Enabling clinician clinical thinking with AI and machine learning
Numerous studies have found that physicians spend almost as much time on “desktop medicine” as they do with patients1. The growing focus on administrative responsibilities at the expense of face-to-face patient care not only challenges physicians wishing to provide optimum patient care, studies have found that the time spent inputting or retrieving data contributes to work-life imbalance, dissatisfaction, and high burnout rates.1 Interventions that have been shown to alleviate the burnout caused by data overload include a team approach to clinical decision-making, reductions in workflow, and improvements in Electronic Medical Record (EMR) technology.2
Enter machine learning and artificial intelligence (AI). The application of these methodologies can potentially offer clinicians even greater support. This can be realized using algorithms that help turn vast quantities of EMR data into meaningful patient insights, augmenting a clinician’s own intuition as they administer patient care.
Hearing the “first whispers” of unexpected patient deterioration
The dilemma of recognizing the early signs of sepsis can be compared to hearing a family member walk through the next room. You intuitively recognize this person’s footsteps, even though you may not be able to articulate it in words. That same family member’s footsteps, however, when walking through a crowded noisy street, would be much more difficult to recognize. According to Paul Mullen, General Manager of Acute Care for GE Healthcare, this is a metaphor for the intuition of clinicians:
“There is hard-earned clinical intuition that lets clinicians know a patient is declining from sepsis or another form of unexpected patient deterioration, but they can’t easily describe it. The clinician just 'knows it when they hear it’.”
As Mullen points out, while clinicians are trained to “hear” sepsis, it becomes far more difficult when inside the metaphorical crowded street of a med-surg unit filled with patients, each with ever-changing conditions. Add to this crowded med-surg, the revolving staff shifts and the need to sort through vast amounts of disparate EMR data in order make meaningful decisions. The problem isn’t a lack of data; it’s “hearing” far too much data than can be processed into useful insights. One promise of AI and related technologies is the ability help clinicians note those first whispers of sepsis before it becomes deadly.
Building User-Friendly Technology for Clinicians
According to Mullen, as the data revolution in healthcare technology unfolds, there is an important opportunity to build user-friendly technology with empathy for clinician needs. Designing technology with clinical empathy can not only enhance clinicians’ abilities and intuition, it sets them free to do what they were trained to do; use their critical thinking skills to navigate the chaotic realities they must deal with every shift.
Says Mullen, “To create an algorithm in a lab with an ideal set of data that’s been scrubbed clean is one thing. We have to make an algorithm work amid the chaos of the clinical care setting. GE Healthcare and Roche are well aware of that challenge.”
Adds Yrjö Häme, Data Scientist at GE Healthcare, “An algorithm has the potential to be impactful if it helps clinicians amplify their understanding of the early signs of the disease. For example, if a clinician develops a suspicion of sepsis, the algorithm could then help them assess this intuition against the patient’s data, potentially nudging them in the “right” direction.”
Unlocking the Potential of Data
Despite containing so much valuable data, the complexity of EMRs makes it difficult for clinicians to find the right information at the right time1. User experience was not necessarily the first priority when electronic medical records were first created at the beginning of this century. According to Andreas Tzanetakis, Sr. Global Product Manager of Acute Care at GE Healthcare, “Data systems needed to be structured in rigid, disciplined forms back when they were built. It was a monumental step forward to go from paper records to digital records. Now, we have the opportunity to add a new layer of interactivity between the data and clinicians through algorithms and machine learning. We can now build on that important and necessary earlier work and make the next contribution.”
GE Healthcare and Roche envision future technology that might liberate and enhance human intelligence in ways they hope will be clinically intuitive. If clinicians could have easier access to the right data at the right time, it could potentially empower them to make more timely and confident decisions during the course of care. It could be particularly impactful in the case of sepsis, where the right data could hopefully enable the early detection of sepsis before it becomes deadly.
1. Allocation of Physician Time in Ambulatory Practice: A Time and Motion Study in 4 Specialties, Annals of Internal Medicine, December 2016. https://annals.org/aim/article-abstract/2546704/allocation-physician-time-ambulatory-practice-time-motion-study-4-specialties Accessed April 15, 2020.
2. Effect of Organization-Directed Workplace Interventions on Physician Burnout: A Systematic Review, Mayo Clinic Proceedings, December 2019. https://mcpiqojournal.org/article/S2542-4548(19)30087-6/fulltext Accessed April 15, 2020.