AI in production: Carta Healthcare
This is one of the articles about "A.I. in production." It contains a story about a real company using A.I. in their products or building an MLOps tool.
This is NOT a sponsored content. I don't endorse the interviewed company and its products in any way. They don't endorse me either.
Last week, I interviewed David Scheinker, PhD, Director of Systems Design & Collaborative Research at Stanford Lucille Packard Children’s Hospital and Joh Olson, Healthcare Management Advisor. They are advisors working at Carta Healthcare - a startup whose mission is to automate and simplify the work that causes burn out in the clinical staff, allowing them to focus on patient care.
Imagine that you are a nurse at a hospital. You work with a team of doctors who want to give a particular medication to the patient, but first, you must check whether the patient is allergic to the drug they are about to use. Easy, right? Not quite. All you have is the patient’s medical record and your knowledge.
You must read the entire drug log and the medical record to check if there is any information about the symptoms getting worse and correlate it with the drug log. You know what, it would be cool if you could do it fast because the patient needs the drug now. There are twelve more people whose medical records you must read, so be fast and try not to make any mistakes. Do you still think it is an easy job?
Good luck with that. Fortunately, Carta Healthcare wants to speed up the process by using machine learning to extract information from medical records and presenting the relevant data to the nurse. Because of their software, the nurse can spend more time applying their knowledge and not waste time looking for the data they need.
According to Carta Healthcare, their data extraction solution makes the nurse ten times more efficient because a machine learning algorithm does the tedious and cumbersome part of the work, and the nurse needs only to verify the results.
Once you have a solution capable of extracting useful data from medical records, what else would you do? Carta Healthcare created software used to submit data to clinical registries. They used a closely related function of their software to predict the surgical supplies that will be used the next time a particular case is performed.
They wanted to solve three problems. First, hospitals buy many supplies that are not used. Second, hospitals sometimes fail to buy supplies they need because they aren’t listed. Third, expensive supplies are used because of a lack of information or a surgeon’s personal preferences.
Carta Healthcare created a recommendation system for hospital supply management that tracks supplies usage over-time and the quantity used per surgery. They have built a time series based regression model for predicting usage of every tool or piece of equipment. As a consequence, they are capable of suggesting supply substitutes or detect that surgeons can perform surgery without using a particular item.
According to Joh Olson and David Scheinker, the most challenging part of their recommendation system was cleaning and merging the data they extracted from preference cards (a preference card is a list of items required during surgery) and other parts of the hospital electronic health record. It was problematic because modern electronic health records are a hodgepodge of individual functions designed for billing separately across departments, not for efficient operations or data tracking across the institution.
Once they managed to clean the data, the problem got significantly simpler because they could run the regression for every item in the inventory.
Joh Olson and David Scheinker claim that their recommendation system prevents biases such as the recency bias and helps to make supply related decisions that are data-driven instead of resorting to decisions based on personal preferences. It saves nurses time observing cases and reading through cards.
When I asked about book recommendations, Joh Olsen recommended the book “Five dysfunctions of a Team” by Patrick Lencioni. It is a book that describes the five issues that make teams unproductive and hostile towards one another. David Scheinker recommended a short story, “Mister Squishy,” written by David Foster Wallace. It is an off-beat story that uses an absurd situation to give insight into corporate marketing and corporate culture.
Did you enjoy reading this article?
Would you like to learn more about software craft in data engineering and MLOps?
Subscribe to the newsletter or add this blog to your RSS reader (does anyone still use them?) to get a notification when I publish a new essay!
You may also like