I absolutely think we need to reach out to WEKA for advice as this application has been integrated with other programs, but not an EHR. My weakness is the tech/IT aspect. I could find an extensive FAQ section on their wiki but no reference to integration help. Perhaps our more IT savvy colleagues can start a thread at: https://weka.wikispaces.com/ or join the WEKA mailing list to start a thread: https://list.waikato.ac.nz/mailman/listinfo/wekalist
By the way, I think our initial data mining goal would be primarily classification and regression. If we can find a way to make this as user friendly as possible that would be huge
The presentation created was heavily slanted towards education, so hopefully it would be accepted for presentation at InSpire 2017 in June.
I have completed revamped the document so it serves as a short overview of what we want to accomplish (or at least what might be ideal). Here is the link to Google Docs so please provide input so we can use this when we contact granting organizations and they want a short summary. So far, many orgs don’t support software enhancement, unfortunately. Let me know what you think.
I have not received any input so far regarding the short overview of the project I shared on Google Docs in my past post.
There is a research meeting next Tuesday at my university to discuss long term grants. The problem is that many of grants have due dates this month, which is too soon. Others, such as those by drug companies seem to want only clinical studies on diabetes and not the infrastructure to take care of them. I have sent feelers out to AHRQ and the STTR folks in Washington DC to see if they can help direct us.
In the meantime I hope the LibreHealth folks can move towards a statement of work and budget as I think phase I is very doable and we could start on that relatively soon. Should be an “easy sell” here to fund this early phase
@aethelwulffe and I can work on the SOW for phase 1 next week.
Phase I: Based on survey data, the project plans to customize the existing EHR by adding about 10,000 patients and their associated data obtained from the CDC National Health and Nutrition Examination Study (NHANES) program (2011-2012 time period).13 Sixty NHANES data tables have been identified that would be valuable for education and research. Data tables selected can be viewed through this link.14 Researchers have used WEKA to data mine NHANES data successfully.
Many thanks Tony. While I have feelers out to several organizations about possibly getting de-identified patient data, none have come through, so I think we should plan on using NHANES data. Keep in mind that there are hundreds of data points/tables in NHANES (in addition to the 60 I have identified) covering topics from the common to the rare. Maybe not in phase I, but down the road think about how we could create an educational and research EHR that could be customized (without too much effort) to allow universities to import additional NHANES information into the EHR for analytics. That being said are you ok with the notion that this project should be to customize LibreHealth EHR for both education and research?
Yes, in phase 1 we would need to design a layer (WEKA based?) that would import NHANES and Generated dummy data that could be extended to import “real” de identified data for other sources. Nate DiNiro at YouBase.io has a similar need with a different target output format (JSON) than LibreEHR, so perhaps we can use/design or leverage that as the intermediate data ETL layer.
Another way to think of this project is that the educational/research EHR that is populated with a lot of patient data could be considered the “sandbox” for the production EHR. In other words, we test the utility of the diabetes registry and machine learning tools in the educational/research EHR. These tools are perfected then eventually they become part of the production EHR. I don’t know if this is a better way to state our intent for grants organizations
Have not heard back from WEKA, but perhaps my question was too general. As we get closer the tech side should definitely ask the Wiki about integration issues. The Wiki link is here
Had a good meeting with our Research Sponsored Program Office yesterday:
We are on track for phase I. There should be roughly $20,000 to upload/map the 60 NHANES tables to LibreHealth and workout the rough edges with installation, etc. Tony is working on SOW, time line and so forth. I will rework our overview to serve as the grant application to UWF.
We did find a very interesting NIH grant PA-14-155 on early development of software initiatives like ours but we can’t make the deadline. We have a webinar with AHRQ HIT Grants section this Friday at 0930 CST. The purpose of the phone call is to request their help in long term funding for phase II. In addition they can give us a reality check on the concept and plan. I am under the impression that our development of an educational and research EHR could be considered our sandbox and when enhancements are added, vetted and tested by end users, they could be migrated to the production EHR. Just to reiterate this is a project with the players of IUPUI, UWF and Software Freedom Conservancy.
Wow, revisted this topic and saw that I had an incomplete response to Tony’s last.
Over the last four days I have been continuing to lay out a NHANES data import/export scheme (as code) on top of existing code in all that “spare time”. Naturally, I am, as usual, very interested in providing accompanying billing data, so I am working that in as well to produce the required EDI “history” files. That said, we actually started on this layer of work several years ago.
Tony’s weekend was heavily invested in the current EHR release, and he will be on a plane today, so we can’t expect much of a response on the completed POA/SOW from his side today. On my side, I am getting my tool to where it consumes data for everything from external data sources, and does it via a configuration instead of hard-coding. I just want you to have a proof-of-concept tool available that sucks in the favored data sources. You might not need such, but I follow follow the F.A.S.T. method, or “Fundamental Approach Sales Technique”
Thanks @rhoyt. I think the PA-14-155 was good, but Feb 5 was the last cycle of a PA thats been running from 2014. We should wait to see if it’ll be notified again. The AHRQ grants are probably the most relevant to us. Weka works well with Java natively and toolkit integration with it, might be the most efficient way to integrate.
I’ve not got a chance to look at the progress on the data model merging between toolkit and the EHR projects.
Sounds like we are all moving in about the same direction. I have to rely heavily on you all for the technical details and how this will fit with the LibreHealth initiative, toolkits and so forth. I’ll post again on Friday after the AHRQ meeting