AI for Patient Risk Stratification

𝗔𝗜 𝗖𝗮𝗻 𝗡𝗼𝘄 𝗣𝗿𝗲𝗱𝗶𝗰𝘁 𝗛𝗲𝗮𝗿𝘁 𝗙𝗮𝗶𝗹𝘂𝗿𝗲 𝗙𝗶𝘃𝗲 𝗬𝗲𝗮𝗿𝘀 𝗕𝗲𝗳𝗼𝗿𝗲 𝗜𝘁 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝘀 A fascinating paper published this week in the The American Journal of Cardiology. The important question it asked was the following. Can we predict heart failure before the heart has already begun to fail? The answer, it now appears, is yes. A team led by Prof Charalambos Antoniades MD PhD FRCP FMedSci at the University of Oxford has developed an AI tool that analyses the fat surrounding the heart from routine cardiac CT scans, predicting a patient's risk of developing heart failure up to five years before any clinical signs appear. Epicardial adipose tissue (EAT) is a metabolically active visceral fat depot that is both a sensor and a modulator of myocardial biology and changes its composition in response to paracrine signals from the myocardium. The team hypothesised that radiomic characterization of EAT from routine coronary computed tomographic angiography (CCTA) can noninvasively capture this adverse remodeling and enable early heart failure (HF) risk stratification. The study involved over 72,000 patients across nine NHS centres, followed for up to a decade. The fat around the heart, it turns out, acts as a potential biological sensor. Patients in the highest risk group were twenty times more likely to develop heart failure than those in the lowest. The tool predicted five-year risk with 86% accuracy, outperforming models built on traditional risk factors alone. What is striking is the conceptual shift this represents. We have spent decades in cardiovascular medicine treating disease that has already declared itself, responding to symptoms, managing complications, optimising a heart already under strain. We have been using risk stratification of cardiac disease using various methods like calcium scores. The team are now seeking NHS regulatory approval and adapting the tool for any CT scan of the chest, not just cardiac ones. Every scan, for any reason, could soon carry an embedded layer of cardiac risk intelligence. As the NHS shifts into prevention as part of the long term plan these tools become more important.

An explainable AI model trained on 15,726 patients across 38 cancer types outperforms clinical staging and scoring systems while generating individual, marker-level risk profiles for each patient. 1️⃣ The AI significantly outperformed ECOG performance status, TNM staging, Charlson Comorbidity Index, and modified Glasgow Prognostic Score on survival prediction. 2️⃣ It processed 350 variables per patient: blood tests, urine tests, CT-derived body composition, tumour genetics, clinical records, and treatment history. 3️⃣ Training across all cancer types simultaneously beat cancer-specific models, because shared prognostic signals transfer across disease types. 4️⃣ Explainable AI assigned each variable a personalised risk contribution for every individual patient, not just a population average. 5️⃣ 90% of the model's decision-making came from just 114 of 350 variables, identifying which markers actually drive prognosis. 6️⃣ The five most important prognostic markers were C-reactive protein, free thyroid hormone (triiodothyronine), functional performance status, metastatic spread, and lactate dehydrogenase. 7️⃣ The same C-reactive protein level carried different risk depending on platelet count, meaning identical lab values can mean very different things in different patients. 8️⃣ A "clinician's guide" output presents each patient's markers ranked by risk contribution, designed for use in tumour board discussions. 9️⃣ Tumour grade and primary tumour extent showed near-zero correlation between their values and their actual risk impact, suggesting they need broader clinical context to be meaningful. 🔟 Findings were validated in 3,288 lung cancer patients from a US electronic health record database, with consistent results across internal and external datasets. ✍🏻 Julius Keyl, Philipp Keyl, Grégoire Montavon, Dr. René Hosch, Alexander Brehmer, Liliana Mochmann, Philipp Jurmeister, Gabriel Dernbach, Moon Kim, Sven Koitka, Sebastian Bauer, Nikolaos Bechrakis, Michael Forsting, Dagmar Führer-Sakel, Martin Glas, Prof. Dr., Viktor Grünwald, Boris Hadaschik, Prof. Dr. Johannes Haubold, Ken Herrmann, Stefan Kasper, Rainer Kimmig, Stephan Lang, Tienush Rassaf, Alexander Roesch, Dirk Schadendorf, Prof. Dr. Jens Siveke, Martin Stuschke, Ulrich Sure, Prof. Dr. Matthias Totzeck, Anja Welt, Marcel Wiesweg, Hideo Baba, Felix Nensa, Jan Egger, Klaus-Robert Müller, Martin Schuler, Klauschen Frederick, Jens Kleesiek. Decoding pan-cancer treatment outcomes using multimodal real-world data and explainable artificial intelligence. Nature Cancer. 2025. DOI: 10.1038/s43018-024-00891-1 | Open Access

