AuxilioMD: Machine Learning and Data Visualization Platform

Introducing AuxilioMD

AuxilioMD is the last-mile solution for clinical researchers who want to validate, implement, and eventually sell their machine learning (ML) models for healthcare diagnostics.

For our first project, we developed a patent-pending, ML-enabled diagnostic model platform that can be integrated with any major electronic health record (EHR) to risk triage GI bleeding patients. We built out the front and back end alongside two physicians from YNHH, who published an original research paper that formed the basis for the model.

Now, we're making our project open source!

Link to our website: auxiliomd.com

Link to our codebase:

Y-Combinator Application (good breakdown of idea): YC Application - AuxilioMD

Why we built it

We think physicians should have access to the best tools to do their jobs.

Right now, there are hundreds of powerful AI models addressing a variety of healthcare challenges, but they're stuck in research papers. They need to be clinically validated, connect to the electronic health record system, and integrate into the physician workflow to make them useful in real practice.

Our team (five undergrads) came together through a club, where we were introduced to two physicians who had this exact problem. They had a published research paper showing their risk triage model for GI bleeding could keep patients from being improperly admitted to expensive overnight stays at the hospital - a shockingly common issue.

In terms of background, two of the five had software engineering experience, and the rest focused on product, business development, research (there was a lot of that!) and marketing/design. None of us had direct experience working in healthcare, but we all saw what an important issue this was to address:

Problem Breakdown

In simple terms, many physicians still make informed "guesstimate" decisions when making a diagnosis or when advising a course of treatment (ie admit to overnight stay or not). Sometimes these are "wrong", or suboptimal. For example, with GI bleeding many emergency room (ER) physicians will advise an overnight stay when the patient could be effectively treated at home with medication. This leads to hundreds of millions in dollars of overspending. Clearly, there's a lot of room to bring in more data analysis to the day-to-day work.

This isn't to discount the years of training and amazing work physicians do every day, but the quantitative data they could use to better inform and prove decisions is not always readily accessible on notoriously user-unfriendly EHRs. There's actually a suite of thousands of clinically validated "Clinical Decision Support" (CDS) tools, but there are two main problems with them:

CDS tools powered by AI can be far more accurate and precise, but it's very difficult to bring a model from experimentation into actual usage in the hospital. Frequently, physician researchers don't benefit from their discoveries.

To sum it up: