Leveraging machine learning to rapidly create clinical AI algorithms

 

There are few artificial intelligence (AI) applications for pediatrics and there are many clinical and technical issues that arise that require automation solutions tailored for pediatric care. Phoenix Children's Hospital embarked on a journey to harness the power of AI in solving these clinical challenges and now uses machine learning to create new AI algorithms, often in as little as one day. 

At the Healthcare Information Management Systems Society (HIMSS) 2023 meeting, David Higginson, executive vice president and chief innovation officer of Phoenix Children's Hospital, shed light on the remarkable impact of AI. By using AI to develop algorithms quickly and efficiently, the hospital has significantly reduced the time required for algorithm development, from months to mere hours.

"So what that does is let us try lots and lots of different things. And what I've learned with AI over the years is the chance of you're getting it right the first time is probably small. So it's kind of an iterated process," Higginson explained. But, instead of each iteration taking months to work on, the new system they use speeds things up so several iterations of a new AI can be tested in a week. 

"The one that we have had the most success with is predicting malnutrition in children. When a child comes to a hospital, maybe they have a broken the leg, they're coming into the emergency department. The doctor's not thinking malnutrition, they're thinking, okay, I'm going to fix the leg. I got to treat the thing in front of me. But often the underlying cause is malnutrition and it can go undiagnosed for quite a long time. So we've developed an algorithm and trained it on many years of data and then put it in stealth mode. And I said to the dieticians, how about I give you three people this week and you just go by and see what you think about those people? Sure enough, they came back and said, well, two of the three had malnutrition, how did you know that?," Higginson said. 

They would test the algorithm further with refinements and give the dieticians 10 more patients to look at the next week. This process helped boost confidence in the algorithm to a point where it is now actually placing an order for consults in the electronic medical record (EMR)

"We're finding six to 10 patients a week who have undiagnosed malnutrition. Now, if you think about that from a family member of a child, that's a huge difference. And those things are really impactful in terms of practical AI, and that's kind of spawned other ideas, but that's been kind of one of our great use cases," Higginson explained.

The journey of AI integration to create new clinical algorithms

Five years ago, Phoenix Children's Hospital embarked on a journey to harness the power of AI in solving clinical challenges. The traditional approach of relying on biostatisticians to develop algorithms proved to be time-consuming and often inefficient. He said the team might work on an algorithm for several months and find it does not work well in the end. So Higginson's team opted for a different path, utilizing automated machine learning. This approach involves providing a dataset to an AI system that autonomously creates the algorithm, allowing the hospital to start using it within a matter of hours, rather than weeks.

One of the key lessons learned from using AI in healthcare is that getting it right on the first attempt is a rare occurrence. Thus, an iterative approach is essential to fine-tune algorithms over time. 

While there are now many vendors selling commercialized AI algorithms, Higginson said many are to generalized for the needs of his hospitals, which another reason why they have decided to develop their own, highly customized algorithms. 

"One of the things I've learned with AI over the years is it doesn't translate very well. So I'm always very skeptical of vendors that tell me, 'I've got an AI model that's going to work great,' because geographic factors are a huge influence as well. There are some clinical conditions which obviously translate, but I think we've seen some recent examples where models are trained in one state, lifted somewhere else and don't work," he said. 

For example, he said they created AI models on operational things like our donors and managing their employees, which require very local and customized factors that are completely unique. "Understanding how far is too far for an employee to travel into work all depends on the road density, where they are traveling from. I think the concepts and the ideas are transferable. But I would be a little skeptical of taking that black box and just lifting it somewhere else," Higginson explained.

Tailoring solutions for pediatrics

Pediatric healthcare presents unique challenges that often require tailored solutions. At Phoenix Children's Hospital, they've developed their own patient portal, recognizing that pediatric patients and their families have distinct relationships with healthcare providers. This patient portal addresses the complex dynamics of patient relationships within families and guardianship scenarios. This includes who has access in a divorce or foster home situation, and the ages when patient information needs to be shared with the patient. 

Moreover, the hospital has adapted to the post-pandemic landscape by embracing telehealth services, which have been particularly well-received by pediatric patients and their caregivers. The implementation of hybrid telehealth, where patients and their caregivers join virtual consultations, has transformed the healthcare experience for families, Higginson said. 

The future of AI in healthcare

Higginson encourages a more general application of AI in healthcare, emphasizing its adaptability to a wide range of scenarios. He used the example of AI helping determine no-show rates to better staff the emergency room. Another example is AI can be used to sift through patient emails to doctors via the patient portal to determine the most appropriate recipient within the healthcare team. This could streamline communication and enhancing efficiency so doctors can practice at the top of their license not not spend a large amount of time sorting basic email requests. Higginson said doctors tell him over and over 80% of these messages are about scheduling, medications and billing which have nothing to do with the physician.

"So how great would it be to take that message that came in and run it through a GPT prompt and ask it, which help desk should this go to?" He said. 

Phoenix Children's Hospital's innovative approach to AI demonstrates the immense potential for the technology in healthcare. By adopting a strategic and iterative approach, they have successfully developed clinical algorithms that not only improve patient care, but also enhance the overall healthcare experience for pediatric patients and their families.
 

Find more HIMSS coverage.

Dave Fornell is a digital editor with Cardiovascular Business and Radiology Business magazines. He has been covering healthcare for more than 16 years.

Dave Fornell has covered healthcare for more than 17 years, with a focus in cardiology and radiology. Fornell is a 5-time winner of a Jesse H. Neal Award, the most prestigious editorial honors in the field of specialized journalism. The wins included best technical content, best use of social media and best COVID-19 coverage. Fornell was also a three-time Neal finalist for best range of work by a single author. He produces more than 100 editorial videos each year, most of them interviews with key opinion leaders in medicine. He also writes technical articles, covers key trends, conducts video hospital site visits, and is very involved with social media. E-mail: dfornell@innovatehealthcare.com

Around the web

Compensation for heart specialists continues to climb. What does this say about cardiology as a whole? Could private equity's rising influence bring about change? We spoke to MedAxiom CEO Jerry Blackwell, MD, MBA, a veteran cardiologist himself, to learn more.

The American College of Cardiology has shared its perspective on new CMS payment policies, highlighting revenue concerns while providing key details for cardiologists and other cardiology professionals. 

As debate simmers over how best to regulate AI, experts continue to offer guidance on where to start, how to proceed and what to emphasize. A new resource models its recommendations on what its authors call the “SETO Loop.”