Data science / ML feature-as-a-service

[jrlegrand]

Website examples for Partner Connect:
https://www.johnsnowlabs.com/databricks/
https://learn.microsoft.com/en-us/azure/databricks/partners/ml/john-snow-labs

From Corinne Stroum:
OK, the TL;DR:
You have built a toolkit of great value to data analysts and data scientists. As they always say, we spend most of our time data wrangling.
The desire to go from "this was the medication that was administered/ordered/filled" to "this is what the condition treats" or "this is in X drug class" is important in healthcare ML
ex. "Has patient filled an rx for long-acting insulins in the past 90 days?" or "Has patient been treated with a DMARD in the last 180 days?" are super common ML features
even questions of "How expensive was this Rx at at the ASP level?" are important, as they influence purchasing habits agnostic of GPO practices (immunoglobulin is expensive whether you're looking at ASP or GPO cost, right?)
These cost tiers are important when evaluating change in protocol or prioritizing efforts (it's a balance between "how often is this used?" and "how expensive is it?"; common and high-cost rx are often targeted for evaluation.)
ASP also helps to normalize differences in cost across facilities within the same health system. Cost accounting systems sometimes bake in something as arbitrary as facility depreciation when determining cost of clinical events.
I recommend offering up your data/insights as a service - I showed you how other vendors like John Snow offer their clinical NLP and datasets as partners in Azure and Databricks.
Sample APIs might be focused on feature computation (and can be nicely devoid of exchanging PHI)
"Give me one/many RxNorm/NDC ID(s) and I will tell you what classes they represent"
"Give me one/many RxNorm/NDC ID(s) and I will tell you what conditions they may treat" (NDRFT I think was the vocab that has this)
Here's what this looks like for ML models, but you are doing it for features https://huggingface.co/inference-api
Simply setting up the ability for downstream customers to subscribe to a cleansed rx-focused DB on a daily basis would be big. Tons of startup health tech vendors would want/need this to support their work without the same capital expense as a big vendor

3:43
This is a big vendor that jumps to mind when I think about Feature Stores. They're obviously a bit more, as they do the logic and computation as well.. https://www.tecton.ai/product/