AI could help identify high-risk heart patients Artificial intelligence is stepping up in healthcare, particularly in identifying patients at risk of serious heart conditions. A team at the University of Leeds has developed an AI system called Optimise, which examined the health records of over two million individuals to find those most vulnerable to conditions like heart failure, stroke, and diabetes. The findings were significant: more than 400,000 people were identified as high-risk, representing a staggering 74% of patients who later died from heart-related issues. In a pilot study of 82 high-risk patients, Optimise revealed that 20% had undiagnosed moderate to high-risk chronic kidney disease. Additionally, over half of those with high blood pressure were prescribed different medication to better manage their heart risk. Dr. Ramesh Nadarajah from the University of Leeds highlighted the potential of AI in offering timely care, which is often more cost-effective than treating advanced conditions. This proactive approach not only benefits patients but also helps ease the burden on healthcare systems like the NHS. The research team's next step is to conduct a larger clinical trial to assess the full impact of doctor-led care supported by AI insights. The British Heart Foundation's Chief Scientific and Medical Officer, Prof. Bryan Williams, emphasized the importance of early diagnosis, noting that a quarter of all deaths in the UK are due to heart and circulatory diseases. By harnessing AI, this study opens new avenues for detecting and managing these life-threatening conditions, offering hope for improved patient outcomes and reduced hospital admissions. Now they plan to carry out a larger clinical trial to prove the AI's worth and efficiency.

🚨 AI JUST HIT ROCHE’S EARNINGS CALL 🚨 Roche’s Q3 2025 earnings call quietly revealed something bigger than a quarterly update — it showed where diagnostics is heading. They announced the Kidney Klinrisk Algorithm — an AI-driven risk stratification tool that just received its CE mark in Europe. This isn’t just a new test. It’s the start of a new category of diagnostics — where routine lab results, imaging, and patient data combine to predict risk before symptoms even appear. “By combining AI with routine tests, Roche helps physicians identify patients at risk of kidney function decline early on, enabling more informed and confident decision-making.” 💡 The signal beneath the noise: ✅ AI + Multi-Modal Data — Fusing clinical, biomarker, imaging, and real-world evidence to find patterns humans can’t see. ✅ Biomarker-Driven Precision — Identifying patient subgroups that respond differently, turning reactive testing into proactive insight. ✅ Data Governance & Traceability — Building regulated, audit-ready data environments to support CE-marked and FDA-cleared algorithms. ✅ Speed to Insight — Automating model development pipelines so clinicians don’t wait months for answers that data could reveal in days. For an industry where Diagnostics has been the slowest to digitize, this marks a real inflection point: from test results ➜ to algorithms ➜ to earlier, smarter interventions. Roche may have lit the spark — but the opportunity runs across the entire ecosystem. The companies who can unify multi-omics, imaging, and clinical data under a compliant, AI-ready framework will define the next era of precision medicine.

For the first time, a machine can look at a routine mammogram and predict a woman's 5-year breast cancer risk quite accurately: no history, no demographics, just the image itself revealing invisible patterns. As someone working inside the pharmaceutical ecosystem, this is interesting and exciting. This is the kind of shift that makes you rethink everything you assumed about early detection. Until now, doctors estimated risk using age, family history, and breast density. Useful signals, but broad, indirect, and often late. We've been predicting population risk far better than individual risk. That is now changing. An image-only AI model can predict 5-year breast cancer risk more accurately than breast density alone. The image reveals patterns the human eye cannot see. This is AI potentially adding healthy years to people's lives. Years without chemotherapy, without surgery, without the fear of late discovery. We're starting to see risk while the body still looks normal. From an industry lens, this changes three fundamentals: •⁠ ⁠Screening shifts from age-based to risk-based. Screening frequency and prevention strategies can now be customized based on actual biological risk, not age brackets. •⁠ ⁠Prevention becomes earlier and more precise, but raises hard questions about false positives, patient anxiety, and long-term follow-up. •⁠ ⁠We face a scale challenge. AI can identify risk at the population level. Healthcare systems must be ready to act without overwhelming clinicians or excluding low-resource settings. The technology is moving faster than our operating models. The real leadership test is no longer whether AI can predict risk. It's whether we can deploy it responsibly, equitably, and at scale without creating gaps in care. The future of oncology will not be defined only by better drugs. It will be defined by how early we dare to see risk and how wisely we choose to act on it. #AI #healthtech

