Altis’ IPRO-Δ AI model outperforms RECIST in predicting survival, signaling a major shift in oncology trial endpoint
- Altis Labs’ AI model IPRO-Δ, in collaboration with Bayer, improves survival prediction accuracy by 50% over RECIST in Phase 3 aNSCLC trial.
- Backed by 200M+ imaging data points and 500,000+ patient-years, IPRO-Δ enables earlier and more precise outcome forecasting.
- ASCO 2025 spotlights AI-powered clinical endpoints as the future of global oncology drug development.
Altis-Bayer Collaboration Demonstrates Superior Predictive Accuracy
At ASCO 2025, Altis Labs and Bayer presented data from the Phase 3 NExUS trial showing that IPRO-Δ, Altis’ AI-derived imaging biomarker, predicted overall survival (OS) with a hazard ratio (HR) of 1.72—versus 1.14 for traditional RECIST measurements. The 50% improvement positions IPRO-Δ as a potential industry standard for early efficacy assessment in oncology trials.
AI Enables Earlier, Data-Driven Trial Decisions
IPRO-Δ analyzed baseline and week-13 CT scans from 165 advanced NSCLC patients, demonstrating that AI can stratify patients more accurately and earlier than standard radiologic endpoints. This gives trial sponsors like Bayer the power to make faster, more confident decisions on trial progression and drug value.
Trained on 200 Million Images and 500K+ Patient Years
Altis’ proprietary multimodal database—spanning radiologic, clinical, genomic, and treatment data—empowers IPRO-Δ to deliver unmatched predictive insight. Its scale and diversity enable consistent performance across trial populations and cancer types, making it a scalable AI solution for global pharma.
Bayer Sets New Precedent for AI in Phase 3 Oncology Trials
Bayer’s integration of IPRO-Δ into the pivotal NExUS study showcases pharma’s growing confidence in AI for de-risking late-stage trials. This collaboration is more than a technical enhancement—it’s a strategic shift toward intelligent trial design, better patient targeting, and accelerated drug development timelines.
For more, visit www.altislabs.com