Vendor Viewpoint: EMS in a Nanosecond: The Impact of AI and Machine Learning

Vendor Viewpoint is a platform for our advertising partners to expound upon future directions in technologies, vehicles, accessories, educational offerings, and other products EMS providers rely on to perform their jobs. EMS World’s editorial staff does not endorse or promote any products or companies discussed in this column.

For many of you the concepts of artificial intelligence (AI) and machine learning (ML) are not new. But have you ever envisioned how their ability to do millions of complex things in a billionth of a second could impact you and the public health and safety industry as a whole?

Similar to your personal life with connected smart devices, EMS providers operate under an “Internet of Things,” or the network of physical devices that are connected and sharing data with other systems in the background without you or others knowing. These networks have revolutionized operational processes for collecting, sharing, analyzing, and reporting data. The EMS industry is already a long way down the path of data collection and visualization; now we look toward the challenge of using that same data to draw conclusions and make decisions through a future of AI and ML, although we can’t turn a blind eye to the value of our EMS providers and the advances they have helped lead. 

Coders of Tomorrow

Through the documentation of patient encounters, today’s EMS providers are essentially the coders of tomorrow’s AI technology. The next time you arrive on scene, remember you’re helping to write the curriculum that will be used to train these AI models of the future.

Imagine a world where we no longer have to physically operate or instruct our devices, but instead we share the operational experience with them. Physical interaction will no longer be required to derive the data points around us, allowing providers to capture more information than ever—with less effort. This reduced data burden on providers allows them to be more focused on scene. 

This world starts with named-entity recognition, a subtask of AI that seeks to extract information mentioned in unstructured text to locate and classify named entities and organize them into defined categories such as name, age, gender, race, medical codes and quantities, and so on. From there a subtask of ML known as text analytics and its ability to associate words and phrases with structured data types enables models to identify key data points through trained algorithms.

On Scene

Let’s look at these benefits from an emergency setting. EMS arrives on scene, and a provider approaches the patient and an unidentified bystander. The bystander yells, “My daughter needs help, she’s having a seizure! She’s only 8!” Instantly the relational extraction capabilities of AI engines can identify that the patient is a pediatric female and update her chart. 

AI continues to consume the patient encounter as unstructured text, and as the word seizure and height measurements are picked up and categorized, our system suggests midazolam or lorazepam as potential medications and an estimated dosage for each. The provider selects midazolam, and this action is time-stamped while the medication is linked to its RxNorm code to populate the proper structured data section documenting how it was administered, where the IV was placed, and the patient’s reaction. 

Still our models thirst for more. Was our recommendation accurate? Did the patient require a second dose? What was the outcome? By asking these questions and centralizing more data, ML works to create systems of accurate suggestion or conclusion for certain events, allowing for continuous quality improvement. This leads us to a future paradigm of data speaking for itself through models and algorithms, allowing us to request advice, share ideas, and capture knowledge that would not otherwise be apparent to humans.

As AI, ML, deep learning, and other technological advances open new opportunities for the future of EMS, we can realize operational efficiencies never thought possible. Think of the models monitoring patient activity, acuity levels, time of day, traffic patterns, primary care locations, and more to identify the most appropriate type of treatment and location to receive it. 

With this powerful shift of data to knowledge, exciting new innovations await us to improve patient care and outcomes, create more efficient operations, and better serve our communities.  

Joe Graw is president and chief operating officer of ImageTrend Inc., in Lakeville, Minn.