Hopefully before the end of the year, we should have the ability to spin up transient instances to test.
Nothing new that you aren’t already involved in has happened except we have the beginnings of packaging process for the EHR that will make it easier (down the road) to deploy new sites.
Re CTAKES and NLP, I provided 10 years worth of notes from Dr Bowen that need to be de-identified (using NLP), but the project is stalled at the moment. @hzbarcea started it, but had to do some major updates to CTAKES to make it a ‘service’ instead of a local client. This is one thing that could use funding, but would require a lot of it or a lot of smart volunteer effort.
None of that has anything to do with LH-Toolkit.
Understood, but it would be helpful to say that the Toolkit will be completed in x months.
Is there any reason I can’t personally de-identify e.g. 25 of Sam’s outpatient records? Are they in PDF format?
On another note, I think there needs to be one clinical report that combines the demographics, diagnoses, lab results and medications. There should only be one search button. As it stands the diagnosis can be inputted into the upper left box or the box above ICD10 which is highly confusing. When do you hit “refresh query” in the process? I just tried to write an electronic clinical quality report scenario of those patients aged 18-75 with hemoglobin A1cs over 9% (also known as glycohemoglobin). I could identify the 441 patients with type 2 diabetes in the Demographics vs. Diagnosis report and I could identify the 127 patients with hemoglobin A1cs over 9 but we don’t know if they are the same patients. There could be a bunch of people with undiagnosed diabetes. Needs to be one robust search engine.
Mixed stuff here
Tool Kit is completed has been for a long time, but it’s just that “a tool kit” for building something specific, so I’m not sure what you mean by that. Perhaps you and Saptarshi had some conversations about things to do with toolkit that I was not involved with.
The notes data are in a LibreEHR SQL data format. I could load them up to be edited using adminer, but it more complex that just deidentification. They then need to be connected to synthetic patients that have appropriate diagnoses, etc to be useful. The NLP project is more about the end-to-end process of converting free text notes to coded, discrete information and maintaining the connection to the original material
Let’s talk about enhancements to the reporting separately (not in this thread). I think some training is in order before we make any changes to the model.
I don’t get notifications, so I have to check for updates to reply.
The issue with cTAKES is that it relies on the concept of a pipeline. That means that it’s relatively easy to build a service with a fixed pipeline, but not very easy to build a multi-tenant version or something with a dynamic pipeline. That’s why cTAKES is currently used with relative success in research, but not much beyond that.
@tony is right that with some funding that could be addressed.
I was under the impression that the toolkit could lead to developing FHIR apps down the road but that might be my misunderstanding.
I’m still just looking for 25-50 outpatient encounter notes I can “marry up” with existing NHANES patients to make the student experience more realistic. I am aware that C-Takes will be a research tool only
My phone call today with Dr. Jennifer Croker at UAB (Center for Clinical and Translational Science) was highly successful. Her interest was in the big picture - how can a sandbox EHR help move students and faculty further along in terms of EHR training and data science. She saw great potential benefit. She plans to talk to more of her staff to gather their ideas. This might get us a new data set that includes patient encounters, who knows? Clearly, the more universities involved the better.
The phone call next week will be with the quality/safety manager at UNMC whose interest I believe is in using LibreHealth EHR for simulation scenarios. Not sure what to expect.
Texas A&M was super pleased with the 4 data scenarios I created using LibreHealth EHR reports but they won’t launch the course until next year.Data Exercises using LibreHealth EHR.docx (24.0 KB)
LibreEHR has a minimal FHIR implementation for Patient Demographics and Appointment booking that we (MI2) did for an provider network in Ireland. My understanding is that toolkit has a native FHIR implementation at some level, but I’m not sure how may “resources” it handles. There are 1000s documented.
PS: Re Data exercises… Test them in the LibreEHR NHanes demo. MSUD should really be left for Denver
https://nhanes.librehealth.io/ this uses the ‘default’
@rhoyt I’ll see if there is some easy way for me to get you the free text notes so you can, perhaps, map them to some of the synthetic NHanes patients get us closer.
The demo https://nhanes.librehealth.io seems to be corrupt. Says JSON can’t be parsed. Unfortunately, I have a teleconference today at 1 PM CST
Somebody seems to have hacked it.
Use the misquared site for today.
Same admin user/ passward as the demo
Thanks. Looked at the MSU demo and saw a bunch of GUI errors (see red arrows)
I fixed the site, @tony could you just send up the dynamic finder fix that I submitted that would help with this error. If you need me you should e-mail me directly email@example.com as I do not check this very often.
Thanks for the repair job. Not sure why one of the patients Debora Abbott is listed as an administrator and not sure why Administrator is duplicated “Administrator, Administrator” Happy Thanksgiving to all
Was it merged? I approved it, but it still needed a 2nd approval. If it’s ready to merge (or already merged) I can push it out easy…
I have not heard back from either UNMC or UAB regarding access to de-identified office notes for us to use. In the meantime, I saw that Amazon just launched a new and sophisticated NLP platform known as Amazon Comprehend Medical: https://aws.amazon.com/comprehend/medical/ It would be great to try this platform out as it hints that it is much faster and easier to use than current platforms like cTAKES. Thoughts?
I was listening to a talk about that today. It’s on my list to take a look at.
I would appreciate your opinion as they also have Docker services. Deploy Docker Containers
on Amazon Elastic Container Service (Amazon ECS)
Amazon Elastic Container Service (Amazon ECS) is the Amazon Web Service you use to run Docker applications on a scalable cluster. In this tutorial, you will learn how to run a Docker-enabled sample application on an Amazon ECS cluster behind a load balancer, test the sample application, and delete your resources to avoid charges.