"It could lead to a whole new field of investigation in clinical and physiological sciences and reshape our understanding of human physiology."
There’s a lot of theoretical applications for machine learning in healthcare, but almost as often as we hear about those, we see a new study outlining an actual, practical tool made possible by a series of smart algorithms. Adding to that ever-growing pile is a new tool that can predict hypotension—low blood pressure—in surgical patients as soon as 15 minutes before it sets in.
The methodology was built atop nearly 550,000 minutes of surgical arterial waveform recordings from 1,334 patients’ records (which included more than 25,000 instances of hypotension). It required high-fidelity recordings, but from them it could extract more than 3,000 unique features per heartbeat, leading to millions of data points to base the algorithm on, according to the study.
"We are using machine learning to identify which of these individual features, when they happen together and at the same time, predict hypotension," lead researcher Maxime Cannesson, MD, PhD, said in a statement. Cannesson is a professor of anesthesiology and vice chair for perioperative medicine at UCLA Medical Center.
Once developed, the algorithm was validated against more than 33,000 minutes of recordings from over 200 patients’ records, which featured about 1,900 instances of hypotension.
In validation, the technique did quite well: sensitivity and specificity were 88% and 87%, respectively, 15 minutes before a hypotensive event: They improved to 92% each at 5 minutes before onset.
"Physicians haven't had a way to predict hypotension during surgery, so they have to be reactive, and treat it immediately without any prior warning. Being able to predict hypotension would allow physicians to be proactive instead of reactive," Cannesson said.
The ability to predict hypotension during surgery might allow physicians to avoid potentially-fatal postoperative complications like heart attack or kidney injury, the researcher added.
And unlike many theorized machine learning interventions, this one is closer than many to becoming a clinical reality. A piece of software—Acumen Hypotension Prediction Index—containing the underlying algorithm has already been submitted to the FDA, which granted it De Novo status in March. It is proprietary to devices made by Edwards Lifesciences—the device company behind the machine learning technique. It has already been approved for commercial use in Europe since 2016.
"Although future studies are needed to evaluate the real-time value of such algorithms in a broader set of clinical conditions and patients,” Cannesson said, the work “opens the door” to application of the technology on other physiological signals, like EKG readings for cardia arrhythmia or EEG measurements of brain function.
“It could lead to a whole new field of investigation in clinical and physiological sciences and reshape our understanding of human physiology,” he said.