In the above video, Jonathon Feit made predilections about artificial intelligence in 2019 that have since came true.
My colleague Chris Witt and I designed a set of algorithms around 2009 at Carnegie Mellon University to determine the best place to transport patients based on a range of factors and accounting for critical contexts, like surges due to terrorist events involving hospitals, weather emergencies and infectious disease outbreaks.
Our concept derived from on my personal experience living in Pittsburgh, near one end of a bridge. The other end is equidistant from three hospitals. Here is a theoretical but not impossible scenario: Assume that a car crash at the end of the bridge is precipitated by a myocardial infarction. The impact of the crash sparks a fire that penetrates the vehicle. The ambulance crew arrives on-scene to find a patient who has experienced major trauma, is in the midst of a heart attack (the severity of which is still unknown), and who risks significant full-body burns. The equidistant hospitals specialize in, respectively, trauma, cardiology/STEMI and burns. Where should the crew take the patient? We can make the scenario more complicated by adding a child into the mix, or a disability, or a POLST form that is on file for the patient but which the crew in the field doesn’t know about.
It turns out that humans aren’t great at these multivariate calculations. Computers, on the other hand, are outstanding at them. The reasons for the distinction are fascinating and essential: humans evolved to rely on biases and excellent pattern recognition—not to solve for stochastics (that is, randomness). Most of us also have a sense of ethics, medical and otherwise. We might make assumption about traffic patterns, or the time it would take for a patient to be seen at one facility or the other. We might neglect to consider (or simply fail to incorporate, perhaps because we didn’t know at the time) both clinical and non-clinical details, like whether the patient is a hemophiliac—or if the patient’s religious beliefs preclude interventions such as blood transfusions.
Some patients are also terrified of medical debt—for example, if they get transported to a hospital that won’t take their insurance—than they are of a longer, more painful recovery. Some considerations flip the order of preferred destination for good reasons, or for reasons that may be only look good in the eyes of the patient at the time. If only such information could be included as part of the decision-making calculation, in a split second, while standing at a patient coding alongside a burning car.
Most Precious Cargo
“Optimization” is the math behind the evaluation of who to take where, when and why. Carnegie Mellon University—my alma mater, and where my company was born—alternates with the Massachusetts Institute of Technology as either first- or second- best in America at this type of science. Some years after graduating, I had the opportunity to speak with Baxter Larmon, then Director of the UCLA Center for Prehospital Care. Although he cautioned me against using the phrase “patients as packages” in public, for political reasons more than anything related to the words themselves, on multiple occasions since our first chat I have thought to myself: “How appropriate a description!” Ambulances are, in fact, trucks hauling most precious cargo, handled (hopefully) with care.
Moreover, if our profession learned from the likes of FedEx and UPS, across the board we would likely enjoy more efficient and profitable operations—because logistics companies are exceedingly good at optimization problems whose complexity is based on the relationship between the requirements and its constraints. This might seem obvious—but when applied to a 250-patient population following a mass shooting or the potential to relocate and protect millions of people bracing for storm—the logistics get really complicated, really fast.
Do I simply take my patient to the closest facility? How do I know which facility is the right one, and whether or not it has space and specialization? In San Francisco—this is bonkers, in the beating heart of the world’s technology ecosystem—for years much of this calculation has been done using a physical map and a radio. But now, all the rage among technologists, investors, journalists, and government wonks is the potential impact of ChatGPT and other artificial intelligence programs on…well…everything.
More than three years ago—a few months ahead of the COVID-19 pandemic—I argued all of the above at a conference called Intelligent Health AI in Basel, Switzerland, highlighting that algorithms developed eleven years earlier were still relevant and needed. A video of my presentation is above. Fast-forward to 2023, and only do the algorithms remain an excellent fit for strong decision-making, but in light of the COVID-19 pandemic, weather disasters that are growing in size and ferocity, and the twin deadly scourges of shooting and shooting up, something beyond “get ‘em and go” is warranted during everyday Mobile Medical encounters—and even more so when capacity is going to be constrained. Yet field crews’ reliance on data-driven, “A.I.-powered” decision-making in disaster response, patient routing, heads-up awareness of key clinical context, and family reunification, is still effectively nonexistent in the United States.
Ironically, even further back—in 2017—I was quoted in Becker’s Hospital Review, in an article called “How artificial intelligence apps are changing patient engagement.” I told the author that A.I. “closes a critical set of gaps. Reading pieces of paper when you need to be taking care of a patient is a problem.” It’s even worse when there are many patients in dire need of help.
We have only barely contemplated the value of A.I. to facilitating real-time matching of patients and conditions—to go past conveniences like reducing charting time (seems a bit risky to have ChatGPT write your ePCR narrative…) or improving billing and collections. Push further: past “Minority Report” and “Total Recall.” (Bypass “Idiocracy” entirely, please.) What if we deploy technology now that does more than system status management, telling agencies how long their rigs have been sitting, waiting, waiting, waiting…What if we fix the problem by keeping the delays from happening in the first place, by accounting for clinical and environmental conditions, capacity, economics, service delays, whether the patient should be cared for in place, and even gas mileage.
We can automate notifications so nurses know how long they have to get a patient ready for discharge, which unclogs patient movement from emergency department to inpatient setting, and in turn reduces wall times. No guesswork or politics involved—rather, the argument was to manage operations using real-time data. I first heard this excellent logic from Danielle Thomas, COO of Lifeline Ambulance in Los Angeles, in mid-2022; before year’s end, New York State EMS Director Ryan Greenberg presented the same argument as, essentially, the salvation for his state’s wall time challenges.
Not only is all of this doable—it is happening now, and could become standard if we embrace the idea of “patients as packages.” Dr. Larmon, of Los Angeles, disliked the phrase but Chief Mike Metro, also of Los Angeles (County, not City) was the first to ask me: Why can’t we check patients in into the hospital like we can check ourselves in for a flight? The two approaches are similar and Amazon excels at both (“your delivery is X stops away!”). Artificial intelligence can not only smooth clinical operations but boost profitability, too. Maybe then Mobile Medical professionals won’t have to worry so much about being called “essential,” because the money—like the patients—will just flow.
More from the Author
Transparency for the Future of Mobile Medicine
