Life science companies launch new pharmaceutical products on a regular basis. That doesn’t make the...
Posted by
Ryan Chapa |
This month’s developments offer a clear view of where AI is gaining real traction in healthcare and life sciences. Much of the most substantive activity came through ESMO’s AI & Digital Oncology Congress and affiliated publications, which now function as one of the most active testing grounds for AI standards, clinical validation, and governance. That concentration is itself a signal: oncology remains the domain where evidence, infrastructure, and regulatory engagement with AI are maturing fastest.
The signals highlight where AI has moved beyond experimentation and where strategic teams can start making informed updates to study design, diagnostic pathways, and market models. The focus throughout is on developments with clear implications for 2026 planning, access conversations, and scientific messaging.
Standards and Governance
ESMO introduces the first framework for AI-based biomarkers in oncology
ESMO’s introduction of the ESMO Basic Requirements for AI‑based Biomarkers in Oncology (EBAI) marks a meaningful attempt to bring structure to an area where expectations have been diffuse. For the first time, AI biomarkers, whether simple automation tools or novel predictive signatures, are described within a classification system that distinguishes their risk profiles and evidence needs. The framework outlines how AI systems should be benchmarked, validated across sites, and assessed for fairness, explainability, operational feasibility, and turnaround time. With its publication in Annals of Oncology, EBAI now sits in a position to influence both scientific and commercial narratives.
More importantly, EBAI provides a shared language for teams building or evaluating AI-derived signatures. It sets explicit expectations around what constitutes acceptable validation and what must be demonstrated to justify clinical or access claims. Although not a regulatory guideline, it is likely to be treated as a practical reference point by tumor boards, hospital adoption committees, and purchasers, particularly in systems where evidence hierarchies already resemble ESMO’s approach
Your takeaway: Treat EBAI as a de facto European playbook for evidence packages and product claims for AI assays. Align technical validation plans, clinical utility studies, and payer narratives to the relevant EBAI category. Build HTA education that explicitly references EBAI criteria. Expect hospital committees and national societies to use EBAI as a baseline when assessing vendor submissions.
Pathology and Diagnostic Performance
AI PD-L1 scoring improves accuracy near treatment thresholds
In a multicentre evaluation within ESMO Real World Data and Digital Oncology, the DiaKwant AI algorithm outperformed routine manual PD‑L1 scoring across 142 patients, delivering higher accuracy (88% vs 75%) and sensitivity (96% vs 78%). Gains were strongest around clinical cut‑offs that determine eligibility for checkpoint inhibitors. ESMO notes that AI‑assisted scoring can reduce inter‑observer variability and optimize patient selection.
The findings land at an important moment. As IO indications expand and treatment costs rise, misclassification at PD‑L1 thresholds becomes more consequential for both patients and systems. AI support in this setting is a way of reducing avoidable risk at a point where small interpretive differences carry major downstream impact. With evidence now published and accessible, adoption in high‑volume centres feels increasingly plausible.
Your takeaway: Build the case for reimbursable AI pathology as a companion to immunotherapy selection. Quantify misclassification reduction at decision thresholds and integrate into value stories for oncology formularies and KOL education. Expect high‑volume centers to move toward AI‑assisted biomarker workflows as standard practice.
NICE endorses five AI‑assisted colonoscopy tools for earlier detection
NICE’s recommendation of five AI technologies for real‑time colonoscopy support stands out as one of the clearest HTA-level endorsements of AI diagnostics to date. The evidence base shows that these tools improve adenoma detection without lengthening procedures, which is an important signal that they add value without disrupting an already capacity‑constrained service. By issuing broad approval, NICE has effectively signalled that AI‑supported colonoscopy is ready for integration into NHS pathways.
This endorsement changes expectations for vendors and system planners. Accuracy alone is no longer enough; cost‑effectiveness, workflow compatibility, and integration with service‑level KPIs will all play larger roles in adoption decisions. As detection rates improve, downstream models for incident cancer, staging at diagnosis, and treatment volumes will also require recalibration.
