By: Brian Lynch, COO – Hero AI
The healthcare industry is in a global crisis. I have spoken to people around the world and everyone believes that this is a unique problem with unique underlying drivers, unique to their country … In Canada it’s a Canadian problem, in the US it’s an American problem, and in the UK it is a distinctly UK problem, but it’s both simpler and more complex than that.
There are not enough healthcare providers. As populations age, people are retiring from medicine while simultaneously needing more medical care. The World Health Organization projects that there will be a shortfall of 10 million healthcare workers by 2030, and the UK may account for 350,000 of that if current trends continue.
This is when the economist in me perks up. When you have increased demand and reduced supply, there is a standard modeled outcome. Prices rise for the resources in demand. Many cannot afford those prices in the short term, but luckily efficiency gains start to develop that drive down prices until market equilibrium is met… problem solved.
This doesn’t work when the resources needed are humans, and the supply is capped by the speed of a country to attract and train clinicians.
This reminds me of an old joke that economists love. Three economists go hunting and spot a deer in the woods. The first one lines up his shot, and misses 3 feet to the left. The second lines up and shoots 3 feet to the right, to which the third yells “YES! We got him!”.
As every economist knows, the models average out long-term to an equilibrium, but they don’t necessarily reflect the short-term pains on society. Are people just going to accept that millions will suffer undertreated because we don’t have enough resources while we wait out the long-term efficiency gains that get us back to equilibrium?
As any Emergency physician can tell you, we are already starting to see the crisis in the real world, and the short-term costs are already reaching an inexcusable level. Even in the best and most well-resourced hospitals in the world, patients suffer during high volume surges, especially when hospitals cannot find the staff to ramp up to meet demands. For many patients, a prolonged wait is a terrible inconvenience, but for some, it’s the difference between life and death.
Accident and Emergency Departments (A&Es) are overcrowded with increasingly sick patients. The result? Critical elements and risk factors are being missed as there is not enough human capacity to dig through every patient chart… which brings us to my thesis. Emergency medicine NEEDS artificial intelligence.
The problems in A&Es are not new, and not even unique to medicine. Populations have been aging for years now, but most other industries have been able to drive their throughput per employee up enough to handle these changes through digitization and automation of processes. Healthcare has been working on the digitization part, introducing Electronic Patient Records (EPRs) and virtual care, but we aren’t seeing the same level of productivity gains through digitization as other industries. In fact, the opposite may occur where clinicians see less patients than they did before. Data captured into the EPRs are trapped behind computer screens and in databases, and aren’t factored into the care of patients until someone thinks to read it. So, digitization hasn’t driven nearly as much automation as in other verticals because those other verticals have focused on using digital data to drive automation and efficiency. Healthcare is missing the automation layer that has become table stakes in other industries.
If you don’t believe me, do a quick Google search of “patient dies after waiting” and you will not find a single case, you will find hundreds of cases spanning many years and across almost every geography. While the details vary considerably, the words used in the stories are very consistent:
- The A&E was packed and over capacity
- “Unfortunately” the wrong patient was accidently left waiting too long.
- “Unfortunately” while they waited they deteriorated.
- “Unfortunately” they ultimately passed away.
What the articles miss is that this is not just unfortunate. “Unfortunate” implies it was bad luck, while this is a systemic issue and human tragedy. Data that could have saved these patients’ lives was captured in the EPR but not seen until the postmortem review of that patient’s death.
So, what does this have to do with AI? There are processes for all of these cases, and didn’t someone just need to follow the protocol? Isn’t this a problem that could be solved with a checklist? As Atul Gawande described in great length in the Checklist Manifesto, errors like this are drastically reduced when you introduce checklists, but how many checklists can one clinician go through for each patient? There are so many potential risks and procedures that it isn’t realistic to expect a clinician in an A&E to check off every possible box, and every process related to the standard of care, and keep track of these items that are time sensitive, all while managing a massive queue of people. Our clinicians in A&Es are working tirelessly day in and day out and need modern tools, powered by AI to keep up.
I think it’s clear at this point that without automation, emergency medicine is in big trouble for at least the next decade. So, it sounds simple, focus on automation. How does it work in other industries, and what did they do to improve throughput?
Well let’s take banking as an example – I’m a bit biased in this because I have spent over 15 years building automation into banking processes.
Every time you tap your credit card, about 10,000 columns of data can be attached to that tap; everything about the customer, the transaction, the merchant, the timing, the financial movements, and more. Dimensions on top of dimensions on top of dimensions are captured. The process is designed so that the next stage of transaction processing can happen automatically, and every piece of information needed is pre-processed to perform that function. In other words, the whole process is designed to be machine readable. Banks know that it’s not feasible to have a person look at every transaction and manually push it to the next appropriate process. If they took that approach, we would be complaining a lot more about wait times at the grocery store, or going right back to good old fashion cash.
We haven’t figured this out in healthcare. For us, it has always been a guiding principle that a human must push forward every aspect of care. Afterall, it’s just too risky to leave decisions like “what happens next in this patient’s care” to a machine. So data is captured in a way that is human readable. A few key identifiers, a few vital signs, and a ton of unstructured text or images. A person can easily create/digest a paragraph of text while computers traditionally struggled with this, but it would be impossible for a person to navigate 10,000 columns of data to make a decision. There are years of human readable data hiding in EPR databases, but there just aren’t enough humans to read it.
Enter AI.
With the latest developments in AI, large language models (LLMs) have made it possible to understand clinical text and drive care forward in a way never seen before. At Hero AI, we leverage cutting-edge LLMs and elevate them to clinical grade using our proprietary clinical automation platform. This has empowered hospitals to accelerate care for patients who need it most. Our clients are improving patient outcomes, reducing wait times and length of stay, and literally saving lives. We’ve created AI Sidekicks that can be taught to quickly understand nuanced clinical workflows that work alongside clinical teams to expand capacity and move care forward for patients. For example, our Mental Health Sidekick advocates for patients in need of psychiatric support by automating psychiatry consultation immediately after triage. This has reduced wait times for psychiatry intervention by 55% and decreased length of stay by nearly 2 hours on average. This is just one example of the incredible impact AI can have in healthcare.
A report by the C.D. Howe Institute, a think tank in Canada states “Increasing the supply of health-sector workers is a necessary but insufficient response to the troubles in the sector”. If emergency medicine is going to survive, we have to do better, and in this economist’s opinion, AI is critical to achieving that dream.
If you want to learn more about how Hero AI is working to transform care, reach out to UK@Heroai.ca.
