How to keep up with medical AI

I read a lot of medical AI news, because we make a free monthly roundup on it. It's one of the most interesting fields I know, and one of the hardest to keep up with. This is what I've learned about doing it without it taking over your week.

The short answer: pick a small number of sources you trust, check them at a regular time, and write down why something mattered when you save it. Most of this guide is about why that's harder in medical AI than it sounds, and the ways people usually go about it.

Why is medical AI hard to keep up with?

Mostly because it isn't one kind of work. A radiologist may be watching validation studies and new tools, while a pathologist is following foundation models. A clinical researcher may care about prospective trials, a health system about deployment and safety, and a medtech team about regulation, payment and procurement. Language models in clinical work bring another set of papers, products and evaluations.

Look at September's roundup. CMS ended the automatic Medicare payment route for Breakthrough Devices. Bayesian Health qualified for payment through the standard clinical-improvement pathway. OpenAI connected ChatGPT for Healthcare to Epic data and nine public health sources. A randomised trial tested LLM assistance with physicians in Indonesia, Kenya and the Netherlands. CROWN, a cytopathology model, was evaluated across 202 task-setting combinations. That's payment policy, a deployed clinical system, a clinical trial and a research model.

Some stories also need careful reading. The multi-country trial's abstract doesn't specify the model, and the full results, effect sizes and subgroup analyses are needed to judge what happened and for whom. The Bayesian Health result was associated with clinicians engaging with alerts, so it shouldn't be read as an automatic effect of deploying the system.

How do people usually keep up?

Most people use some mix of these:

• Newsletters. Good for a regular overview, because somebody else has done the selecting. The catch is that their priorities may favour product launches or large papers over the regulatory or deployment detail you need.
• Journal alerts and preprint feeds. The closest you get to new studies, trials and technical results. They're also noisy, and it takes time to separate retrospective accuracy from prospective clinical evidence.
• Podcasts. Easy to fit into a commute, and useful for hearing why a study, clearance or policy change matters. Less good for sample sizes, comparators and caveats you need to look up later.
• Conferences and meetings. A good way to hear what researchers, clinicians and companies are working on before everything is published. They happen only occasionally, and presentations can leave important methods or results to the paper.
• Colleagues and communities. Often how useful papers, safety concerns and deployment lessons reach you. It works until the person who normally sends them is busy, or everyone in the group follows the same narrow part of the field.

None of them covers everything, so most people end up with a patchwork.

What tends to work

• Keep the list short. Three or four sources you trust are better than fifteen you feel guilty about. If one hasn't given you anything useful in a month, drop it.
• Pick a time. Fifteen minutes on the same day each week works better than catching up whenever you can.
• Write one line when you save something. Why it matters to your work, in a sentence. Note whether it was retrospective or prospective, single-site or multicentre, and what the comparator was when that matters. A bare link rarely gets looked at again.
• Go back to the source for anything you'll act on. A summary tells you what something is about. For the exact result, patient numbers or caveat, read the original. A benchmark, clearance or company-reported deployment measure doesn't answer every clinical question.

That's most of it. It takes discipline, and medical AI gives you enough research, regulation, products and health-system news to fill every spare minute.

One more option

This is why I built Iris. You describe your work in your own words, something like "prospective validation of radiology AI and reimbursement in NHS hospitals", and every month it reads across the field and writes you a briefing on what changed for you, with every source linked so you can check anything yourself.

It's one option, and a good routine with a couple of newsletters works well too. If you'd like audio, our free monthly podcast, Keeping up with Medical AI, covers the month's biggest stories in about five minutes, with the written report and every source underneath. When you'd like a briefing built around your own work, you can set up your own briefing.

If you work in medical AI and think I've missed something, I'd like to hear about it: joe@irisbriefings.com.