Using AI Tools to Reduce the Risk of Adverse Drug Events in the ICU

Each year, an estimated 1.5 million intensive care unit (ICU) patients are at risk of experiencing a potentially life-threatening adverse drug event (ADE), or an unexpected negative reaction to medication.

Several factors account for this significant risk, including the intricate medication regimens often required for critically ill patients (who are already highly susceptible to complications), the demanding workload of ICU staff, and, sometimes, simple human error.

A proven strategy for mitigating the risk of ADEs in the ICU is the integration of critical care pharmacists (CCPs) into the healthcare team. These specialized pharmacists closely monitor patients' medication regimens, using a patient’s personal health history and current status to inform decisions about dosage, medication adjustments and the early detection of potentially dangerous reactions before they escalate into crises.

This is the fifth article in an ongoing series on artificial intelligence in the health sciences. See other articles in series.

Research indicates that the involvement of a critical care pharmacist can lead to a 70% reduction in ADEs. Scientist and clinician Andrea Sikora, PharmD, MSCR, FCCP, FCCM, an associate professor in the Department of Biomedical Informatics at the University of Colorado School of Medicine, aims to further improve these outcomes. Sikora's research focuses on developing artificial intelligence-based tools capable of predicting ADEs long before they occur, thus helping to guide decision-making and stave off potential problems.

We spoke with Sikora about the importance of critical care pharmacists and the promising potential of AI in this field.

Why are ICU patients at such high risk for adverse drug events?

The average ICU patient is on 13 to 20 medications, which is risky in itself. On top of that, the medications we're using tend to be high-risk neurotherapeutic index drugs, like neuromuscular blockers, which paralyze your whole body, or big, broad-spectrum antibiotics that have lots of adverse side effects associated with them.

What are critical care pharmacists, and what role do they play in the ICU?

A critical care pharmacist (CCP) has done postgraduate residency training after getting a PharmD. They conduct comprehensive medication management, or CMM, which is a very personalized type of surveillance. For example, as a CCP, I might look at a patient and see that they're on an anticoagulant. The physician's done a good job – they’ve ordered the right anticoagulant at the right dose, and the patient needs it. But I might notice that the patient has some risk factors for bleeding, and their hemoglobin has been trending down, so I’ll suggest we get one more (hemoglobin) level to check out where we're at or think about doing a preventative dose decrease while we're not sure about their organ function. Some of those smaller tweaks are what help prevent medication errors.

How can AI tools support CCPs and potentially further reduce errors?

One of the challenges with ICU patients and medications is that we're often operating in situations of uncertainty and comparing risks for different things. For example, you might have a patient who has sepsis and also has heart failure. A very standard treatment for sepsis is to give fluids for low blood pressure. On the flip side, giving fluids can be really bad for heart failure patients and can make their blood pressure worse. So, the question is: Should you or should you not give fluids? How that decision is made is often based on best judgment. There is no data driving it on an individual patient level.

The idea behind these tools is that I would be able to give you a prediction. If you give fluids, you have a 50% risk that the patient ends up on a mechanical ventilator. But if you don't give fluids, you have a 100% chance that they're going to have an acute kidney injury. The team might look at that and decide that they're probably going to be on the ventilator anyway, but we really don't want this acute kidney injury. You're providing some objective data and prediction modeling for risks to help guide your decision-making. These are tools that can make CCPs more efficient and better at their livesaving role in the ICU.

Is prediction modeling currently being used in ICUs?

Prediction modeling for ADEs, pharmacist workload, and general ICU complications are still in the development stage, but AI models called early warning systems (EWS) are currently being used.

Being in the ICU puts you at high risk for infections and pneumonia. What would usually happen is you have to develop full-blown pneumonia with a fever and poor oxygenation, indicating that we need to give you antibiotics now. And by the time you start treatment, you have put the patient at higher risk of a bad outcome. Some of the AI and machine learning-based models like EWS can detect signs of the earliest onset of infection that would not meet the human eye. The models have learned to look for those types of patients and sets off an alarm so we can start treatments earlier before things get out of hand. Early warning systems are showing reductions in mortality, which is very exciting.

You’ve written a book on mentorship, Pay It Forward: A Path to Mentorship in Medicine, and the importance of team science. Tell us about the inspiration behind that book.

The National Institutes of Health would say that team science is the way that the best science happens. I have biostatisticians, epidemiologists, informaticists and computer scientists on my team now. They're all bringing in unique expertise that's informing all of this work.

I became more passionate about mentorship when it became obvious to me that everything that I had done and achieved was the result of another individual being the helping hand or the insight along the way. Mentorship is far more than just a senior person providing advice or insights for someone just starting out. It is truly a bi-directional relationship characterized by intentionality of the individuals toward mutual growth and shared altruism. Team science and rich learning ecosystems are essential for scientific discovery as well, but mentorship provides unique and irreplaceable benefits to the people involved. The book is kind of a map toward understanding best practices.

Guest Contributor: Christina Frank is a Brooklyn-based writer who specializes in health and medicine.