Skip to main content

Artificial Intelligence in the Healthcare Arena: Implications for Emergency Medicine

This is really interesting. Elon Musk has another company called OpenAI, an artificial intelligence company. OpenAI built a neural network that was designed to teach *itself* how to play a very difficult video game and then the AI played against the best players in the world on stage in front of millions of people. 

What’s really interesting is that the AI used strategies that the humans had never considered and could not even initially understand. 




The implication for the ER is really interesting. Optimizing ED workflow is really a “game” in the same sense - you have two teams (providers/nurses/techs vs patients), with limited resources and functions, and each side has specific goals (for patients it is to get out quickly and have their healthcare issue resolved, for hospitals/doctors it is to keep the patient alive and get them out as quickly as possible by using the fewest resources). 

Imagine if somebody created a video game that mimicked the people, forces, goals, and parameters of the emergency room, and then created an AI bot to play it, in order to figure out the best strategy to "win." That strategy could then be studied and learned from by humans.

For those who are naive to the challenges of Emergency Medicine, here are some of the huge problems in our specialty:

1. How should one staff?

How many doctors and APPs (Advanced Practice Providers, i.e., physician assistants) should you staff and at which hours? It depends on the volume and acuity of patients coming in, but you can't predict that accurately enough, so your best guess is usually based on averages seen in previous months and years.

However, if you staff for the average, you'll be understaffed 50% of the time (putting people's lives at risk), and overstaffed 50% of the time (wasting money).

2. Should you have more doctors, or more APPs, and in what ratio?

APPs are less expensive than doctors, but need oversight and have a reputation for ordering too many tests (although there are plenty of doctors who are guilty of that as well).

3. Should you pick up the next new patient, or discharge one of the patients who is ready to go?

This is a fascinating question that requires knowing a few other variables to solve properly (Single coverage site? Acuity of the next patient? Other doctors twiddling their thumbs?)



Comments

Popular posts from this blog

Mastering Medicare Cheat Sheet

Amy Schiffman and I have created an awesome 1-page Medicare Cheat Sheet. You can get it for free by joining the Mastering Medicare subscriber list . Here's a sneak peak

How To Automatically Upload a Zoho File Attachment to Dropbox from Zoho Creator using Dropbox API v2 and Zoho postUrl

I'm an idiot, so this took about 6 hours for me to figure out, but apparently many people have not been able to figure this out, and I could not find a complete and working solution online anywhere. If you want to be able to upload a file to Dropbox from Zoho Creator, you have to create a custom function like below (here I called the function "Dropbox", under namespace "Create"). In the function, the ID is being passed to it from the underlying form entry. void Create.Dropbox(int id) { url = "https://api.dropboxapi.com/2/files/save_url"; headers = {"Authorization":"Bearer XXXXX","Content-Type":"application/json"}; data = {"path":"/FFFF/TTTT.XXX","url":"YYYY"}; datastring = data.toString(); response = postUrl(url,datastring,headers,false); } ... where XXXXX is your secret Dropbox Token, and YYYY is the url of the file you are uploading, and FFFF is the Dr

How to track CME - guest post published on Knack blog

I am happy to announce that my detailed post on how I built CMElog.org using Knack , Zapier, Dropbox, Mailchimp, Mandrill, and Google Drive has been published on the Knack blog: In this post, I dive into details of how to use Zapier to build automated services, like CMElog.org, in which medical providers' continuing medical education credits can be automatically extracted, identified, copied for back up, organized, and accessed.