Practically every emergency department has dramatically changed over the past decade. A vast majority have upgraded with high-resolution 64-slice CT scans, cutting-edge diagnostic lab equipment, and teleradiology, which all generate patient data that converge for the provider to digest in the electronic medical record.
American health care has been successful at bringing patient data together in one location for physicians, nurses, and other providers to use. One could argue that there is almost too much information to be able to make nuanced decisions in real time. This is where machine learning and artificial intelligence can really affect patient care positively.
As emergency physicians, we've all had near-misses. Several weeks ago, I was working in the ED, and a patient came in feeling generally unwell; he had chest congestion and a temperature of 99°F. The patient did not meet any sort of admission criteria using various Medicare tools, but he had a persistent tachycardia of 102 bpm. Arguably, he appeared to have a common cold or perhaps non-typeable flu. His chest x-ray, labs, and lactic acid were normal. Everything was normal.
But I made a case for admission to the hospitalist, who thought I was an idiot. He agreed, however, to admit the patient to observation status. The next morning, the patient was transferred from the observation unit to the ICU because he had developed pneumonia and pulmonary edema and appeared to be septic.
Was this heroic medicine? Not really. Looking through the retrospectoscope, subtle evidence suggested that this patient had something. I could have easily missed it and sent him home. This got me thinking about how artificial intelligence could play a role in health care. The hospitalist and the case manager correctly argued that my patient didn't meet any admission criteria; the risk was the hospital could eat the cost of admission. This is the first barrier of care: It can bias and take decision-making away from clinicians.
What if the decision-making were based strictly on rules data from the electronic medical record and past, present, and even future projections based on data modeling? Couple that with clinical findings, interactions with the patient, and nurses' findings, I think we could be on to something.
Today's electronic medical record is a treasure trove for data analysis. Algorithms are being developed by EMR vendors, researchers, and clinicians that digest data in nanoseconds and present evidence-based clinical decision-making tools for physicians to digest. Are we there yet? Not by a long shot, but I am excited about how all of this will play out in the next five years or so.
The imperfect science of the ED triage system could also benefit from AI enhancements. Dr. Dominique Jean Larrey is credited with creating the first triage system by stratifying patients into three categories: unlikely to live regardless of care, likely to live if they receive immediate care, and likely to live without care. Today's modern ED has implemented the five-level Emergency Severity Index for the same purpose. This widely used clinical decision-making tool has some subjective components, and is based on available resources in the ED. AI can develop algorithms based on anonymous data collected over several thousand visits and better predict the right treatment by the right provider in the right location of the ED.
We are at a crossroads in medicine where machines will help clinicians make better decisions, and when we make better decisions, the whole health care system wins.Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.