Home » News » Currently Reading:

Being John Glaser 1/8/09

January 7, 2009 News 4 Comments

A byproduct of the implementation of an electronic health record is the creation of a relatively large clinical database. The core value of this data is supporting the provision of care. There are other uses — secondary uses — which are important.

Perhaps the most important secondary use is assessing the quality, efficiency, and safety of care. Care patterns can be contrasted with national guidelines; a physician’s practice can be compared to that of other physicians; and the health of a population of people with a chronic disease can be assessed.

However, several organizations have begun to look at additional secondary uses of this data. Two areas show great promise.


One area is post-market surveillance of medications. The graph above (Brownstein, PLoS ONE, 2007) was developed using data from Partners HealthCare. It shows a dramatic increase in admissions due to heart attack. The arrows indicate when our physicians began to prescribe Vioxx and when they stopped prescribing Vioxx. This example raises some interesting possibilities — could we begin to monitor a medication soon after it is introduced and do a much better job of detecting problems earlier?


Another area is leveraging EHR data for clinical research. An area Partners is working on centers on genome association studies (www.i2b2.org), e.g., are there genes associated with depression treatment success? The graph above (Kohane, Internal Partners Analysis, 2008) is a bit complicated, but it shows that studies that leverage EHR data (the lowest set of lines) can cost five times less than studies that rely on manual chart extraction (the top set of lines). In addition, it appears that EHR-based studies can be done in one-tenth the time. These gains in efficiency and speed could dramatically alter clinical research.z

There is still much work that remains, e.g., developing sound methods for dealing with the often poor quality of EHR data. However, both of these examples show very compelling potential secondary uses of clinical data.


John Glaser is vice president and CIO at Partners HealthCare System. He describes himself as an "irregular regular contributor" to HIStalk.

View/Print Text Only View/Print Text Only

HIStalk Featured Sponsors


Currently there are "4 comments" on this Article:

  1. John – do you believe that it would be wise for Government and industry to develop a simpler data model based on the Government-adopted HITSP standards, using deidentified data, to spur better research? By simpler data model, I would think of one that looks at limited demographics, basic encounter information (ex: dates, times, CPT, ICD, DRG codes), labs, medications, and allergies. Possibly problems, vitals, and social history could be added where providing organizations have data available that meets coding requirements.

  2. A great example of “assessing the quality, efficiency, and safety of care” is the NSQIP (National Surgical Quality Improvement Program) started by the VA, but now currently managed by the American College of Surgeons. Their results from 1991 – 2001 include:
    * 27% decline in post-operative mortality
    * 45% drop in post-operative morbidity
    * median post-operative length of stay fell from 9 to 4 days
    * patient satisfaction improved

  3. I think that there is value to a core data model that is consistently defined and implemented across all EHRs. This model would significantly ease research, post-market surveillance, quality measurement and public health.

    However we find that many of the data problems result from (a) gaps in data, e.g., we have data for 6 months then there is no data for a year and then there is data for 6 months – what happened during that year?, (b) data that is in the note (even though there is a place to record the structured data element – NLP is a major contributor to the efforts I described) and (c) incomplete data, e.g., not all of the medications or problems are recorded – this leads you to infer, for example, diabetes based on HgA1C and medication data.

    More structure would help. But we will always have this reality of messy, noisy clinical data.

  4. Some view transactional and translational repositories as separate entities with unique requirements.

    A transactional repository, or CDR, must accurately reproduce the clinical data in other systems and be able to retrieve data by patient. Accuracy is critical as patient care decisions are made based on patient data in the CDR. As the CDR is critical to a healthcare organization’s operation and is a “single point of failure” it must be available 24/7 at all cost. This translates into costs associated with redundant, mirrored backup and hot-site hosting.

    A translational repository must be able to retrieve data across many patients and may, using conventional databases, require a different database design from a CDR. A translational repository need not be overly concerned with accuracy as minor differences in data in a translational repository and the source system will be “noise” and not statistically significant. A translational repository does not carry the same uptime requirements as a CDR and, therefore, costs much less to operate.

    Operating both a CDR and a translational repository costs more than running a CDR and using it as both a transactional and translation repository. But new conceptual models are explored that eliminate the CDR, or virtual database , with web services used to retrieve data from source systems. In this model there is no CDR and only a translational repository.

    I am interested in your view of this discussion.

Subscribe to Updates



Text Ads

Report News and Rumors

No title

Anonymous online form
Rumor line: 801.HIT.NEWS



Founding Sponsors


Platinum Sponsors





























































Gold Sponsors
















Reader Comments

  • Sam Lawrence: Except in this case, coding = medical billing, not development. Though the same warning may be true...
  • BeenThere: Partners will find the savings from their cuts of coders as fools gold. There are a lot of hidden costs running an outs...
  • JC: If there is not there can be. VistA has a reference lab interface that can create the manifests/labeling and such as we...
  • Tom Cornwell: Great stuff from Dr. Jayne as usual. One small typo, last sentence of second-to-last paragraph: should be 'who's' not 'w...
  • HIT Observer: What I find most interesting here, is people defending their common practices rather than truly taking this as invaluabl...
  • Bob: There's no incentive for the provider to spend time doing a price comparison for the patient. Nor is it a good use of th...
  • Peppermint Patty: Veteran - can you clarify what was "fake "? Was something made up (definition of fake) or did you disagree with Vapo...
  • Pat Wolfram: Such a refreshing article. Thanks -- there really can be a simpler version of an acute HIT implementation. But I do ...
  • Woodstock Generation: Bravo to HIStalk's Weekender recaps and other news/opinions. I read it first thing on Monday mornings..................
  • Veteran: #fakenews...

Sponsor Quick Links