Monday, April 13, 2009

The Limitations of Administrative Data

The data issue described in the Boston Globe this morning is really unrelated to Google, Microsoft, or any Personal Health Record (PHR) provider. In the US, there are two kinds of healthcare data - clinical data in Electronic Health Records (EHRs) and administrative billing data that is used by payers, researchers, and the government. Billing data is imprecise, but it is a starting point to describe the care given by a doctor or hospital. The only thing that's new in 2009 is that PHRs now enable patients to see the kind of billing data that's been used for 20 years for reimbursement, quality measurement, and population health. Blue Cross of Massachusetts and Medicare (in a few pilot states) share billing data with patients via Google Health, so this is not just a BIDMC implementation.

As a society we're likely to see increased data transparency between patients and providers, which will lead to several improvements:

1. Doctors will likely begin using more structured problem lists based on SNOMED-CT, a standardized clinical vocabulary of symptoms and conditions. This will enable their EHRs to better share data with PHRs as well as to more accurately measure quality. The Healthcare Information Technology Standards Panel (HITSP) has harmonized the national standards needed to reduce the dependency on billing data for PHRs, quality measurement, and population health.

2. Eventually, billing data will become more detailed as ICD-10 replaces ICD-9 billing codes in 2013. It will take several years for ICD-10 to be widely adopted and improve data granularity.

3. In the future, patients and doctors will work together to ensure records are up to date and accurate. It's a shared responsibility. Now that the Stimulus Bill requires doctors to make records available electronically to patients, the limitations of billing data will become more widely understood.

In the meantime, BIDMC will take the following actions to accelerate this work:

1. I'm working with the National Library of Medicine to map the most common Problem List terms used at BIDMC to SNOMED-CT, enabling BIDMC to use a clinical vocabulary and not just a billing vocabulary.

2. I'm working with Google to evaluate the impact of sending our existing free text problem lists instead of billing codes. It will reduce the number of features available to patients, since Google's educational materials are based on billing codes, but it may be more informative to patients to see the text their clinician wrote, not the diagnosis on the bill. Showing problem lists is what we've done in Patientsite for 10 years.

3. We'll hold a conference call with e-Patient Dave, his doctor, Google, and me to review Dave's clinical and administrative data (with his permission), to capture a real world example of the differences between these data sources.

24 comments:

  1. Dear Dr. Halamka,

    There is one little detail that I just cannot understand. The cloud based PHR have been in existence for a while. e-Patient Dave told his story. But what has happened with all the other patients from your institution that transfered their data to GoogleHealth or Microsoft HealthVault, prior to Dave. It is inconceivable that no other problems have surfaced until now.

    So, is Dave the first:
    * person to do the transfer,
    * to notice inconsistencies or
    * to just pay attention?

    ReplyDelete
  2. That's a great response and great you took the time to write it up and post so others don't just see this as a negative, which as a society we seem to thrive on what might be called dirty laundry, but it's not, it is in the "rinse" cycle:)

    Using SNOMED to map is a great idea and when people test the water, which is what we want to happen, things get better with improvement. I can't think of a time where I ever wrote perfect code the first time through.

    When you stop and think of what we had before hand, which was nothing, great strides have been made and will continue forward.

    Good stuff!

    ReplyDelete
  3. John - Right on target. Thanks for your responsiveness. As to your earlier post about meaningful use and quality measures, I'd like your comment on this observation. It appears to me that the quality indicators in use by MAeHC are HEDIS measures. What is the incremental value of collecting these indicators through the physician EMR versus the health plans' data warehouses?

    ReplyDelete
  4. John, thanks for responding, and thanks for our numerous exchanges before I wrote my post that started all this.

    I have to say, I cannot fathom how you can say with a straight face that the data PatientSite transmitted was better than nothing. As the Globe article makes clear, what arrived in Google was not only full of erroneous statements, it did not include the lab history and all the other things I thought would arrive there. Plus, as I wrote in my post, it didn't even include my medication history, which IS in PatientSite.

    From my perspective, as someone who thought he'd end up with a good set of data in a common format, what PatientSite transmitted was indeed worse than nothing. It raised so many questions that I had to comb through it, item by item, which means I might as well not have even bothered.

    You are right to start detailing the different data vocabularies that are available for medical information. Last night I posted a longer list on the e-patient blog.

    Does ICD10 have the ability to encode the fact that a condition was being tested for, as opposed to the condition being a diagnosis? (I wrote about this issue in my post on vocabularies, linked above.) If not, please don't even think of using it to represent clinical history.

    Re the issue of automatically linking to Google's reference information: with respect, I have to say it is freaking useless to link me to a completely generic article about migraines when I don't have them. Please, really, seriously, think through the whole use case.

    As you know, I'm happy as always to continue the work of creating the data e-patients need in order to be responsible for their care. But I urge that when you and Google continue working on it, this time you involve patients and their advocates who can scrutinize the results. That would be sensible, since we (the patients) are the ones who'll use the data.

