Doctors Said He Was Dead, A Machine Learning Model Saved Him

Hello everyone and welcome to my newsletter where I discuss real-world skills needed for the top data jobs and specifically the AI Agent Role. 👏

This week we discuss how a large language model is saving lives.

Not a subscriber? Join the informed. Over 200K people read my content monthly.

Thank you. 🎉

A little over a year ago, Joseph Coates was told there was only one thing left to decide. Did he want to die at home, or in the hospital? Suffering from POEMS syndrome, Joseph Coats experienced severe complications that prevented him from getting the only potential life-saving cure, a stem cell transplant. POEMS syndrome is a rare blood disorder that can impact various organs. Coates's hands and feet were numb, his heart had enlarged, and his kidneys were failing. Every few days, doctors needed to drain fluid from his abdomen.

As both the doctors and Joseph’s family gave up hope, his partner, Tara Theobald, refused to do the same. She reached out to Dr. David Fajgenbaum, a physician in Philadelphia who is conducting research into AI applications in medicine. Tara sent an email detailing Joseph’s condition and sought his help.

Fajgenbaum and his colleagues helped create a machine learning model for drug repurposing. This involves using drugs that are currently approved to treat certain conditions as therapies for rare illnesses that don't have effective treatments. Essentially, a machine learning model can comb the literature and find drug side effects that are actually beneficial for patients suffering from those potentially life-threatening conditions.

A day after Tara sent the letter, Dr. Fajgenbaum replied, offering an unconventional therapy. He proposed the combination of chemotherapy, immunotherapy, and steroids for Coates. The therapy worked, and Coates was strong enough to receive the stem cell treatment four months later.

The doctor did not devise the therapy himself; a machine learnig model suggested it. Fajgenbaum had previously treated his rare condition (Castleman disease) with a drug called sirolimus, which is usually used for kidney transplant patients.

Fajgenbaum founded Every Cure, a nonprofit that uses AI for drug repurposing. The nonprofit received over $100 million in funding from TED's Audacious Project for Advanced Research Projects Agency for Health. Every Cure will use the funding, in part, to conduct clinical trials for AI-based repurposed drugs.

This type of AI use might benefit hundreds of millions of people worldwide who suffer from rare, difficult-to-treat diseases.

Each of these conditions might only impact hundreds of thousands of people in a country. The National Institutes of Health (NIH) defines rare diseases as conditions that affect fewer than 200,000 people in the US. But it quickly adds up when you take into account the entire world.

"If you comb through enough drugs, you eventually find the side effect you're looking for," Dr. Matt Might said, "and then that becomes the main effect."

Like Fajgenbaum, Dr. Might is working on drug repurposing with the help of AI. He and his team at the University of Alabama developed an AI model that helped them treat a 19-year-old patient suffering from chronic vomiting.

They used a prompt similar to what you'd type in ChatGPT to find answers to a specific question: Show us every proposed treatment there has ever been in the history of medicine for nausea. The model suggested the young patient inhale isopropyl alcohol through the nose, which is usually used as an antiseptic.

Large language models alone and in concert with agents are going to revolutionize the field of medicine and this is a great example of an outlier model saving lives.

Thanks for reading and have a great day. 👏

I’ve put together an introductory course on mastering the basics of Copilot Studio for those who are thinking about becoming an agent engineer on Microsoft’s ecosystem. For this article only get 50% off the entire platform including the agent course by using the code YOUTUBE25 at checkout. I’ll take this away in a few days so hurry.

I’m using content from around the globe that does the best job of explaining the concept. I’m also outlining the information you need to take away from the video. I’m also working on a complete course for Copilot Studio and then one for AI Foundry. Microsoft is going to dominate the agent space and I want you to be prepared for these roles using either interface.