As artificial intelligence (AI) becomes more integrated into healthcare, much of the conversation focuses on the technology itself. But behind every AI tool is a foundation of clinical data—and much of that data are generated by nurses.
In a Q&A, Kelly Gleason, PhD, RN, FAAN, informatics nurse and associate professor at the Johns Hopkins School of Nursing, discusses the role nurses play in creating the data that power AI, and what that means for the future of care.
A large share of the data that feeds AI comes from electronic health records. What role do nurses play in creating that data?
There’s really nobody who’s with a patient as much as the nurse. They’re the main creators of electronic health record (EHR) data; they document what they see and that determines how we understand what’s happening clinically.
When we talk about AI in healthcare, we’re often talking about systems that rely on structured data. And the main creators of that structured EHR data are nurses and patient care technicians.
What kinds of data are we talking about?
It’s a wide range of information that nurses collect throughout a shift. That includes vital signs, shift assessments, height and weight, and “intake” and “output,” so how much a patient is eating or urinating.
Nurses also document elements of a patient’s social and clinical history. All of that becomes part of the structured data in the EHR that other clinicians rely on and that AI systems can use.
How are those data used in AI systems?
A lot of the AI tools we’re seeing right now are built to identify patterns in that data. For example, early warning scores, deterioration indexes, and alerts like sepsis alerts or stroke alerts are often driven by the kinds of data that nurses document.
So when those systems are running in the background, they’re pulling from data that are being entered in real time at the bedside.
Why does that matter for how we think about AI in healthcare?
It matters because the people who enter the data understand the context behind it. Nurses see the patient in person, so they may know why something looks the way it does in the chart.
For example, a patient’s heart rate might be elevated, but the nurse knows it’s because the patient was just moving or anxious. Or it might be elevated and concerning. That context doesn’t always show up in the structured data itself.
Nurses know how the data are being entered and what it represents, so if they aren’t involved in how these systems are designed, then something could be missed.
What risks come up if the data are incomplete or biased?
There are many risks. For example, there are known issues with certain clinical tools. Pulse oximeters, for example, don’t always read as accurately for dark-skinned people. So you can have situations where a patient’s condition isn’t fully reflected in the data these systems use, which creates a faulty baseline for clinical standards AI systems adopt.
Does that mean AI can miss things that clinicians catch?
Yes, and we’ve seen examples of that. In one study, we found that nurses were documenting more frequently on patients in the hours before hypoxia was identified. That suggests they were concerned about the patient even before objective measures showed a problem.
That’s a good example of how clinical judgment and observation still play a critical role. Data are important, but it doesn’t always capture everything that’s happening in real time.
What does this mean for the future of AI in healthcare?
AI architecture needs nurse clinician involvement—not just researchers, but bedside nurses who are actively entering the data.
As we become more data-driven and rely more on tools that use that data, it’s important that the people who are creating it are involved in how those systems are built. Otherwise, you risk designing tools that don’t reflect how care actually happens.
What’s the biggest takeaway for people thinking about AI in healthcare?
The technology only matters to the extent that it helps us take care of patients. Nurses are going to use it as much as it helps them do their job.
At the same time, as we become more reliant on these tools, we have to think carefully about how they’re designed, how they’re used, and what happens if they fail.
AI has a lot of potential, but it depends on the quality of the data behind it. That data starts with nurses.
