-
The teleconference with AHRQ HIT section (Mr. Wyatt) went reasonably well but they stressed they would like to see a research question. They were aware that there is a gap when it comes to students and instructors having access to a hands on EHR. I have asked my library science folks here to do a new lit search to confirm there is this gap across nursing, pharmacy, HIM, etc. I think we have already confirmed the gaps with HI instructors in the survey I sent. Please put on your thinking caps regarding research questions that could be associated with outcome measures. We plan to look at funded AHRQ grants to see what has been funded and plan to talk to an Educator PhD later today about questions. Perhaps the sequence of events needs to be building the educational and research EHR first with testing and some type of outcome data. We might have to host this for x number of universities to try in actual courses and study some type of outcome. In spite of my severe impatience disorder, we might have to wait and get a second grant to integrate machine learning. Is the diabetes registry something we need a grant for or just add it??
-
This week saw a ârequest for inputâ from NSF regarding innovative things to do with government data. I thought our use of NHANES data was innovative so sent in our proposal. Ironically, there is a $50K award for those they like. A newer proposal is now on Google Drive.
Mr. Bob, I have research questions: Is it possible to achieve meaningful improvement of clinical documentation practices and patient care through the use of a fully functioning EHR? Is it possible to train a new generation (or old) of clinicians and administrative staff to use an EHR system and achieve a net benefit for the practice and patients? Is it possible to significantly advance clinical and syndromic research methodology and drive EHR innovation through use of a tool that allows the researcher to experience the data in the same way that the clinicians do?
-Yes, I know these are strange, but the first question a Woods Hole research vessel answers is âDo boats float?â
There is good evidence that EHRs improvement documentation quality but at the price of speed and workflow (but highly dependent on the practice)
It is possible to train staff to use an EHR and achieve a net benefit to a practice but everyone needs to be on board. Very few practices would have done it without HITECT Act support and the fear of penalties if they didnât.
Iâm not entirely sure where you are going with the question about research methodology but I do believe it would be important to allow researchers to âget under the hoodâ with an open source EHR so they could actually access data tables and analyze the data. Almost all analytics is with very large EHRs that daily transmit data to an enterprise data warehouse
After looking at all funded grants by the AHRQ HIT group I have some additional thoughts. When you get a chance please look at this summary on [Google Docs](https://docs.google.com/document/d/17ha_DUhcroaRgRKKTknT7wlSPwWl3r7TxiQrxPmFDT8/edit?usp=sharing)
I was unable to comment directly on that document (view only) ⌠so
- One possible path is to use existing funds to populate LibreHealth EHR with about 10,000 patients and work out any issues of installation or use. â I definitely think we should start here and get this out of the box
Iâm not sure how the diabetes registry âintegrationâ is different from what LibreEHR already does at the clinical level, perhaps you can expand on that? What does the registry provide to the patient or the doctor that would help? the EHR already has the ability to record the diagnosis and have rules created to trigger recommended actions on the part of the clinic.
I think a project around using WEKA (as in http://community.pentaho.com/projects/data-mining/) would be valuable. The question is are we talking about using it for One Clinic to datamine their own data or adding some kind of managed collection of de-identified data collected from multiple LibreEHR and possibly OpenEMR/OpenMRS sources?
I changed the link so everyone can edit/comment. I certainly agree that we should move forward with importing the patient tables from NHANES with existing monies. The big question is what would be the next step and therefore the direction we would go for funding. Predictive analytics/data mining is a very hot topic in healthcare so if we could integrate WEKA and test it using the NHANES data (but add more patients). I think this would be the testbed before you would every release it to the production EHR. As I hinted in my note, what is the research question and which faculty might test it for us for usability. Several of us could create machine learning scenarios or case studies based on the NHANES data.
You are asking good questions about the registry. The registry is set up based on the American Diabetes Association with all of the annual requirements embedded. The lab work is color coded and it has the ability to generate patient level and clinic level reports. In a prior note I provided the link to the article we wrote on the registry, including the technical side and where you could find the components on GitHub: http://perspectives.ahima.org/developmentofawebbasedregistry/
I donât see anything wrong with putting the registry on the back burner. It was just a thought because we created it very recently, used open source components and will now roll it out to several free clinics in Pensacola.
