Heh…tried to indicate that it isn’t impossible, but there is a LOT of little things…and now my “free SAS viewer” is going all flaky on me. Anyone got a non-sas, non-R source of everything relevant?
It seems to me that periodicity of the data (dates of service etc…) is an important factor in profiling a patient.
I do think we will end up using NHANES, partly because it would look good for govt grants and the data dictionaries are very good. We might not get any good documentation if we purchased data sets from Marshfield Clinic, but I will keep an open mind. I added ICD-10 codes for the common conditions noted in the Excel spreadsheet I circulated. For the lab work we could easily add LOINC codes. As we don’t have any true encounter (patient visits) we won’t need E/M codes yet
One other minor point about coding patients. We could also use medications to tentatively code patients. In other words, if they are on drugs for diabetes and there are no other reasons to be on these drugs it is probably safe to label them diabetics
FYI. I have a teleconference with the Medical Algorithm folks this Thursday at 4 PM CST. This will be just a general info session to see if there is any possibility of future collaboration once the educational EHR has been built. I have no idea as to pricing and whether we are able to use the API successfully. If anyone wants to join the session, let me know and I will provide the login information
I’ve been out of town. I’ll work on the SOW in the next day or two but I need a very specific list of what we plan on doing, seems to be a bit of spinning has happened, so I just want to be clear.
I’m probably the most guilty of spinning but while we are waiting to build the educational/research EHR I am looking at future possibilities once it is built. I should hear something in the next week from the Marshfield Clinic to see if they can offer patient data that is superior to NHANES data and if it is affordable. Regardless, the overarching goals seem to be to ETL the 60 or so data tables selected from the 2011-2012 NHANES data set to LibreHealth EHR. We would benefit from the demographic generator Art created to add realistic names, etc. to the patient files.
We should discuss what we plan to do with the final product. Should we solicit universities, such as UWF and IUPUI to use the educational EHR for upcoming classes so scenarios and case studies can be created and we can gain valuable input? Seems like we should do that before we simply release the software to the public.
Can some part of this be done by a GSoC student? The NHANES data import project, atleast seems like something a student can spend the summer doing, which will be useful to a lot of people.
I would like to offer my services as the “vendor” in providing tools for managing NHANES data in LibreEHR, with the goal of producing a useful platform for both educational and research purposes. I will present a statement of work shortly.
We might find something suitable for a GSoC student in the future, but don’t think we should worry about that until we at least have an initial version. Having something tangible to present will help drive discussion on many fronts.
There probably won’t ever be a “final product.” The data set and tools should evolve as new needs are identified, and additional input from academic institutions will be valuable. With the initial version we can host an instance in the cloud on Amazon AWS or similar service for UWF, IUPUI and others to review.
It’s clear that NHANES provides a useful baseline dataset. It’s biggest limitation is that each survey response represents a snapshot of the patient at only a single point and doesn’t provide indications of how individuals’ state changes. Unfortunately, the SEQN between NHANES datasets get recycled between years and don’t indicate the same individual. (e.g. in the 2011-12 set SEQN#1 is a 22 year-old, and in 2013-14 SEQN#1 is a 69 year-old).
However, for teaching purposes, there are a number of approaches that could be used to overcome that problem. For example, we could select a handful of “interesting” patients (find ones in NHANES with the “most diagnoses” or “biggest list of medications”) and then manually create progress notes/lab results documenting care of those interesting patients. Another possibility would be to see if we can identify survey responses between the NHANES datasets that are “likely to be the same person” (e.g. see if we can find a 66-68 year old in the 2011-12 data set that looks similar to the SEQN#1 record in the 2013-14 dataset and just assume that it’s the same person.) This wouldn’t be valid from an analytics standpoint, but still could be useful for demonstration purposes.
I’m in complete agreement with what you have stated, except I don’t think any of the NHANES survey time periods includes the same patients. I believe they are all unique.
Clearly, I agree that this will be a constantly changing process to create, enhance and maintain an educational and research EHR. As we gain more experience and input we can tweak according. Ongoing grant money would be important to support the “vendor”
I have already paid for web hosting for a platform I am not using so I am assuming we could use that to host multiple instances. Happy to pay for this type of thing if it would be useful
For development and demonstration purposes the project (librehealth.io) has plenty resources from AWS to proceed. Part of the project (for Kevin’s benefit) is to create a deployment package that the Universities can use and reuse to install the tool set and EHR and to reset it between “classes”.