The headline that caught my eye this week was "AI Trial to Spot Heart Condition Before Symptoms." Here's my take: Artificial intelligence holds substantial promise to improve quality and reduce costs in healthcare. One example from Leeds involves an algorithm that scours medical records for early warning signs of atrial fibrillation (AF) before symptoms appear — potentially preventing thousands of strokes. The results suggest that by analyzing existing medical records for patterns that human physicians might miss, AI can flag high-risk patients for early intervention. The trial has already identified cases like a 74-year-old former Army captain who had no symptoms but can now manage his condition effectively. This is particularly significant given that AF contributes to around 20,000 strokes annually in the UK alone. As Professor Chris Gale notes, too often the first sign of undiagnosed AF is a stroke — an outcome this technology could help prevent. The broader implication here is about AI's role in healthcare: not replacing physicians but augmenting their ability to identify risks earlier and intervene before conditions become critical.  

Case Tuesday: Cardiac CT A patient presents with chest pain. The question is urgent: is this a heart attack waiting to happen, or something else? A CT coronary angiogram is performed. For the radiologist, this means carefully assessing coronary arteries, looking for stenosis, calcifications, and subtle plaques. The challenge: Coronary CTs generate hundreds of slices, often complex to interpret. Subtle plaques can be easily overlooked. Quantifying calcium scores and stenosis consistently takes significant time. This is where #AI is showing real promise: Automated calcium scoring to assess cardiovascular risk Plaque detection and quantification to support precise diagnosis Tools that standardize reporting and improve communication with cardiologists The radiologist’s expertise is essential in interpretation and clinical context but AI ensures that the assessment is faster, more reproducible, and more actionable. The impact: Earlier detection of coronary artery disease. Better risk stratification for patients with chest pain. Closer collaboration between radiology and cardiology teams As Chief Medical Officer at GE HealthCare, I see cardiac CT as a shining example of how AI doesn’t just enhance workflows it helps us move toward preventive, precision medicine that saves lives before catastrophe strikes. Do you see AI as the tipping point that will make cardiac CT more widely adopted as a first-line test for chest pain? #CaseTuesday #CardiacCT #AIinHealthcare #Radiology #HeartHealth #GEHealthcare

Heart failure remains one of the biggest drivers of potentially avoidable hospital admissions in the United States and it continues to create major pressure on both patients and value based care systems. A powerful example of this came from a 12 month review across six acute care hospitals, where 3,233 emergency department visits for heart failure were tracked. Nearly 89% of those visits resulted in inpatient admissions, and an overwhelming majority of those admissions were considered potentially avoidable under the CMS PQI 08 quality indicator. The financial impact was massive, totaling over $27.6M in potentially avoidable costs of care. What is most encouraging is how much improvement was possible with smarter triage workflows. By updating heart failure triaging algorithms, especially for patients presenting with shortness of breath or lower extremity edema, over 92% of triaged heart failure patients were managed in an ambulatory setting, and nearly 85% avoided an ED visit within 24 hours. Within one year, the system nearly reached its cost reduction target with: • 11.2% reduction in potentially avoidable admissions • $3.35M reduction in total costs of care This is the kind of progress that matters because it protects capacity, improves outcomes, and keeps patients out of the hospital when they do not need to be there. The bigger opportunity now is how we scale this with AI. AI can help strengthen cardiovascular care by supporting earlier risk detection, improving decision support at the point of triage, and identifying which heart failure exacerbations can be safely managed outside the ED. When paired with clinical judgement, the goal is not to replace clinicians, it is to reduce uncertainty, standardize escalation pathways, and prevent avoidable deterioration. The future of heart failure care is proactive, data driven, and patient centered. AI can help get us there faster, but only if we build it around trust, safety, and real clinical workflows. Learn more: nej.md/48VcXDB Explore the full issue: nej.md/4j0CAHN Follow Zain Khalpey, MD, PhD, FACS for more on Ai & Healthcare. #HeartFailure #Cardiology #CardiovascularHealth #AIinHealthcare #DigitalHealth #ValueBasedCare #PopulationHealth #ClinicalInnovation #HealthcareInnovation #CareDelivery #PreventableAdmissions #HospitalAtHome #PreventiveCare #RemoteMonitoring #ClinicalDecisionSupport #HealthTech #MedTech #QualityImprovement #LeanHealthcare #PatientOutcomes

