ASHP Pharmacy Futures 2024 Insights.

Andrea Skora, PharmD, MSCR, BCCCP, FCCM, FCCP, clinical associate professor at the University of Georgia College of Pharmacy, explained that artificial intelligence (AI) extends beyond popular applications like ChatGPT and into technologies that include computer processing chips. And includes a computer cooling fan. during a presentation at the American Society of Health System Pharmacists (ASHP) Pharmacy Futures 2024 in Portland, Oregon. Additionally, Skora explained that the category of AI can include mathematical and statistical modeling, including machine learning.

“Machine learning is the ability of a computer to learn without being explicitly programmed,” Skora said. “Before machine learning, you would have a computer program, and you would say, 'Whenever you see red, It means stop.' So your program read means stop.”

With machine learning, a computer would be able to recognize patterns in the stoplight sequence itself, such as whether or not cars stop as a result of a red light or go as a result of a green light. Even with exceptions to this behavior, the computer will be able to understand the existence of these exceptions rather than abusing the exceptions in the rule.

“[With machine learning, the computer is] Learning that rule, as opposed to us explicitly telling that rule,” Skora said. “It comes from different mathematical and statistical modeling methods, like random forests and Bayesian networks; These are all types of machine learning.

Due to the current popularity of artificial intelligence, the term is often used without clarifying exactly what it is referring to, notes Skoura.

“I think AI is so hot and excited that you're going to hear, 'Oh, it's based on AI, or it's based on machine learning,' and I think it's important that we all take a step back and say, 'What do you really mean?' In light of this, I saw a funny comment the other day that one of the best rebrands of all time is linear algebra which is now called machine learning,” Sikora said. “[For example,] Logistic regression is, in a very technical sense, machine learning, but logistic regression has been around for a long time. I understand what people mean. [AI] is helpful, because [it could be referring to] Machine learning, which is really new and fancy, and this technique was created in the last decade, or you might have something like logistic regression, which has been around for a long time.”

Specifically, Sikora defined AI as the science and engineering of creating intelligent machines capable of achieving human-like goals through a constellation of technologies. Additionally, Scoura defines machine learning as the ability of a computer to learn without being explicitly programmed to do so, with a number of different methods (such as random forests and Bayesian networks) that give computer algorithms examples. and allow learning from experiences (data sets). Predefined, rigid rules-based methods.

Additionally, AI is being used in many ways in healthcare, including diagnosis and treatment (ie, clinical decision support and symptom analysis), computer vision (ie, radiologic image analysis of electrocardiograms). [ECG] and electroencephalogram), workflow (ie, patient flow optimization and process failure detection), predictive modeling risk stratification (ie, readmission rates, hospital-acquired infections, and complications emergency), mobile apps (i.e., health, mental health, and chronic disease management), health medicine (i.e., genomics), discovery (i.e., clinical trials, hypothesis generation, and proof of concepts), and matching engines. (ie, patients with similar profiles and treatments with similar cost-benefit ratios).

“What's interesting about AI in the healthcare setting is that it can be so impressive in one very narrow activity and then be so aggressively bad in another area. Potentially looks at things like the ability to diagnose. [ST segment elevation myocardial infarction] It's better than a cardiologist what it reads in an ECG, and it's very impressive, and it's probably going to redefine radiology right now,” Skora said. “I think it's really, really cool. There is, but I think that immediately in our mind, we can transfer the meaning [more]”

Key takeaways

1. The growing role of AI in healthcare. AI is significantly impacting healthcare through a variety of applications such as diagnostics, treatment planning, workflow optimization, and drug discovery, due to its ability to improve patient outcomes and operational efficiency. shows the
2. Machine learning capabilities. Machine learning, a subset of AI, enables computers to learn from data without explicit programming by using advanced statistical methods to identify patterns and predict outcomes in complex healthcare scenarios. .
3. The importance of clarity and impact assessment. Understanding the specific definitions and applications of AI is critical to effective implementation, and evaluating the real-world clinical impact of AI models to ensure they translate into meaningful improvements in patient care. .

According to Scoura, along with the tasks and skills we know AI is particularly adept at, there are areas where AI is completely incapable of accomplishing such tasks.

“It's still a possibility. [if you tried using an AI for a new task] It can actually be pretty bad in that particular job,” Sikora said. “Again, what's trying to balance. [AI] What is and what is not [capable of]I think this is an important path.

Currently, AI has been found to be particularly beneficial for use in the field of pharmaceutical discovery, Skoura noted.

