clai

Clinical Augmented Intelligence Group (CLAI) | The Average Patient Fallacy clai.mgh.harvard.edu/research/n-of-…

Clinical Augmented Intelligence Group (CLAI) | Building a Digital Clinical Team clai.mgh.harvard.edu/research/agent…

Clinical Augmented Intelligence Group (CLAI) | Optimization Instability in Autonomous Agents clai.mgh.harvard.edu/research/agent…

This work shows diagnostic signal decay is tractable. SFI-aware calibration offers a practical, label-free strategy for deploying AI models across diverse healthcare contexts. Congrats to leads Federica Spoto & Jingya Cheng + the whole team! 🎉

The results? Dramatic improvements across 6 clinical phenotypes: • F1-score gains: +4.2% to +34.9% • AUC gains: +4.7% to +40.1% Even "well-diagnosed" conditions like diabetes & hypertension showed major benefits. Signal decay affects ALL EHR-based phenotyping.

🚨 NEW: Two CLAI studies published! We're tackling a fundamental problem in clinical AI: diagnostic signal decay, the hidden variability in how diseases are documented across healthcare systems. 🧵 Here's what we found & how we're fixing it:

new preprint! 🌟 An N-of-1 Artificial Intelligence Ecosystem for Precision Medicine A multi-agent AI framework that delivers personalized diagnostics, treatment planning, and real-time adaptation—for one patient at a time. 🧬🤖 doi.org/10.48550/arXiv… #PrecisionMedicine #AI #Nof1

(3/3) We believe this is essential for building safer, more equitable AI in healthcare. We're keen to hear feedback from clinicians, data scientists, and patients. ​ #PatientSafety #AIethics

The problem is that rare, life-threatening events get dismissed as statistical noise during training. ​we offer a solution: a framework to measure the "Rare Case Performance Gap" and force models to pay attention to what's clinically important, not just what's common.

(1/3) Excited to share CLAI's new preprint on a critical flaw in medical AI: the "Average Patient Fallacy." doi.org/10.48550/arXiv… ​#AIinMedicine #healthTech

Clinical Augmented Intelligence Group (CLAI) | The AI Black Box clai.mgh.harvard.edu/research/cot/

New study and AI tool from our group.

"#DigitalTwins in #healthcare: Where every heartbeat, breath, and cell has a virtual echo." - #ChatGPT

AI is well on its way of democratizing healthcare - Robin Farmanfarmaian @Robinff3

Will doctors have to see more patients with the time that ChatGPT saves them?

"We currently simply don't know what is the data in the NLP (i.e. ChatGPT) models..." - Issac Kohane, MD, PhD @zakkohane

"[doctors] are drowning in administrative tasks" - Rebecca Mishuris, MD, MS, MPH (Mass General Brigham)