LibreSOW.pdf (185.6 KB) I have put together a proposed statement of work for review.
I am requesting $5000 for roughly 50 hours of work to be done over a period of roughly 4 weeks.
Looks great to me so will send over to our Research Section for funding, in a timely manner
I’m very excited about the plan laid out by Kevin Yeh.
While the educational and research model is being built I thought we should continue our search for serious long term funding. Today I located a NSF grant nsf.gov-Software Infrastructure for Sustained Innovation SI2 SSE amp SSI Software Elements and Frameworks.pdf (795.4 KB) for software infrastructure innovation. There is an educational sub-domain that lists the following “The Directorate for Education and Human Resources (EHR) is interested in proposals that focus on innovative software infrastructure that supports the directorate’s research areas, namely STEM learning and learning environments, STEM workforce development, and broadening participation in STEM. For example, EHR is interested in research studies on how software tools foster science and engineering learning”
I would like Sunbiz to look at this and see if it seems attractive. The proposal would be due September 2017. There is a subdomain leader and we could run our project by him and see if he thinks we stand a reasonable chance of funding.
Glad to be moving forward on this. My SOW promises status reports to the LibreHealth Community. When I’m ready for the first one, I’ll create a new thread specific to the data project.
While we are waiting for phase I to begin, I wanted others to know that a friend of mine located [WEKA (analytics) for web services] http://gridlab.dimes.unical.it/weka4ws/howtos/installation/, created by an Italian university several years ago. I’m not making any recommendations, just noting that the components are freely downloadable so might be worthy of discussion down the road
Marshfield Clinic emailed me today to let me know that they could supply many of the data fields we planned to use from our NHANES data, but not all fields. In addition, while they could supply encounter notes in a PDF version, they had to be individually “redacted” which would be expensive and slow. Even though they did not include a price, I told them thanks but no thanks.
I sent an email to our research folks about the status of funding Kevin Yeh, but they are on Spring Break so might not hear anything until next week
Thanks for the update. I’ve been making steady progress on the NHANES import process.
Currently have scripts which can download ALL of the XPT files for a given year from the CDC website and convert them to CSV using R. The are roughly 120 files in the full NHANES survery, the fields of interest from the initial document are spread across 23 of them.
It also analyzes the corresponding “.DOC” files to look at what fields are present in each “.XPT”. .There are roughly 3100 fields in the full NHANES survey. I will provide a MySQL dump or an excel file with the variables and descriptions if anybody wants to peruse the full list. It’s much easier than trying to look at the webpages on the CDC website one at a time.
Next steps are to handle the import process into LibreEHR.
Will hold off on creating the “new thread” for updates at the moment.
I guess I need to get off my tuckus, now that I’m back from Ireland and work up a SOW for the infrastructure part of this gig.
Also we need to discuss the process by which the money moves. I think LibreHealth should be the prime contractor and distribution method. That way the non-profit aspect can be maintained for the contributor and Librehealth hires the development resource. Just my thoughts…
We are very fortunate to have Kevin Yeh on this project. I am incredibly frustrated with my own university because of their stalling to pay the $5K. The money is part of a larger fund of money I personally gave to the university to promote data science. I have been bugging them twice a week and while they have reorganized their research department there is no excuse for their inaction. I am inclined to just pay the money out of textbook sales this week (as a contract) and forget the university. Kevin has started work and SHOULD BE PAID ASAP. What should be the mechanism?
I’ll have another forum posting shortly because there is another similar project out of Canada and there is the tEMR project by Blaine Takesue at Regenstrief. I’ll listen into a demo today but this is EHR training for clinicians so different from our project.
Looks like there will be a panel on educational EHRs at the AMIA inSpire conference in June and I will be on the panel. A great time to be discussing this. Don’t know if either Kevin or Subiz will be attending. More to follow
Will discuss this tomorrow with @downey at our weekly meeting. Nothing in stone at this point, so we can make sure things get handle to everyone’s needs.