Glass.Health: AI Diagnostics and Clinical Notes

Hello, and welcome back to the AI my clinic newsletter. Today, I am writing about Glass.Health and their AI-Powered Clinical Decision Support tool. I’ve been watching and reading a lot of Dr. Dereck Paul’s (CEO) social media output, and I have to say, the journey is quite fascinating - from medical school and residency to the Y Combinator cohort of Winter 2023. Will they be the SpaceX of medicine? Let's take a look, Part 1.

Subjective (an intro to Glass):

Glass.Health was founded in 2021 by Dr. Dereck Paul and Graham Ramsey. They teamed up while Dr. Paul was still in medical school at UCSF, with the goal of creating an online AI platform to store and organize medical information. I guess Dr. Paul had the same problem we all face in medical school - TOO much information, TOO many resources, TOO scattered. So the premise is great; create a central repository for medical information as you go through your training. They had a lot of traction off the bat, many doctors who loved their product, and were even able to raise capital. In 2022, they raised their first round from Breyer Capital and Y Combinator. It was so successful, Dr. Paul had to put his residency, Internal Medicine (IM) at Harvard, on hold to pursue this dream full time. Talk about dedication. 

And this is where things went for a crazy twist. Imaging being Dr. Paul, flipping gigs from an Internal Medicine first year resident (the consensus worst year of residency) to a venture capital backed entrepreneur that is ACCEPTED to Y combinator Winter Class of 2023, when suddenly…

ChatGPT is released

… and it virtually changes the internet as we know it. I can only imagine if you had an AI tech startup in that Y combinator class, you probably heard very quickly from your advisors and investors, especially since Sam Altman, the CEO of OpenAI, was formerly the president of Y combinator (Here’s a really cool video series by him on how start a start-up). So, the Glass Health team made a pivot and a pivotal moment in internet history. They built a second product called Glass.AI, a tool that can be used by physicians to create differentials based on clinical data and preliminary medical plans to improve patient care. So my part 1 of this review will focus on this product.

Objective (what the product actually is):

When you log into Glass.Health, you can select Glass.AI. A page comes up where you can insert a patient summary into a text box. 

Careful not to include any patient identifiers, I added a generalized description (with multiple variations) of a case I encountered in the hospital. Here is the case:

Glass AI then has two options: Draft Differential and Draft Clinical Plan. For the drafting differential, the most likely differential was:

It also included other differential diagnosis including, zenkers diverticulum, achalasia, myasthenia gravis, bulbar palsy, ALS, stroke, esophageal cancers, esophageal stricture, GERD, Laryngeal Cancer, neurological disorders with dysphagia, medications induced esophageal disorder.

For Draft Clinical Plan, the algorithm succinctly summarizes my patient data,  provides an assessment for why aspiration pneumonia is their most likely clinical diagnosis, and writes a list of other diagnoses it could be. Then, it gives us diagnostic studies to consider ordering for our patient:

And a treatment plan:

Well done.

Assessment (what I think of the tool):

In evaluating Glass Health's Glass.AI, I focused on two key features: the Draft Differential and the Draft Clinical Plan. My test case revolved around a 68-year-old woman experiencing difficulty swallowing, suspected of having aspiration pneumonia. Impressively, the Draft Differential correctly identified aspiration pneumonia as the most probable diagnosis. This alignment with my clinical assessment was reassuring. The tool also expanded my perspective by suggesting other potential diagnoses, such as achalasia—particularly relevant given the patient's history of breast cancer and radiation treatment—and less likely conditions like myasthenia gravis, ALS, and bulbar palsy (which admittedly, I did not even consider). The detailed rationales provided for each of these differentials was helpful and through provoking as well.

However, one aspect I found lacking was the inclusion of serious red flag pathologies in the differential, especially since shortness of breath and orthopnea could be indicative of heart failure or primary lung disease. Inclusion of these conditions would ensure that nothing big is missed, even if the suspicion is low.

Regarding the Draft Clinical Plan, I was impressed by the range of suggested diagnostic studies and the clear explanations justifying each. This feature not only helped in narrowing down the differential but also in guiding the clinical decision-making process. Nevertheless, I believe the tool could be enhanced by prioritizing the suggested tests, streamlining the workflow for busy clinicians. For example, simpler tests like blood cultures, sputum cultures, and neurological exams could be prioritized over more complex studies like pulmonary function tests or fiber optic endoscopic evaluations of swallowing.

And most importantly for me, the intern resident, the clinical plan is not exactly something I can copy paste into the electronic health record (EHR). While I understand this product is more of a guide than a dot phrase (a function in EHRs that allow faster note writing), adding a layer of organization to allow for copy pasting would greatly increase my usage of the product. For example, grouping treatments based on individual problems or system-based categories and separating analysis from the plan would make for easier copy-paste interface.

Plan (do I plan on using the product?):

My first experience with Glass Health was positive. By inputting a history of present illness and with just a few clicks of a button (a significant improvement over the hundreds of clicks required in EHRs), I was able to generate a thorough differential, diagnostic plan, and treatment plan. I do like this tool, but currently, I see it more as an educational supplement to fill in the gaps of my differential. It is not yet at a stage where I could copy and paste this clinical plan without significant and time-consuming edits. That being said, having listened to Dr. Derek Paul's multiple interviews, I realize that this tool is intended to supplement my clinical assessment as a physician. AI will not replace me… Darn. However, I do see AI streamlining note creation, requiring less extensive adjustment before publishing to EHRs. I would bet that Glass Health is in a position to make that happen, and I believe in Dr. Paul's direction and momentum. Exciting stuff.

If you have any thoughts, or if you have used this product before, please reply to this email and let me know how it has helped you. In Part 2 of my Glass Health series, I will be analyzing their first product, The Notebook, which promises to revolutionize medical knowledge storage and education. The future is bright.

Hope to hear from you soon,

-Maxime Rappaport, MD