If storm clouds are rolling in and you don’t want to get caught in a sudden downpour, you may err on the side of caution. But maybe the sun is still shining. Do you risk it? After all, there is a 75 percent chance it won’t rain.
Now imagine you’re a physician covering an intensive care unit. While you’re putting orders in the EMR, an alert pops up on the screen. The system’s artificial intelligence algorithm detected that one of your patients has a 25 percent chance of developing sepsis in the next eight hours.
What should I do? The question runs through your head.
The decision is certainly weightier than whether to bring an umbrella. Patients with severe sepsis can have a mortality rate of 50-80 percent or higher. The patient’s life could be at risk and the window to act is short.
Should I administer fluids? Should I break from the standards of evidence-based medicine and proactively prescribe antibiotics?
The algorithm that provided the alert is based on population-level data with a history of success in predicting outcomes. Given the same parameters 25 percent of the population would develop sepsis in the next eight hours.
How meaningful is that for the individual at hand? The patient appears healthy right now —what should I do?
Should you start treatment — even if the potential diagnosis doesn’t exist yet — and it has a 75 percent chance of never existing at all? Herein lies one of the key challenges of applying AI to clinical decision-making. Though still in their infancy, AI algorithms for diagnoses are beginning to provide probabilistic information. When those probabilities are not clearly tipping the scales in one direction or the other, they present high-stakes dilemmas, particularly in the context of standard evidence-based medicine, as with the potential sepsis patient.
Evidence-based medicine was designed to guide physicians to make decisions by prioritizing scientific evidence combined with clinical expertise and patient values. It has been considered modern medicine’s gold standard for several decades, replacing individual clinical authority and opinion. Evidence-based medicine helped incorporate scientific methodology and statistical analysis to try to reduce uncertainty and improve clinical decisions.
Evidence-based practices are certainly not perfect. Studies may present contradictory information, or datasets and research methods may be inadequate, among other potential issues. Also, evidence-based medicine, which rests on population-level findings, does not always provide consistent guidance for patient-level decision-making.
Often, a gap between the study population and the individual patient, who may have many factors influencing their health outcomes, exists — creating a margin of error that cannot be reconciled with a randomized controlled trial. In fact, the “controlled” trial may produce results that do not translate to an “uncontrolled” environment.
However, despite the limitations, evidence-based practice has improved care and outcomes tremendously. In many ways, adding AI to the clinical decision matrix builds on the principles of evidence-based medicine, with potentially improved methods that may bridge the gap from the population to the individual better.
Algorithms pull from vast datasets beyond the capability of human consumption, making the resulting insights considerably data-driven. Fundamentally, this data may provide the answers that physicians are looking for.
What is the probability that another patient just like the one in front of me improved with the same clinical decision-making? Was there a better decision to be made? What can I learn from the petabytes of data that are out there?
The potential to revolutionize healthcare is limitless.
The current state is not perfect. The decision-making is still limited by several factors — including the accuracy of the data. Given our current state of EMR keeping, system restrictions, interoperability and the entry of data, our data accuracy and integrity should give some pause.
People make thousands of decisions a day on perceived probabilities as well as algorithmic ones. Tech giants like Netflix, Google and Amazon use algorithms to predict optimal outcomes and suggest movies, content or products based on viewing history, age, demographics and other factors. We know these algorithms may not always get it right. Every suggested movie or product may not be satisfactory. However, picking the wrong movie isn’t life and death — just like choosing not to carry an umbrella isn’t the end of the world.
However, starting the wrong course of treatment or failing to intervene appropriately do have severe consequences in healthcare.
How do we make the insights currently provided digestible enough or actionable enough to be useful at the bedside?
It’s a challenge that will require stakeholders from all corners of the industry to decide: How will we conduct research in a world generating petabytes of data daily at each health system? How can we reconcile such findings with evidence-based medicine in a way that is meaningful and appropriate for the individual patient at hand? What kind of protocols do we need to create a path forward that is both prescriptive and flexible?
AI presents a window into previously untapped and life-saving insights, signaling the need for a paradigm shift in the fundamentals of medical decision-making. If we can make the shift appropriately, we will add one more positive revolution to healthcare.
More articles on artificial intelligence:
AI chatbot startup Babylon Health attacks physician for ‘2,400 Twitter troll tests’
Wearable AI could prevent 30% of heart failure readmissions: U of Utah Health study
AMA, Google & 50 more health IT leaders unveil standard for AI in healthcare
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