    ReplyDelete
  5. No kidding - the limits. Six degrees of separation is more like it. 1. The US took a system which was largely created for mortality reporting and made a clinical modification so it would be useful for morbidity too. 2. It is required to be reported for billing. 3. Payment policy is designed around the codes. 4. True use of the codes changes dependent upon the payment policies of the insurance company (not supposed to happen under HIPAA, but check with your billing office). 5. ICD-9 is now over 30 years old because ICD-10 implementation, which would help, was delayed for a decade because the transition would be too hard. 6. Now its supposed to somehow be used for true clinical condition feedback to the patients.

    It's like we are trying to use a PC with 64MB of memory to support the EHR. It just won't work very well.

    ReplyDelete
  6. I'm not an expert in lots of stuff, but when it comes to ICD-9 and SNOMED encoding in the real world I'm about as good as you'll get.

    There's no way Dave's data is unusual. What's unusual is that he noticed and he complained.

    Also, John lets Google Health off too easily. The problems with administrative code data are as old as the hills. We considered all those in the 90s and decided to avoid populating clinical problem lists from admin sources. The non-dated aspect is a problem.

    Lastly, SNOMED is indeed the preeminent source today for diagnoses and findings. However that's only about 1/5 of the SNOMED entities. It includes a much larger data set related to medicine, medical science, pharmacy UK), dentistry and even veterinary medicine.

    ReplyDelete
  7. Folks - thanks for all your comments.

    e-Patient Dave and I will definitely followup on our blogs after our call to review data and make specific recommendations.

    ReplyDelete
  8. Hello John,
    Thanks for providing your side of the story and do side with you on this one. The healthcare data that is the most digital, adheres to some form of standards, however limiting, that can become liquid is indeed claims/admin data. We have a long way to go before we have clear clinical data moving through the system and yes, SNOMED-CT and ICD-10 will help in the future.

    But what is of concern here is the simple issue that if consumers, who are only now beginning to wake up to the need to manage their health records read such stories as Dave's, will is stunt adoption of these consumer-facing healthcare IT solutions. In the near term, we will need a process that assists the consumer in insuring that the data they have is indeed correct, or at least point out which data/claims may be problematic.

    Did a post on this as well exploring the issue of bad data and PHR adoption that you may find of interest. www.chilmarkresearch.com

    ReplyDelete
  9. Johnnysmooth,

    With respect, I have to ask: what on earth is the use of transmitting erroneous misleading data, just because it happens to be in electronic form?

    Have you even read my post http://is.gd/q6t2, where it details the result of this approach? Google Health immediately threw up a "requires immediate attention" alarm for a non-existent medication conflict, and reported that I have intestinal parasites, mets to the brain and spine, anxiety, depression, aortic aneurysm, and migraines, none of which is true, and most of which never were true.

    Please, read my post and then say whether you think it's useful to transmit this stuff.

    ReplyDelete
  10. Dave,
    Read your post and stand by my position. Why don;t you read MY post before future comments.

    ReplyDelete
  11. Oh, John Chilmark! I tried opening your signature profile and it said it was protected. And I didn't spot the link to your blog. (Good old Blogger, not turning them into clickables unless we add the href codes. My mistake; apologies.)

    Having said that, I DID read your post yesterday - loved seeing Oscar in a HIT blog post. :) But you said "Either way, end result is an inaccurate picture of the consumer’s health." So I'm not understanding why you feel it's sensible to transmit this data.

    But perhaps I'm misconstruing. When you said you agree with JohnH on this, did you mean you support transmitting ICD data, or did you mean something else?

    Again I apologize... as JohnH knows, I (like him) strive to be respectful in everything, but participating in these conversations is quite demanding given the toll it takes on my day job, and sometimes I'm too hurried. Sorry.

    Let me know what you think.

    ReplyDelete
  12. Old expressions are the best:

    Garbage In. Garbage Out.

    This actually goes to a deeper issue with interoperability and EHRs, HIEs and PHRs: how do you trust data from a source? How do you know that they won't be contaminating your record with garbage? Can you easily "undo" the operation to get rid of crud that some other system dumped into yours?

    ReplyDelete
  13. I think the focus on making better use of coding systems is well and good, but aren't we overlooking the part of the medical record with the deepest value -- free text? As a (formerly practicing) physician, I rarely paid attention to coded data - billing or otherwise. When opening the chart, I would seek out the most authoritative note(s) and read what other doctors that I trust had to say. From their contextual voices, I could assemble the patient's story.

    Google, of all services, should understand the value of text, and of the ability to apply smart (semantic) searching to that text. It seems to me that uploading the textual documents (perhaps starting with discharge summaries and clinic notes) would yield much more useful data than the current approach using (billing) codes.

    David McCallie MD

    ReplyDelete
  14. All the comments seem to miss a critical fact. Dave story did not happen in just any hospital. It happened at BIDMC.

    BDMIC EHR is designed and implemented by noneother than John Halamka, CIO and Dean for Technology at Harvard Medical School, Chairman of the New England Health Electronic Data Interchange Network (NEHEN), CEO of MA-SHARE (the Regional Health Information Organization), Chair of the US Healthcare Information Technology Standards Panel (HITSP) and a member of the Google Health Advisory Council.

