Are you a free-hander or a navigator?
Are you an artisanal surgeon or do you look forward to a future of engineered processes and computational medicine?
As AI (artificial intelligence) and data push their way into the OR and physician’s offices, surgeons and their staff will be grappling with new and profound questions—including how to frame the practice of orthopedic and spine care itself.
In the following editorial we’ll check in on AI’s progress in medicine, so far.
Artificial Intelligence as a Creative Engine
By now, I expect 60, 70, maybe 80% of orthopedic surgeons have taken ChatGPT for a test drive. And, I expect, the experience was good. Maybe great. Playing with ChatGPT, a large language model, has changed our awareness of what AI is and what it can do.
In the knowledge and creative industries, generative models like ChatGPT have been transformative. More than half of all software code is now written by generative AI models. The Hollywood writers’ and actors’ strike is based on a fear that AI generative models will mimic writers, creators and, actors in hyper-realistic ways.
Bottom line, among its consumers, the tool of artificial intelligence has evolved from an analytical tool to a creativity and productivity app.
Artificial Intelligence as a Disintermediating Engine
The three basic manifestations of AI modeling (generative models like ChatGPT: deep learning models like AlphaGo and the combinatory models that created the novel drug Halcyon) have been deployed to create novel therapeutic drugs and proteins.
AI, in these applications, is a powerful disintermediating force.
Mike Nally, who was one of Merck’s top executives and currently CEO of Generate Biomedicines and CEO-Partner of Flagship Pioneering, described at a recent AI panel discussion how AI is being used to accelerate drug development.
“When you make this transition from artisanal craft to engineering, it introduces a new scalability dimension,” explains Nally. “If you’re relying on individuals or a team of geniuses, that’s really hard to scale. When you’re relying on computation, it’s very easy to scale. And so, I think, given the productivity challenge that is associated with research, Artificial Intelligence offers the prospect of taking advantage of scalability of technology.”
“We see this in our organization. We can generate [protein or molecular] sequences instantaneously. We can work on 15-20 programs with a group of about 200 and be comparable to the output of my former employer [Merck Pharmaceuticals] with 72,000 employees. These rules are being changed as we put computational techniques at the center of some of these processes.”
Nally’s company, Generate Biomedicines, creates breakthrough medicines using machine learning and AI techniques. His AI tools study millions of proteins and kick out generalizable rules which decode biologic functions—like pain relief, tissue healing, viral or infection reduction, you name it.
Nally is doing this Merck-scale with 200 knowledge workers. Instead of a 20-year development cycle, which is fairly typical of drug development, Nally is looking at a fraction of that time—40-50 weeks, for example.
Added Nally, “This productivity change will force us to think radically differently about business models, about access to healthcare models, because, ultimately, if you can solve productivity, you can get more medicine to more patients. Which is why a lot of us do what we do.”
The challenge, of course, is the next step in the process of bringing novel therapeutics to the patient. A one-year clinical study still takes one year.
Artificial Intelligence as a Predictive Model Engine
Most experts in AI applications in medicine see predictive modeling as the most promising way to crack the productivity problem at the clinical level.
Again, from Mike Nally’s discussion: “We can have better predictive models. We rely on animal models today. Animals are bad predictors of human biology. There are better techniques that have to be introduced.”
So far, the effort to apply AI, whose internal engines are fundamentally P-Value generators, to improve the clinical aspect of medicine has fallen short.
In a 2022 editorial in the journal Nature, authors Marwaha and Kvedar write: “A recent systematic review by Zhou et al, the authors surprisingly show that AI’s impact so far has been quite limited.”
The Nature authors go on to say: “They reviewed 65 randomized controlled trials (RCTs) evaluating AI-based clinical interventions and found that there was no clinical benefit of using AI prediction tools compared to the standard of care in nearly 40% of studies.”
Furthermore, the authors write, “The clinical benefit of using deep learning (DL) predictive models over traditional statistical (TS) risk calculators was only minimal, and there was no benefit in using machine learning (ML) models over TS tools.”
“Somewhat counterintuitively, most of the AI tools in these trials exhibited an excellent area under the receiver operating characteristic (AUROC; a common performance metric for predictive models) during development (median AUROC 0.81, IQR 0.75–0.90) and validation (median AUROC 0.83, IQR 0.79–0.97): a humbling reminder that robust predictive utility does not guarantee clinical impact at the bedside.”
Another way of thinking about the AI revolution in medicine is with the following analogy.
Math was the tool that scientists used to understand physics.
AI and Data Science, in combination with the Law of Large Numbers, will be the tool that scientists and healthcare professionals use to understand biology.
Is Biology more complex than physics?
I think so. One scientist described the complexity of trying to unravel the mystery of biology as trying to map the fingerprints of GOD.
Risks to Turning Medicine Into an Engineering Process
As authors Marwaha and Kvedar essentially argued in their Nature editorial, Artificial Intelligence predictive models have so far been humbled when confronted by the complexity of clinical medicine.
Personally, I’m wary of people who think they apply AI to break the clinical medicine system down to its component parts and make it an engineering science.
Still, AI is coming. But where?
About 80% of FDA approvals in the AI area are for medical imaging. So, imaging.
Also, Microsoft and Epic (the largest supplier of electronic medical record systems in the United States) recently announced that they have teamed up to bring AI-based ambient listening systems into the examining room to draft a physician’s clinical notes, link to reimbursement codes and then, finally, to be added to the EPIC electronic medical records.
So, logistics.
Which, studies have shown, means less paperwork, less burnout, more time for direct patient care and fewer clinical mistakes.
That’ll work. Also, stay tuned.
