How AI can actually help clinicians
Human failures are system failures
“You could have killed him!”
This moment is seared into my brain. Even writing this now I am feeling a little nauseous.
I walked in at 7 AM to get ready for my residency assignment. I walked up to get a report for my floors and I was asked about a patient from the night before when I was staffing.
I told them exactly what happened.
The patient came in with a skin infection and cultures were not back yet I tried to get a hold of someone but was not able to so I approved the antibiotics because, from the limited information I had it was the right medication. There were no labs either and I had handed it off before I left for the night.
My heart is already starting to sink. The worst thoughts are racing through my mind.
Usually, you don’t get asked about a patient that you took care of unless something odd happened.
Then as you can probably guess, I got chewed out. It turned out that the patient had a rectal abscess (very bad infection), had cancer and he was neutropenic (low white blood cell count). This changes everything. Thankfully for me, the cultures were sensitive to the antibiotic I had verified. Which meant they should work.
Regardless, of the outcome I was on the brink of tears. I just stood there like a zombie. It was like one of those movie scenes where someone is talking to you but you don’t really hear what they are saying.
The rest of the day I was in a daze.
I had every emotion rushing through my head.
I was angry that they didn’t understand. I was sad because that was someone’s father, husband, or brother.
I contemplated quitting medicine that day.
Spoiler alert….
I didn’t quit medicine and had a pretty fulfilling career.
I’m telling this story because the system failed. We only rise to the level of our system. Anytime you are relying on human input you are going to have breakdowns.
In this case, the failures were:
The patient came in with very little information on the chart because he was a direct admit
The treatment team could not be reached to fill in the gaps
The pharmacy team (me included) did not do a good job of passing along the information
3rd shift pharmacist is only per pharmacy personnel and doesn’t always have time to look at everything
3rd shift staff in general is a skeleton crew so labs weren’t seen until hours later
This is not an issue with just this hospital this could, and has, happened at every hospital. Even the most prestigious ones. Saying that we have operational and system issues in healthcare is like saying water is wet. The real question is how can we prevent this from happening?
The Fix
TWO WORDS…
ARTIFICIAL
INTELLIGENCE
Thank you for your support…..
I say this in jest and also kind of seriously.
Let's break this scenario down to each step.
Patient Record
The biggest issue we all had was that the patient was directly admitted from the facility that he was at to the hospital. Every decision we make in medicine is based on data. Past medical history, labs, notes. When you take that away we are just going in blind. The only person that really knows their history is the person that admitted them. In this case, all we knew is that he had an infection and was in for surgery the next day.
This is the exact reason it is vital that we have an open exchange of health records. There are legislation and voluntary agreements out right now but I will believe it when I see it.
Some of you may be thinking why didn’t we ask the patient? That is because they were on a lot of pain medications and were not awake. The nurses tried to no avail.
Also, to those thinking that the surgeon should have called back. If someone’s life is dependent on returning a page then I am sorry we are all screwed.
The only reason why we don’t have more mistakes in our system is you have healthcare workers performing herculean feats daily.
Automation
The rest of the issues could have been solved by an automated system. It could be an AI model or a decision tree.
One of the biggest issues with this whole scenario was that the labs resulted overnight but were not looked at until the morning when nurses were preparing for shift change. That is when the pharmacist was called. Thankfully the drug he was on was the correct one but that doesn’t mean that the next time they are put on the right medication.
I purpose that we implement a system that would automatically prescribe the right antibiotic based on predetermined guidelines. In this case, you are not relying on the staff to come back and check.
Yes, we should be checking. But there is a hospital full of sick people. None of them are going to be like “we know you’re busy with this guy, we will stop being sick for a couple of hours.”
And if you have a code arriving you can throw all your timelines out the window. A code for those that do not know is when someone is literally dying and we are trying to stop that from happening. Some can be an hour long. I have been part of those.
So what would this system look like?
I am glad you asked!
I am thinking we could do it in two different ways
The first way, and the one that would be easier to build, would be to have an automated solution that is linked with the hospital's antibiogram that sends a suggestion to the provider with a prompt that they can agree or disagree with the change. An antibiogram “is a periodic summary of antimicrobial susceptibilities of local bacterial isolates submitted to the hospital's clinical microbiology laboratory.” It is what we use to decide what antibiotics are the best to use for your specific location. Each hospital can have a different preferred antibiotic based on the hospital population.
The pros are that you keep the clinicians in control, and it doesn’t require them to keep track of the antibiogram or keep checking for labs.
The cons are you keep the clinicians in control…😅
Now before I have the mob running after me…
Sometimes people are stuck in their ways and will do what they want, and that can lead to some issues.
Another con is additional alerts can get buried or ignored.
The second way, and the more complicated to build, would be the same as above but it makes the changes on its own and sends a message saying the change was made.
The pros for this are that it is automated and could be instantaneous. One of the biggest issues is trying to get a hold of a physician or infectious disease team to change the antibiotics or get them de-escalated. They are so busy and if we’re relying on pagers we will have problems.
On the flip side, the con is that you are taking the clinician out of the decision-making process.
To me, the compromise lies somewhere in the middle. Where your hospital chooses which infections can be run through this system. Let’s say for example if you have a drug-resistant organism growing then you need the physician to make the call. But, for superficial skin and soft tissue infections, you can have the algorithm decide.
Then there is the very large elephant in the room, one that is overlooked when AI is talked about. Who is liable?
Is the model liable, the physician, or the hospital? I personally do not know the answer to this. I will say that if there is a breakdown in the system then it should be the company. I know that is not going to sit well with some people, but if you are going to take the control away from the clinician then you can’t expect them to take on the liability. This is one of the biggest issues that is hindering the adoption of AI now and also in the future.
There is a very famous acronym in healthcare…C-Y-A. It stands for “cover your a$$.” It is exactly what it sounds like. You make sure that you have documented everything and left a trail so if something goes wrong you have a leg to stand on. I personally hate this mentality because it halts innovation and out-of-the-box thinking. Unfortunately, this mindset is adopted with the litigious nature of our society.
Would a system like this help in my story? Yes, I believe so. It would have prompted a change if necessary. In my case, I was lucky that the patient was put on the proper regimen.
Is a system like this replacing a clinician? I don't think so. Right now the workflow is usually you call someone to figure it out or grab your trusty Sanford Guide and just prescribe what it tells you. And if it is a very bad infection then you usually call in the infectious disease specialists.
You know where something like this could make a massive impact. In those small hospitals and clinics that don’t have enough resources. A system like this could help them focus on the most extreme cases and have better outcomes.
Thank you so much for your support! Please do not hesitate to reach out if you have any questions or leave a comment.
You can connect with me on LinkedIn and I am trying to be more active on Twitter. Would really appreciate a follow there!
If you enjoyed what you read join 800+ founders, investor, clinicians, and executives learning how to create better healthtech products every week!
I hope you have a great day!
How to create better healthtech products. Learn from someone with almost 10 years of healthcare experience. Join 800+ founders, investors, clinicians, and executives for free.