📘 🤗🙂 Summary more of What I Learned at #ERA25 on AI in Nephrology @ERAkidney “CME 9 – Artificial Intelligence for Nephrologists | June 4, 2025, Vienna” European Renal Association 🧠 Foundational Concepts Speaker: Valentin Goutaudier Valentin Goutaudier 📍 Talk: “AI and Machine Learning Demystified” •When AI/ML is most useful: 1.Complex learning processes 2.Situations with large data volume 3.Heterogeneous diseases, such as CKD with multiple etiologies 4.Personalized prediction, e.g., risk stratification or therapy response •Benefits of ML: •Learns from nonlinear data patterns •Enables clinical decision support at scale •Works well when structured and unstructured data are integrated •Limitations: •Vulnerable to poor-quality or biased data •Limited external validity (generalizability) •Overfitting and “black-box” interpretability concerns ⸻ 🧪 AI Tools in Practice ✅ CKD Progression Prediction Speaker: Len Usvyat Len Usvyat Renal Research Institute (RRI) @RR_Institute 📍 Talk: “Exploring the Power of AI in Nephrology: Useful or Dangerous? (PRO)” •Tool: Klinrisk •Predicts CKD progression in patients with T2DM •Uses only routine lab data—no biomarkers or EHR needed •Outperforms KDIGO heatmap and KFRE in AUC at 2–4 years •AUC up to 0.86 at 4 years 🧬 Kidney Biopsy & Deep Learning •AI excels in histopathology: •87% of approved AI/ML tools are in radiology/pathology •Can detect subtle morphologic changes and predict graft failure •Ideal for automating glomerular lesion detection and quantification 🩸 Anemia Management in Dialysis •Tool: Anemia Control Model (ACM) •AI-powered decision support for ESA/iron dosing •Implemented in over 100 clinics since 2013 •Outcomes: •25% ESA reduction •12% hospitalization reduction •No safety signal in 12 years of real-world use 💻 LLM Integration into EMRs •EMR vendors now embedding LLMs (e.g., GPT-based tools) for: •Inbox triage •Drafting documentation (progress notes, prior auths) •Coding and billing •Patient-specific decision support between visits ⸻ ⚖️ Regulatory & Ethical Considerations 🔐 Speaker: Pierpaolo Maio Pierpaolo Maio 📍 Talk: “AI in Nephrology – Useful or Dangerous? (CON)” •Strict Compliance Required under the EU AI Act •AI systems in nephrology often qualify as “high-risk” •Must meet strict requirements: CE marking, ISO 13485, post-market surveillance •Privacy & Risk Stratification •Data classified by societal value vs. risk •Only high-value, low-risk use cases (Type A) are widely supported •Type C/D (low value or high risk) face restrictions •Governance Priorities •Continuous human oversight •Transparent algorithms •Legal risk shared across developers, institutions, clinicians •Regulatory Watchpoints •MDR requires clinical evaluation pre- and post-market •New Product Liability Directive (PLD) proposal expands liability to clinicians •Mandatory incident reporting for all serious AI-related errors #AI #Nephrology