Your takeaway: HTA‑level backing raises the bar for evidence. Vendors must show cost‑effectiveness, not only accuracy. GI oncology brands and screening stakeholders should refresh assumptions on incident cancer rates, stage at diagnosis, and eligible pools in NHS settings and anticipate follow‑on adoption in other markets
AI‑powered three‑dimensional tumor pathology emerges as the next data layer
ESMO’s reporting on emerging AI‑driven 3D pathology highlights how digital pathology is beginning to move beyond two‑dimensional interpretation. By reconstructing spatial architecture and capturing features that are difficult to represent on standard H&E slides, these tools may provide richer prognostic and predictive insights. Although early, the work gestures toward a future in which tumour heterogeneity is quantified with greater granularity.
As digital pathology continues to expand, 3D data layers may become another differentiator for precision oncology. Developers of next‑generation biomarkers will need to demonstrate not only accuracy but also incremental value relative to established methods. If 3D features consistently improve predictive performance, they could influence risk stratification, trial design, and companion diagnostic pathways.
Your takeaway: Track 3D pathology as a potential contributor to next‑generation biomarker strategies. Its value will depend on comparative evidence, but early signs point to meaningful gains over traditional histology metrics.
Therapy Personalisation and De‑escalation
ArteraAI and NHS data suggest omission of radiotherapy for a prostate cancer subgroup
Data from an NHS cohort of roughly 4,000 men indicate that ArteraAI’s multimodal biomarker can identify a subgroup within intermediate‑risk prostate cancer whose outcomes remain similar whether they receive adjuvant radiotherapy or not. The study offers a real‑world perspective on how AI-derived risk segmentation may refine decisions that have historically been driven by broader clinical categories.
The implications extend beyond radiotherapy. If AI-driven subtyping becomes common, long‑standing risk categories may need to be re‑evaluated in trials, guidelines, and budget‑impact models. The ability to stratify risk more precisely could influence follow‑up imaging, the use of systemic therapy, and the economic assumptions that underpin care‑pathway planning.
Your takeaway: AI‑enabled risk segmentation is beginning to reshape expected demand for radiotherapy and related services. Modelling frameworks and study designs should prepare for future states where AI-defined risk groups replace current strata.
Risk Stratification from Routine Imaging
Clairity secures Series B to scale FDA‑authorised image‑only breast cancer risk platform
Clairity’s Series B funding round underscores the momentum behind imaging‑based population‑risk tools. Clairity Breast, the first FDA‑authorised image‑only model predicting five‑year breast cancer risk from routine mammograms, has shown strong performance across large internal and external datasets. The RSNA data indicate that the model distinguishes risk categories more sharply than breast density alone.
The introduction of Clairity Heart, which aims to estimate cardiovascular risk using the same mammograms, points toward a platform model in which a single imaging modality powers multiple disease‑risk assessments. This approach offers operational and economic advantages for health systems aiming to consolidate screening workflows and improve early identification.
Your takeaway: Imaging‑based multimorbidity scoring is emerging as a realistic pathway for large health systems. Screening funnels, epidemiology assumptions, and early‑intervention planning should begin incorporating the possibility of multi‑condition risk outputs from a unified imaging pipeline.
Enterprise AI and Discovery Partnerships
Merck KGaA expands Valo collaboration into neurology
Merck KGaA’s expansion of its relationship with Valo Health signals that AI-native discovery platforms are becoming embedded in core portfolio strategy rather than positioned as peripheral capability boosters. The addition of neurology to their existing cardiovascular and oncology work suggests increasing confidence in Valo’s Opal platform and its potential to advance programmes across diverse mechanisms.
The collaboration also illustrates how digital twin models, multimodal datasets, and algorithmic design tools are converging into integrated discovery ecosystems. The ability to iterate on hypotheses quickly and assess candidates computationally is beginning to shape expectations of speed, quality, and translatability.
Your takeaway: AI-first discovery partnerships are now shaping planning in complex disease areas. Expect heightened emphasis on demonstrating translatability, not just computational speed, when differentiating platform value.
Digital twins gain traction in Alzheimer’s trial design
A new paper in the Journal of Alzheimer’s Disease highlights how AI‑driven digital twin models can simulate disease trajectories and treatment responses. By using synthetic control arms and model‑informed enrichment, the authors demonstrate pathways to reduce sample sizes without sacrificing statistical power—an appealing proposition in conditions where recruitment is challenging or costs are high.
Regulatory commentary within the paper indicates that the FDA and EMA have already accepted these tools in specific contexts, especially where conventional trials face feasibility constraints. As the evidence base grows, digital‑twin methods may become a standard complement to traditional trial designs rather than an experimental alternative.
