🍼Tiny Patients, Smart Nutrition: AI's Breakthrough in Preemie Care
11:14 Share

Welcome to AI Unraveled, a show created and produced by Etienne Neumann. He's a senior software engineer and a passionate soccer dad from Canada. Sounds like he's got his hands full. He does. If you find our deep dives insightful, please like and subscribe on Apple Podcasts. We appreciate it. Today, we're tackling a really fascinating topic. It's where cutting edge technology intersects with some of the most vulnerable lives. Well, that sounds intriguing. What is it?

AI in the Neonatal Intensive Care Unit. The NICU. Wow. Yeah. Specifically, how AI can improve nutritional care for premature babies. That's such a critical area. I mean, when a baby is born prematurely, their systems are, you know, they're not fully developed. Right. Even something as basic as

feeding becomes a huge challenge. Absolutely. A lot of times their digestive systems just aren't ready to handle regular feeding. And in those cases, they often have to rely on what's called total parenteral nutrition or TPN. TPN. Yeah.

Yeah. Which is intravenous feeding. Exactly. And for these tiny babies, you know, TPN is a lifeline. It really is because it delivers a really precisely calculated mix of fluids, electrolytes, glucose, you know, all those essential nutrients, amino acids, fats, vitamins. Everything. Everything directly into their bloodstream. Yeah. Bypassing their immature digestive system. Yeah. It's truly a marvel of modern medicine. It is. But there's a catch, right?

It's incredibly complex. To figure out the ideal TPN formula for each baby every single day, it requires close collaboration between neonatologists, dieticians, pharmacists, nurses, all working together. They're relying on the baby's weight, which can fluctuate daily, of course, their lab results, their overall condition, which can change so rapidly. Right. It's a very dynamic situation. Very much so.

And that brings us to a really crucial point that was made by Dr. Nima Akaipour at Stanford Medicine. OK. He said the TPN is actually the single largest source of medical error in NICUs worldwide. Wow. So the very thing that's designed to nourish these delicate infants. Yeah. Is also the most likely place for a mistake to happen. That's right.

But what's interesting is, as Dr. Agaikor's team points out, these errors aren't usually due to incompetence. It's the inherent complexity of the process. Think about it. You have all these daily calculations for numerous nutrients. You need absolute precision when you're transcribing those orders. And there's potential for miscommunication in a busy, high-pressure environment like an NICU. It just seems like the opportunities for something to slip through the cracks are tremendous.

They're numerous. They really are, even with the most dedicated and skilled professionals. So facing this challenge, researchers at Stanford asked, could AI help? That's the question, isn't it? Can AI reduce these errors, bring some standardization to this process? And ultimately make top notch,

nutritional care more accessible to more premature babies. Exactly. So you might be wondering, how can a computer system possibly handle something as individualized as a preemie's nutritional needs? Right. It's a good question. Well, the answer lies in something called a deep learning algorithm. It's a sophisticated type of AI. It's like those systems that recommend movies you might like. Yeah. But instead of movies, this AI is learning patterns in medical data. And it's not something that they just came up with overnight, right? Right. This

This AI had to be trained on a massive amount of real world information. Oh, absolutely. The Stanford team fed their deep learning algorithm a data set of almost 80,000 previous intravenous prescriptions, ones that were given to premature babies. Wow. And crucially, this data included information about the outcomes, you know, how those babies responded to the different formulas, their health status afterwards, all that. So the AI learned from experience. Right.

From tens of thousands of real life cases. And it began to identify the connections between all these different factors in a baby's electronic health record. Right. Things like weight trends, their lab values, their gestational age, you know, their overall clinical picture. And it linked those factors to the baby's nutritional needs. Exactly. And how they responded to different TPN formulas and dosages. So that digital version of a patient's medical history became the AI's training ground. Precisely.

It was all about recognizing patterns that might be too subtle or complex for even experienced clinicians to consistently spot in a busy and ICU setting. So how does this AI actually work in practice? Yeah. Because it's not generating a completely new formula from scratch every day. No, it's not.

Instead, the researchers took the patterns that the AI had identified. Right. And they used them to develop 15 standardized TPN formulas. OK. But these weren't just random. So the AI has these 15 like templates. How does it determine which one is the best fit for a specific baby at a given time?

So it looks at the preemie's current data, all those vital indicators, weight, labs, et cetera. And based on what it's learned from those 80,000 past cases, it recommends one of the 15 standard formulas. And it suggests how long that specific formula should be used before reassessing the baby's condition. So it's kind of like this incredibly experienced data-driven assistant working alongside the medical team. That's a great way to put it.

