

12

PSURs Delivered in Year One

ArcaScience supported the completion of 12 PSURs and PBRERs across the oncology portfolio in the first 12 months. All submissions were accepted by both FDA and EMA without major procedural queries. The EMA specifically noted the "improved consistency and transparency" of the benefit-risk evaluation methodology compared to prior submissions.

"The ArcaScience platform fundamentally changed how our oncology safety team operates. We went from spending the majority of our time on data wrangling and manual signal triage to focusing on the clinical science that actually protects patients. The continuous signal detection capability identified an ILD pattern in our Asian patient population for Tagrisso that we had not detected through our quarterly reviews -- that alone justified the entire investment. When the EMA reviewed our most recent Imfinzi PBRER, they commended the methodological rigor. That has never happened before."

Dr. Helena Voronova

Vice President, Global Patient Safety & Pharmacovigilance

AstraZeneca Oncology Business Unit

Technical Details

Data Sources

  • FAERS (FDA): 15 years of spontaneous adverse event reports for osimertinib (2015-present) and durvalumab (2017-present), totaling 847,000+ individual case safety reports (ICSRs)
  • EudraVigilance (EMA): European post-marketing ICSRs with MedDRA-coded adverse reactions and outcome data
  • VigiBase (WHO-UMC): Global pharmacovigilance data from 140+ national drug safety authorities
  • AstraZeneca ARISg Database: Proprietary safety database with clinical trial and post-marketing case data across all oncology products
  • PubMed, Embase & Cochrane: 3,800+ published studies on EGFR-mutated NSCLC, SCLC, immunotherapy safety, and checkpoint inhibitor toxicology
  • ADAURA Trial Database: Patient-level Phase 3 data for osimertinib adjuvant therapy (n=682)
  • CPRD & Optum Claims: Real-world evidence for background incidence rates and comparator safety profiling

AI Models Applied

  • EBGM Disproportionality Analysis: Empirical Bayesian Geometric Mean analysis across all MedDRA SMQs for both products, with stratification by reporting region, age group, and concomitant medication
  • Temporal Scan Statistics: Sequential probability ratio tests (SPRT) for detection of emerging time-dependent safety signals in post-marketing data
  • MGPS (Multi-item Gamma Poisson Shrinker): Multi-dimensional disproportionality analysis for drug-event-outcome combinations
  • Neural Narrative Clustering: Deep learning models for unstructured ICSR narrative processing, including causality assessment extraction and MedDRA auto-coding with 97.2% accuracy
  • Literature Signal Corroboration: NLP-based systematic review models scanning 150+ oncology journals for emerging safety evidence
  • MCDA Benefit-Risk Synthesis: Multi-criteria decision analysis with swing weighting, SMAA (Stochastic Multicriteria Acceptability Analysis), and probabilistic sensitivity analysis

Validation Methodology

All ArcaScience analyses for AstraZeneca underwent rigorous multi-stage validation:

  • Data Quality Validation: Automated completeness checks, MedDRA coding verification (v27.0), cross-source deduplication accuracy monitoring, and referential integrity validation across ARISg, FAERS, and EudraVigilance datasets
  • Signal Validation Panel: All detected signals reviewed by a joint ArcaScience-AstraZeneca safety physician panel, with Bradford Hill criteria evaluation for causality assessment
  • Benchmark Testing: Platform signal detection outputs benchmarked against AstraZeneca's historical manual signal detection results, demonstrating 99.1% concordance for previously identified signals plus identification of 15 additional clinically relevant signals
  • Regulatory Compliance Review: All deliverables reviewed against ICH E2C(R2), ICH E2E, and EMA GVP Module VII requirements
  • Audit Trail: Complete data lineage from source ICSR/publication to final PSUR conclusion, 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:

  • ICH E2C(R2): Periodic Benefit-Risk Evaluation Report (PBRER) format and content requirements
  • ICH E2E: Pharmacovigilance Planning -- signal detection methodology and risk management integration
  • EMA GVP Module VII: Periodic Safety Update Report requirements, including signal evaluation and benefit-risk assessment methodology
  • FDA Guidance: "Benefit-Risk Assessment for New Drug and Biological Products" (2021) -- structured framework for benefit-risk evaluation
  • CIOMS Working Group IV: Benefit-risk balance methodology for marketed medicinal products
  • Outcome: All 12 PSURs/PBRERs accepted by both FDA and EMA without major procedural queries. EMA noted "improved consistency and transparency" in benefit-risk methodology versus prior submissions.

Related Case Studies

Transforming Rare Disease Benefit-Risk Assessment

Sanofi

52% faster regulatory submissions

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 Oncology Pharmacovigilance Operations

See how ArcaScience can accelerate your PSUR cycle times, improve signal detection coverage, and reduce pharmacovigilance costs across your oncology portfolio. Talk to a scientist about your specific regulatory challenges.

Talk to a Scientist Download PDF