As a new nurse, did you ever work with a seasoned nurse and just wish you could download their brain into your own? What impressed me most was their ability to recognize when a patient was experiencing a significant change in status before the rest of us. Their wealth of experience enabled them to recognize patterns in patient data—sometimes in intuitive ways that they couldn’t even articulate themselves. So, what if there was a way to support newer nurses in recognizing patterns in patient data to enhance their ability to recognize changes in a patient’s condition or a patient’s risk for an adverse events? That is the potential that machine learning has in nursing when it is used to drive clinical decision support (CDS) systems.
Machine learning (ML) is a branch of artificial intelligence (AI) that focuses on building computer applications that can learn and adapt without being programmed with specific instructions. Supervised ML is one type that involves training an algorithm with known inputs and outputs. In healthcare, we have a wealth of data about our patients such as lab values, vital signs and other assessment data, comorbidities, and much more. We also know the outcomes of patients we cared for in the past and whether each of these patients experienced an adverse event, such as sepsis or a fall. Data on these historic patient outcomes can be used to train machine learning models to predict adverse events or other outcomes for our current and future patients. This support to recognize patterns in patient data could help nurses and other care team members intervene earlier and implement interventions to mitigate the identified risk and reduce adverse patient outcomes.
During my work as a nurse informaticist, I managed a pilot project of a sepsis machine learning model decision support system. When I turned to the literature to find evidence-based practices to implement this novel form of CDS safely and effectively, I found that the literature on this topic, particularly studies that focused on nurses, was sparse. As a result, I have made the integration of ML-driven CDS the focus on my dissertation studies.
In my first qualitative descriptive study, my aim was to explore perspectives surrounding ML-driven CDS design, development, implementation, and adoption. I interviewed nurses about their understanding of how machine learning decision support works, their preferences for design features, and their perceptions regarding the role of nurses in the development and implementation of these systems. Nurses also shared their preferences for training strategies and discussed how peer and leader attitudes influence their trust and use of these systems. The aim of my second study was to determine whether there are associations between the format and level of complexity of ML-driven CDS displays and nurse satisfaction with the outputs of these tools, including how nurse characteristics, such as numeracy and graph literacy, might influence these relationships.
The landscape of AI in healthcare is changing rapidly, and many of you likely have seen recent news stories about large language models like ChatGPT being piloted for use in healthcare. Nursing as a profession needs to engage in a discourse about how various forms of AI might influence and change nursing practice. For example, do we want large language models to script nursing notes and messages to patients? What are the risks and benefits of these tools? Though they might potentially save time and increase efficiency, how might they impact nurse critical thinking skills and patient-centered care? Is there a danger that nurses will become too reliant on these decision support systems? What are the risks of propagating existing healthcare disparities using potentially biased data?
My research aims to amplify the voices of nurses to learn about their experiences with and perceptions about ML-driven CDS and aid in answering those questions. Nurses can provide unique insights on data inputs in ML-driven CDS systems as well as identify which healthcare problems may or may not be appropriately tackled by ML-driven CDS. Through this research I hope to provide nursing with a voice in the design and implementation of these emerging technologies.
Ann Wieben, MS, BSN, RN-BC, is Clinical Research Program Specialist at UW Health and a PhD candidate in the University of Wisconsin-Madison School of Nursing. She is a member of Sigma’s Beta Eta at-Large Chapter.