But it's really important to emphasize that the AI is not taking over. It's not making all the decisions. It's a support tool. It is. It's there to help the medical team to reduce those complex calculations, lessen the risk of manual errors and routine adjustments, which frees up the clinicians, the doctors, nurses, pharmacists and dietitians to really focus on the more critical tasks, the complex cases that need their expertise.

It sounds like it's really streamlining the process. Yeah. And the initial feedback on this AI system has been quite promising. There was that evaluation they did with experienced neonatologists. Oh, yes. The Stanford team presented 10 experienced neonatologists with clinical information for, you know, hypothetical premature babies. And they showed each doctor several possible TPN prescriptions.

Some were actual prescriptions given to real babies. Others were generated by the AI. And the neonatologists actually preferred the AI-generated prescriptions. Yes. They consistently preferred the AI-generated prescriptions to the real ones. That means the AI wasn't just producing acceptable formulas. It was potentially...

Finding better, more optimized nutritional plans, you know, taking into account the insights from that huge data set in ways that individual clinicians, even very experienced ones, might not be able to do consistently. That's really remarkable. Yeah. So if this holds up in further research, the potential benefits for these babies could be huge. Like what are some of the biggest impacts this could have?

Well, I think one of the most immediate and significant impacts would be a reduction in medical errors. If you standardize formulas based on this kind of data and automate the recommendations, we could see a substantial decrease in those calculation, transcription, and mixing errors, which at the end of the day means safer care for these babies. And it goes beyond just preventing those mistakes, doesn't it? I mean, it could also improve the consistency and the quality of care. Yes.

The AI brings in this element of objectivity and data driven best practice, which means that those critical decisions are informed by, you know, the outcomes of thousands of previous cases. Right. Which could lead to more consistent and better care no matter which clinician is making the recommendations. And efficiency in the NICU is paramount as well. Of course. Automating the routine parts of TPN prescription frees up so much time for the whole NICU team.

Doctors can spend more time on complex diagnoses and treatments. Nurses can focus more on direct patient care and monitoring. Pharmacists and dieticians can apply their specialized knowledge to the most challenging cases. So it's really improving the workflow. Yeah. And looking beyond a single NICU, there's also the potential to improve access to high quality care in places that might not have all the resources of a major medical center. Absolutely. Think about NICU.

So what about NICUs in less developed regions? They may not have the same level of specialized expertise or the vast experience found in, say, big academic hospitals. An AI like this could provide expert-level recommendations. It could essentially bring a higher standard of care

all over the world. Remember those researchers talking about making doctors better and making top-notch care more accessible? Well, AI could play a huge role in achieving that. That's a powerful vision. We have to remember that this is all still in the early stages. It hasn't been implemented in hospitals yet. Right. It's still being researched and developed.

So what are the next steps? What needs to happen before we see this AI actually being used to help babies? Well, rigorous clinical trials are absolutely essential. They need to prove that the AI not only generates preferred prescriptions, but also that it actually improves patient safety, treatment effectiveness, and overall outcomes compared to the way things are done now. Makes sense. I mean, with any new technology, especially in health care,

There are always important considerations to keep in mind. Like we need to be absolutely sure that patient data is being kept private and secure. Absolutely. And there needs to be transparency. Yeah. We need to understand how the AI is making its recommendations. Right. And then there's that whole question of accountability. Like if an AI recommendation, however unlikely, leads to a negative outcome.

Who is responsible? These are complex issues that require careful thought and planning. They really do. But the goal here isn't to replace humans in the NICU. It's not about diminishing the role of human touch, intuition, and judgment. It's about giving clinicians better tools to manage this complex and critical task, reducing the potential for error, and leading to the best possible outcomes for these vulnerable babies. It really highlights the potential for AI to be a force for good in healthcare.

This goes way beyond just automation or entertainment. This is about using technology to solve real life-saving challenges. It is.

And by learning from a huge amount of past experience, this AI could make intravenous feeding for preemies much safer, more consistent, and more effective. - While we wait for those clinical trials to happen, this research offers a really hopeful look at how technology can play a crucial role in improving care for these tiny, fragile lives. - It shows that innovation,

when guided by compassion and a focus on human well-being, can lead to some truly amazing advancements. Speaking of advancements, if you're interested in advancing your own skills in areas like cloud computing, cybersecurity, healthcare, and more, you should check out Etienne Newman's AI-powered Jamcac app. It's a great app. It can help you master and ace over 50 in-demand certifications. You'll find links to it in the show notes. So as we wrap up, here's something to think about. Beyond the NICU, beyond nutritional support,

How else could we use AI ethically and effectively to make other complex medical processes better? That's a great question. And how could it improve patient outcomes across different areas of healthcare? It's a fascinating area that deserves a lot more exploration. Thanks for joining us for this deep dive on AI Unraveled. And don't forget to check those show notes for links to Etienne's AI Power Jam Get Tech app.