Your takeaway: Digital‑twin approaches are maturing into practical design tools for neurodegeneration studies. Teams should begin defining governance, vendor strategies, and internal guidelines for when these methods are appropriate.
Where to Focus Next
Oncology decision-making: Benchmark every AI-enabled biomarker or imaging signature against the expectations outlined in EBAI. Align validation packages, clinical-utility evidence, and messaging to the framework’s categories so adoption committees and payers can understand claims through a structure they already recognise.
Pathology and detection workflows: Update funnel projections and treatment-eligibility assumptions based on AI-supported scoring in PD-L1 and enhanced adenoma detection in colonoscopy. As variation decreases and detection improves, downstream volumes and staging distributions will shift.
Risk stratification and screening: Plan for imaging-derived risk platforms that identify high-risk groups earlier and more consistently. Screening programmes, epidemiology models, and early-intervention strategies will need to account for multimorbidity risk scoring emerging from a single imaging pipeline.
Trial design and evidence planning: For neurodegeneration and other hard-to-recruit diseases, incorporate digital-twin simulation into feasibility assessments and ask where model-informed enrichment can reduce costs and complexity. Prepare internal guardrails for when these methods are appropriate and how they should be validated.
Cardiovascular pathway planning: Adjust service-planning assumptions as AI-ECG tools bring diagnoses forward, particularly in hypertrophic cardiomyopathy. Expect changes in referral mix, diagnostic workload, and genetic counselling requirements as earlier identification becomes commonplace.
Imaging procurement and platform strategy: Anticipate a shift from single-module AI tools to platform-level contracts spanning multiple pathways. ROI assessments should reflect cross-service impact rather than the performance of isolated modules.
November’s developments point toward a landscape where AI can be evaluated, benchmarked, and adopted using familiar standards. Although oncology remains the most advanced, radiology, cardiology, and neurodegeneration are showing similar momentum, with real-world performance, maturing governance, and platform-level deployment now shaping expectations.
The task for 2026 is to translate these signals into updated assumptions across trial design, diagnostic pathways, access strategies, and procurement models. Organisations that refine their plans early will be able to shape adoption rather than respond to it.
The opportunity now is to act while the benchmarks are still forming. Choose one assumption or decision that this month’s evidence challenges and refine it ahead of next-quarter planning.
Book a 30-minute consultative call to discuss how to translate these market signals into action for your 2026 roadmap.
Sources
ESMO introduces the first framework for AI-based biomarkers in oncology (EBAI): https://dailyreporter.esmo.org/spotlight/the-first-esmo-guidance-for-ai-based-biomarkers?utm_source=chatgpt.com
AI PD-L1 scoring improves accuracy near treatment thresholds: https://dailyreporter.esmo.org/news/ai-tool-outperforms-manual-pd-l1-scoring-in-a-study
Federated training takes collaboration beyond data sharing: https://www.businesswire.com/news/home/20251028507233/en/OpenFold-Consortium-Releases-Preview-of-OpenFold3-An-Open-Source-Foundation-Model-for-Structure-Prediction-of-Proteins-Nucleic-Acids-and-Drugs
NICE endorses five AI‑assisted colonoscopy tools for earlier detection: https://www.nice.org.uk/news/articles/new-ai-tools-could-help-save-lives-by-spotting-warning-signs-of-bowel-cancer-earlier
AI‑powered three‑dimensional tumor pathology emerges as the next data layer: https://dailyreporter.esmo.org/opinions/ai-powered-three-dimensional-tumour-pathology
ArteraAI and NHS data suggest omission of radiotherapy for a prostate cancer subgroup: https://www.theguardian.com/society/2025/nov/03/nhs-hospitals-to-test-ai-tool-that-helps-diagnose-and-treat-prostate-cancer
Merck KGaA expands Valo collaboration into neurology: https://www.businesswire.com/news/home/20251119408649/en/Valo-Health-Announces-Collaboration-with-Merck-KGaA-Darmstadt-Germany-to-Discover-and-Develop-Novel-Treatments-for-Parkinsons-Disease-and-Related-Disorders
Digital twins gain traction in Alzheimer’s trial design: https://pmc.ncbi.nlm.nih.gov/articles/PMC12639399/