This is exactly the kind of thing we should be using AI for. AI can now predict breast cancer risk up to five years before a diagnosis by analyzing subtle tissue asymmetries. Researchers at Duke University have unveiled AsymMirai, a streamlined deep-learning algorithm designed to predict breast cancer up to five years before a formal diagnosis. Unlike previous complex models, this system focuses on "asymmetry"—the subtle differences between left and right breast tissue visible in mammograms. By isolating these specific structural variations, the AI achieves high-level accuracy while remaining significantly easier for radiologists to interpret. This shift highlights a critical movement in medical technology: using artificial intelligence not for content generation, but for identifying life-saving patterns that the human eye might overlook. The study, which analyzed more than 210,000 mammograms, confirms that breast asymmetry is a powerful yet underutilized biomarker for long-term health risks. Lead researcher Jon Donnelly notes that the simplicity and reliability of AsymMirai could revolutionize public health strategies by tailoring mammogram schedules to individual risk profiles. By providing a clear five-year window for intervention, this technology offers a proactive approach to oncology, potentially catching cancer in its earliest, most treatable stages and shifting the focus from reactive treatment to long-term prevention. source: Donnelly, J. J., et al. Asymmetry is all you need: Deep learning for predicting breast cancer risk. Radiology.

𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 𝗖𝗼𝘂𝗹𝗱 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺 𝗖𝗹𝗶𝗻𝗶𝗰𝗮𝗹 𝗗𝗲𝗰𝗶𝘀𝗶𝗼𝗻 𝗦𝘂𝗽𝗽𝗼𝗿𝘁. 𝗛𝗲𝗿𝗲'𝘀 𝗪𝗵𝗮𝘁 𝗧𝗵𝗮𝘁 𝗠𝗶𝗴𝗵𝘁 𝗟𝗼𝗼𝗸 𝗟𝗶𝗸𝗲. This week, OpenAI released a visual tool for building multi-agent workflows(1). I've been curious about agents for a while, but never had time to learn. Playing with this tool got me thinking about a longstanding challenge: how do we translate complex clinical pathways into effective CDS tools? 𝗧𝗵𝗲 𝗣𝗿𝗼𝗯𝗹𝗲𝗺 𝘄𝗶𝘁𝗵 𝗖𝘂𝗿𝗿𝗲𝗻𝘁 𝗖𝗗𝗦 Today's EHR-based decision support faces a fundamental tradeoff. Tightly scripted rules are reliable but inflexible. Loosely scripted ones give clinicians room to adapt but sacrifice consistency. Multi-agent AI workflows might offer a way out of this bind. 𝗔 𝗣𝗿𝗼𝗼𝗳 𝗼𝗳 𝗖𝗼𝗻𝗰𝗲𝗽𝘁 To explore this, I translated Children's Hospital of Philadelphia's Suicide Risk Assessment Pathway (2) into a multi-agent workflow. Here's how it works: • Input: Clinician's risk formulation, screening results, risk and protective factors • Acuity script: Determines patient acuity level • Intervention agent: Recommends response level • Response agents: Four specialized agents provide tailored clinical guidance based on severity 𝗔 𝗗𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗞𝗶𝗻𝗱 𝗼𝗳 𝗖𝗗𝗦 Now imagine this in your EHR. Instead of rigid decision trees, you'd have multiple specialized LLMs that can: • Access relevant patient data • Provide contextual guidance • Answer follow-up questions in real time • Adapt to clinical nuance while maintaining evidence-based standards Some extras that make this promising are the ability to use MCP and RAG! This isn't just automation. It's augmentation that preserves clinical judgment while providing robust support.   Bimal Desai MD, MBI, FAAP, FAMIA