💻 AI Transforms Medicine - Predicts Drug Effects, Summarizes Patient Records, Forecasts Cardiac Health, and More 📈
Weekly Updates on AI in Medicine 🤖🩺🚀
"A perfect storm for innovation usually takes place when there is urgency to do it, policy that suggests we should do it and industry saying it’s the right thing to do. And that’s exactly what is happening right now. It’s a top of mind issue." - Dr. John Halamka, President of Mayo Clinic Platform
Dear Med AI Enthusiast,
Welcome to The ‘Med AI’ Capsule weekly newsletter, your inside look at how AI is transforming medicine.⚕️🤖💊
In this fast-paced field of artificial intelligence in medicine, I'll be your guide to the most groundbreaking news updates and thought-provoking ideas.
Whether you're a medical pro, tech enthusiast, or curious mind, The ‘Med AI’ Capsule delivers an insider's perspective on the exciting possibilities and cautionary areas for AI in medicine.
Let the adventure begin! I'm thrilled to have you along for the ride. 🚀
In today’s capsule:
3 News Updates
2 Research Updates
1 Upcoming Event, and more…
Reading Time: 5 minutes
Med AI News Updates 📰
1. 🧬 AI Predicts GPCR Responses to Drugs 💊
Why Important: The AI algorithm can predict with over 80% accuracy how GPCRs (G protein–coupled receptors) will respond to drug-like molecules. This could greatly improve drug development targeting this key protein family.
“Our ultimate goal is to be able to predict how individual variants that people carry respond to drugs, allowing for the custom tailoring of prescriptions and paving the way for precision medicine.” - Kirill Martemyanov, Neuroscientist
Caution: The algorithm's accuracy still needs improvement. Further research is required to validate predictions across more genetic variants and determine real-world clinical utility.
2. 💻 Japanese Doctors Adopt AI Patient Summaries 🇯🇵
Why Important: Ubie's new LLM feature automatically summarizes patient interviews for doctors in Japan. This helps them quickly grasp patient concerns and improve efficiency.
"The [LLM-driven] summarised text is very easy to understand, and we are now able to grasp and communicate with patients more promptly than ever before. In the past, we had to selectively copy and paste text from Ubie into the [EMR]. Still, the time and effort required for this process have been greatly reduced, which is also helpful from the perspective of improving operational efficiency. Currently, we turn on the medical interview summary function for most of our medical examinations, and it has already become indispensable for us," - one doctor user
Caution: Broad adoption requires ensuring reliability, privacy, ethics - especially with new generative AI tech. Still in early stages.
3. 🤖 UK Unveils Regulatory Sandbox for AI Tools 🇬🇧
Why Important: Dubbed the AI Airlock, the tool will allow developers to safely test healthcare AI before deployment. This aims to support innovation while ensuring proper regulation.
“We need to ensure that AI is safe and properly regulated, but in a way that doesn’t stifle innovation and access to the latest of medical technologies to improve patient care.” - Paul Campbell, MHRA head of software and AI
Caution: It won't launch until 2024. It is unclear if the sandbox can adequately replicate real-world implementation at scale.
Med AI Research Updates 🔬
1. 🤖 Machine Learning to Predict Heart Disease in Hypertension
Key Findings: ML models incorporating clinical factors and coronary artery calcium scoring predicted obstructed arteries in hypertensive patients with up to 83% AUC. XGBoost ML algorithm performed best, significantly improving risk assessment over traditional models.
Conclusion: ML algorithms combined with calcium scoring may accurately predict obstructive coronary artery disease in hypertensive patients.
Limitations: Single retrospective study with limited sample. Needs validation in broader populations and real-world clinical practice.
2. 🩺 Comparing AI Models for Medical Therapy Recommendations
Key Findings: Claude had the highest quality scores for treatment recommendations across ophthalmology, orthopedics, and dermatology. All models showed gaps in risks/benefits, patient resources, and had errors like vague advice. GPT-3.5 was safest with lowest potential harm ratings.
Conclusion: LLMs can generate valuable recommendations but need refinements to improve quality and safety.
Limitations: Small sample of specialties and expert evaluators. Prompting strategy and benchmarks may not fully capture AI nuances.
Upcoming Med AI Event 🧑💻
Med AI Industry Showcase 📢
Basys.ai - Uses proprietary AI to automate prior authorizations for health plans and systems. The company follows industry standards like FHIR to enable accurate, interoperable data processing.
AiCure - Provides an AI platform for life science organizations to gain insights from patient-level data and their own data. The platform can help with model development, analytics, and clinical annotation.
Artelus - A health-focused startup based in Bengaluru, India. They build primary screening tools that enable doctors to diagnose a large number of patients simultaneously for a variety of diseases, including diabetic retinopathy.
Let’s wrap it up with a fun fact!
According to a report by McKinsey, 56% of companies have adopted AI in at least one function within the organization. That’s more than half of the companies surveyed! 🤖 (source)
Stay tuned for our upcoming editions as we explore the latest breakthroughs and dive deep into AI's transformative power, shaping a healthier future.
Warm regards,
P.S.: If you're a medical professional intrigued by artificial intelligence, but not sure where to start, feel free to reach out to me for personalised guidance.
You can also check out and join our vibrant Med AI WhatsApp Community for Medical Professionals.
Thanks, Avneesh for yet another newsletter on AI in healthcare. Seeing the amount of inroads being made into healthcare delivery by AI, it looks as though the medical curriculum, under- and post graduate, will need a lot of rethinking if young doctors are to stay relevant at all.