I hope we can get some input about the machine learning aspect from Sunbiz and Dallas Snider, a computer science at UWF
I took a look at the link with my sister who is a Diabetic and Diabetic, public Health Educator. The data entry forms could easily be absorbed by LibreEHR. It might be good to send that data to the tool for the reporting (at least in the short run)
Another option is this kind of thing: http://hsc.unm.edu/community/chwi/pathways/program-model.html which is much farther up the chain of care coordination.
Dr. Hoyt, Tony knows me, but you and I have never met.Iâm starting to ramp up on this topic.
Starting to familiarize myself with some of the terminology in the thread and also plan to find out more about the content and format of the NHANES dataset.
My interpretation of âintegrating the registryâ with LibreEHR would mean standardizing the forms/collection of diabetes patient information so that all the sites are providing the info consistently.
As an example, there was a request in the forums a long time ago to add âblood sugarâ as an additional field on vitals, which was probably easier for that particular siteâs workflow, but not consistent with how other locations are likely to track it.
Glad to meet you. Having a diabetes registry is a nice EHR feature and I suspect you can simply customize LibreHealth EHR and accomplish the same goals we did with our open source project. Would be great to find funds to do it but it is highly likely we would have to do a clinical study to prove benefit.
Sunbiz sent me an educational grant I am still looking at. Seems like all of our energy right now should be in creating an educational and research EHR version. In the background we can explore how we might integrate the EHR with machine learning, given its Java API and so forth
By way of introduction to everyone Dr Yeh is a long time contributor to OpenEMR and has worked with me on many advanced projects including the new tab based UI and was our Subject Matter expert for the Peace Corps project. Welcome to the forum @yehster
We have missed youâŚWell I have certainly missed you. At everyone else: Kevin is one of my heroes, and I ainât got that many.
Dr. Hoyt, A quote from your paper:
âSeveral noncommercial diabetes registries that could be used by free clinics are available. The Diabetes Collaborative Registry is a global cross-disciplinary registry endorsed by several medical specialty organizations and sponsored by two pharmaceutical companies.17 The registry was designed to adhere to the Physician Quality Reporting System (PQRS) program so that physicians could be reimbursed for thirteen quality measures, reported via a registry.18 Free clinic physicians are not eligible for reimbursement through PQRS because they do not bill Medicare.19â
A few questions from me: Even though free clinics arenât eligible for reimbursement, are there any âabsoluteâ reasons why a free clinic couldnât participate?
What if LibreEHR integrated with the Diabetes Collaborative Registry so that both free clinics and clinics that accept Medicare could submit data with LibreEHR?
Does it make sense to analyze data collected from your Web Based diabetes registry at this point?
It might be interesting to compare the patient population of Our Lady of Angels St. Joseph Medical Clinic with the NHANES population.
On the more general topic of Informatics Education: Are there specific âconcepts/lessonsâ that ought to be taught?
For diabetes as an example, there could be a lesson which presents a few patients in the system, such as a ânewly diagnosed patient,â âone with well managed blood sugars,â and âone with poorly managed blood sugars.â Could explain to the student a little about the pathophysiology so that the lab results arenât just numbers. At the same time they could see the underlying data structures that are being used to track the data. As an exercise a student could then create a âclinical ruleâ to flag patients of concern.
Another approach to educational patients records, might be to just enter a set of interesting conditions, perhaps use a USMLE Step 2 Clinical Skills study guide as a reference.
Along similar lines for brainstorming:
Whoâs the student who benefits from the educational content we want to create? Possible answers:
- Software engineers who needs to learn more about medicine.
- Clinicians who need to learn more about technology
- Students in an academic setting who are just starting to get their feet wet.
My proposed âlessonâ on diabetes earlier would be most interesting for Student 1 or 3, but not necessarily as interesting for the Endocrinologist looking to get an EHR into the office.
I think the advantage of the DCR is that the registry can compare different doctors in the practice and compare them to a national average. This is valuable if you get reimbursed for quality measures related to diabetes. The registry we created offers more features but does not compare physicians and doesnât compare them to a national average.
It is âbackwardsâ to have a free standing registry, but most free clinics donât have EHRs. Sadly, because most free clinics are staffed by âolderâ physicians they donât want to use an EHR. That being said, I think the goal is to have the registry as part of an EHR and not free standing. I may be wrong but Iâm assuming there is no way to automatically populate DCR from any EHR. It sounds like a manual process.