“One area that I'm particularly excited about is the concept within discovery. The idea that we can have AI-based algorithms that are basically looking at how proteins are structured, And then trying to find different chemical molecules and different drugs. [that would be] A good idea,” said Skora. “When you look at the funnel of all the chemical compounds. [in relation] As far as FDA-approved stuff goes, it's a pretty tight fit. Machine learning can help us with this.

Within clinical pharmacy, Sikora explained that AI models excel in their ability to perform large-scale predictive work, as they can discover patterns within data quite effectively.

“One of the problems with more traditional regression modeling is that it struggles with too many variables in the equation at some point,” Skora said. “I'm working with some biologists, and they keep saying, 'Your data is too dimensional,' and all the data that [are] There is only one medication administration record. [intensive care unit (ICU)] The patient, and I just limited it to 24 hours, and they're like, 'Okay, well, that's still like 40 medications that were given. What about food? What about frequency?' These models can't necessarily handle that, whereas AI is showing the ability to actually see all the drugs that were given to a patient in the ICU, and not just the drug, but the dose, the frequency. , root, all these types of different properties or variables within the system. So, it's very interesting.”

Skora noted that these AI models are also helping to predict disease, especially in complex data sets.

“With something like sepsis prediction, you have different early warning systems that are based on AI that are finding sepsis patients hours before clinical recognition. But then I just have a suggestion with a computer scientist. was working on who was really interested in heart rate variability, and he thinks that his machine learning-based models of heart rate variability can predict it even 24 hours in advance,” Skora said. . “The idea that we can learn about sepsis. [for a] A day before the patient is a very incredible concept.

AI can also generate relatively new variables from variables whose existence was not yet recognized by humans.

“Again, heart rate variability is a great example of this,” Sikora said. “Typically, we think of vital signs as heart rate, diastolic, systolic, and things like that. Heart rate variability has to do with the space between your heartbeats, and There are 27 different variables in it. It's crazy to think about the kind of analytics required to calculate these variables from different waveform data, and then put them all together in a meaningful way. There are things that I think AI is opening up that we just didn't have the potential for 20 years ago.”

Skora explained that the clinical impact of these models is an important avenue when learning about new AI models in development, as it has a direct impact on patient care. With all the interest in AI right now, Skora noted that the AI ​​models being developed within healthcare aren't looking at the clinical implications for patients.

“You're going to hear a lot of people who have really neat AI-based this or that, and maybe even [applies to] By the end of that,” said Skora. “But then you have to ask the question, what? [this AI] actually do for patients? To clean the data, process the data, train your algorithm, and make sure the algorithm does what you said it was going to do. But many times now, we're struggling with actually getting this product into the clinical space and seeing the impact on patient outcomes.”

During a presentation at the Society of Critical Care Medicine (SCCM) 2024 Critical Care Congress, Skora noted that SCCM presenters discussed the use of AI in a population of approximately 50,000 patients with 100 different variables to determine the mechanics of exclusion. Failure can be predicted. According to Skora, presenters were more enamored with AI's ability to predict who would fail extubation than they were with how AI would be applied in the clinic.

“I was in the crowd, and I was like, 'OK, did you compare that data to the Tobin index? And they were like, what's the Tobin index? I explained, well, the Tobin index of extubation. The current medical standard for predicting failure, and it's actually just dividing one number by another, you can do it in your head. [with] division,” Sikora said. “They were like, 'Oh.' And I was like, 'Yeah, so we probably have to compare it to this clinical standard, which is a standard, right?'

Further, Skora noted, it was unclear what AI actually achieved and how it could be used and provide benefit in clinical practice. In particular, Skora learned during the presentation that the excision failure prediction model proposed in SCCM, for example, had not been evaluated for its practical use in a clinical setting.

“I think, right now, what there is a lot of excitement about. [AI] can do that we haven't necessarily translated it into the medical space,” said Skora. “There's a lot of interest right now in what AI can do in terms of operations. So, things like inventory management, diversion analytics, automation of various operations and all that are quite exciting.

At ASHP Midyear 2023, Skora noted that he attended a talk where presenters explained that for the first time they could use AI to find a diversion by a nurse or doctor. For Sikora, it was an important tool that has immediate, practical use in a clinical setting.

“It was really cool, that you could see that they did such a good job of learning patterns of drug use and learning ways to change drug use that even the first time someone had their Turned around, so the AI ​​can find it,” said Skora. “Isn't it nice that before the person goes too far down that path, you can intervene and maybe help the person? Very interesting, but this kind of analytical power is brand new, and I think it's redefining the field. With that, comes great responsibility.”

Reference

Infrastructure needed to implement artificial intelligence in SCORA A clinical pharmacy. American Society of Health System Pharmacists Pharmacy Future 2024; June 8-12; Portland, Oregon.