I am Dr. Ryan Melvin, Ph.D., the Department of Anesthesiology and Perioperative Medicine’s data scientist. My background is in physics and statistics, but my passion is figuring out what the heck data is trying to tell us.
Before UAB, I did predictive analytics (telling the future with math) for one of the top 10 US banks. They’re still in business, so I couldn’t have been that bad at it, right?
This departmental data science blog has two purposes. First, I want to provide examples of what statistical learning, machine learning, and data science in general can do for you. Through these examples, I also hope to peel back the curtain a little bit and demystify the black box of machine learning. Second, I want this blog to be an educational resource for statistical tools and appropriate use of machine learning. This second kind of post will range from pitfalls of those statistical tests you learned back in undergrad all the way to how much you should trust a cancer-spotting AI.
You may already have examples in mind of questions that traditional (sometimes called parametric) statistics can answer, such as “Do these data sets have different means?” or “Which of the three drugs had the greatest impact?”
The questions that machine learning algorithms do particularly well at answering look more like “I wish I could know if a patient is at higher risk for sepsis just one day in advance.” or “I wish I could know if ventilator asynchrony has occurred just 10 seconds after it happens.” You’ll have to forgive my lack of medical knowledge in those examples, but hopefully you get some idea of the form of the questions. There’s a piece of knowledge or prediction and a time frame within which you need that knowledge or prediction. If you have a question like this or even a question that might be better suited for traditional statistics, reach out to me at rmelvin@uabmc.edu, and we’ll get started!
If you have some extra time on your hands and want to know more about how machine learning can impact medicine and medical research, check out my talk on the subject at https://youtu.be/opADRqZHv5A .
If you have quite a bit of extra time on your hands, allow me to make two book recommendations: Deep Medicine by Eric Topol and You Look like a Thing and I Love You by Janelle Shane. The title for that second book was generated by an AI, but the book itself was not. At least, I’m pretty sure it wasn’t.