I have discussed integrating the diabetes registry created for St. Josephs with WEKA so we can perform some basic analytics, but we havenât done it so far. NHANES patients represent a pretty good cross section of Americans. Free clinics are largely inner city poor and minorities. They are sicker.
The lab work in the registry we created is color coded to show what is within standards and what is out. I supposed it could be linked to pop ups that quote the American Diabetes Association.
I think an educational and research EHR could be helpful for a wide variety of students (with or without the registry). Medical students, nurses, pharmacists, HIM and Health Informatics students. They can generate reports, perform analytics, evaluate EHR usability and perform canned exercises
FYI. About 3 weeks ago I saw a study by Vanderbilt that analyzed patient data they got from the Marshfield Clinic Research section. I emailed them to ask whether they would share de-identified patient data. Yesterday they finally answered and wanted more details. I sent them the tables from NHANES and asked what the charge would be to basically provide the same type of data we plan to get from NHANES.
The advantage of the Marshfield data might be that it would include encounters and xray (imaging) reports, which are not available with NHANES. Have no idea as to cost but will share with the group as soon as I hear something
Intuitively weâd expect the free clinic population to be sicker, but it could be helpful to verify objectively.
- Are A1c levels statistically higher in this population, initially and after follow up?
- Is prevalence of neuropathy and decreased kidney function higher?
- Do therapy choices differ? Possible problems with access to meds?
- Follow up rates/compliance?
The NHANES data is provided in SAS format since thatâs a standard tool for doing statistical analysis, it might make sense to just try and get the St. Josephâs data into SAS. Basic statistical analysis seems like it could provide some useful answers before proceeding with WEKA and âmachine learning.â A later research goal might be to see if WEKA can identify relationships, both ones th> I may be wrong but Iâm assuming there is no way to automatically populate DCR from any EHR. It sounds like a manual process.at are expected (discovered through âmanual techniqueâ) and novel ones that we didnât consider.
We could also just try to do subset analysis of the NHANES data itself.
https://www.ncdr.com/WebNCDR/Diabetes/getstarted/datacollection We probably need to get more info from the DCR folks as to what their process is for submission certification and figure out if itâs something to pursue. Who knows, maybe we could get funding from AstraZeneca.
A tangential but related topic. Are there âsmart glucose metersâ? e.g. ones that keep a meaningful history, and possibly share the data automatically? I see that there are ones that integrate with iPhones, but if there were ones that didnât require the extra expense of the iPhone, it would be interesting to provide them to lower income patients and see if there is demonstrable improvement.
Incidentally, Iâm still thinking about getting âgood recordsâ into LibreEHR for more general academic purposes in addition to these thoughts about analysis.
SAS is of course very expensive, looking into cheaper/open source alternatives. R might fit the bill.
There are very few published papers from free clinics. What I can tell you at this point is that the diabetic control is similar to other clinics. In the case of St. Josephâs clinic most of the newer diabetic medications are donated free to the clinic. Many patients are well controlled on 4-5 medications that they could not afford on their own. These patients care closely followed and the care may actually be better than an average low income urban or rural clinic.
What I found on the DCR web site was a 7 page paper form for every patient. This means hiring a full time (FTE) employee to input and maintain a registry. This would not be the case if the registry was part of the EHR and auto-populated.
I reviewed every drug company that sells diabetes drugs and offers grants. I was either turned down via email or on their web site it states they donât support tech projects. They want clinical studies that show improved outcomes.
There are smart glucose apps but they port the info to specific web sites. Keep in mind that diabetes patients need more than a dozen tests and exams every year as standard care so they need 3-4 visits each year as a minimum.
After looking at the statistical packages vs. machine learning with WEKA, I have to say that I think the latter has many more features and has a less steep learning curve. Both offer decision trees and linear and logistic regression. Machine learning always generates ROC/AUC data and confusion matrices that can be used for comparison and publication.
I donât know if I mentioned this but Health Catalyst, a major health data analytics company just launched a new web site http://healthcare.ai/ that is an open source think tank/blog with webinars, mainly based on R programming. From what I can tell they have boiler plate R script for common functions to save time. They are brand new so canât say much more than that. Their company also using Python and WEKA.
The other reason I am pushing WEKA is that it is validated and open source with a Java API