    In other words, he is definitely one of, if not the, greatest expert in the country on EHRs and the all important INTEROPERABILITY of EHRs, a topic on which he has spoken and published at length.

    So, it is unfathomable that Dave story would come as anything close to a surprise to him. He had to know that talking of interoperability between these 2 data systems was a joke, one that unfortunately is now played at patients expense.

    It turns out, digging into what has been said over the last few years, that multiple blog posts constantly convey that the data in the EHR was of very little value. It is just that nobody paid attention to them.

    In a blog post announcing GoogleHealth interoperability with BIDMC EHR John Halamka wrote that an initial group of 5,000 patients with Gmail addresses were offered to transmit their BIDMC data to GoogleHealth.

    I am now asking, in this age of openess and transparency, that John Halamka release the data and let all of us know how many of these 5,000 have used the opportunity.

    There is just something that doesn't make any sense and I sincerely hope that someone in the Obama administration is paying attention to what is going on here!

    John, thank you much in advance!

    Gilles Frydman
    e-Patients.net

    ReplyDelete
  15. Dr. Halmaka:
    It seems that at the core of this post are some fundamental data issues (both definition and reconciliation) - which (for Health IT) are being attacked on numerous fronts; progress has been made, but much remains.

    For another perspective on healthcare data (from a colleague) I would urge you to read this post:

    http://dmsblog.burtongroup.com/data_management_strategie/2009/04/dear-mr-president-a-data-model-for-my-electronic-health-records-nearly-killed-me.html

    ReplyDelete
  16. I disagree with one of the points brought up by Gilles (Kosher Frog) and perhaps others.

    The data in electronic medical records is actually quite valuable. Problem lists (maintained by clinicians), medications, allergies, test results and other parts of the medical record that are viewable through patient portals like PatientSite are not only useful to both patients and clinicians, but are generally accurate. (They should ideally be combined with resources to assist in the interpretation of this information.) However, health care information system also contain clinically worthless (and potentially alarming and dangerous) information, such as codes used for the purposes of billing.

    We may not be able to get away from the use of ICDx codes for billing in the US healthcare system anytime soon, but we must not assume that these "diagnoses" align closely with clinical reality.

    Caveat emptor. It is up to the people using this information to understand when and why and from where it was collected. If health care providers and health information professionals believe that this "administrative data" represents clinical reality then shame on us.

    However, since our patients may not have the background to understand this, it is up to systems developers (both on the sending and receiving end) to a) explain the when, the why, and the source of each data element, b) provide education and disclaimers to those viewing it, and c) provide customer support to those using the information. This last piece is crucial, so we avoid finger pointing about the source of the problem.

    If we are going to create data connection tool, we need to do so in a thoughtful way. Just because we can do something doesn't mean that we should.

    Daniel Z. Sands, MD, MPH
    Cisco and
    Beth Israel Deaconess Medical Center

    ReplyDelete
  17. Danny, that's a superb, thoughtful comment. Wish I'd said it myself.

    ReplyDelete
  18. Danny makes a very good point which no one is denying :-)

    Vocabulary here is all important. Danny, as a physician, talks of EMR (Electronic Medical Record) while I have been arguing about the problems of the EHR. I always consider that the EMR does contain the physician's notes and the data elements that help build a valid longitudinal view of the patient medical history. That is the part that obviously didn't get transmitted to Google Health and it raises an interesting question.

    Is the BIDMC EHR system more interoperable while transmitting information to entities subjected to HIPAA regulations (doctors, hospitals, insurance companies) than it is with Google Health.

    The level of complexity associated with the total obscurity of the system makes it impossible to know any of this.

    But one thing is sure: the current EHR systems are certainly not going to help us move from the doctor-centric century-old model of medicine to the open, transparent and democratic model of participatory medicine we are all promoting lately.

    ReplyDelete
  19. Many good comments from everyone. To me the patient should be able to see the same information that physicians have. Certainly, coding the information and describing the diagnosis and treatment to a patient is highly beneficial, but we need to start with medical notes, reports, and results.

    For patients and physicians to have confidence in a PHR or any system, the information must be correct. This requires a process of checking and releasing information rather than transmitting data that is not verified.

    ReplyDelete
  20. Dear Dr. Halamka,
    Sharing healthcare data in a way that works well for patients, doctors, staff, and all the other people who are interested in this is going to be exremely difficult, if not impossible. But bravo to you for getting into the ring! You had the chutzpa to give it a try. Bravo to Patient Dave for raising some of the important issues that need to be worked out. Bravo to HealthVault and Google Health for stepping up. These are critical issues for all of us, and I'm so grateful that a few are willing to lead the way.

    ReplyDelete
  21. What I find interesting about this whole discussion is that no one has raised the issue of why doctor's offices and hospitals submit inaccurate or "exaggerated" codes for billing in the first place.

    The way we pay for our health care has everything to do with the inaccuracy of the administrative data. A physician may desire a particular lab result to better monitor his patient's known condition, but his billing staff is aware that the patient's insurance will not reimburse for that lab test for that reason. Instead, they have to add a diagnostic code for something that the patient does not have, to ensure the insurance company will pay for the test.

    In other cases, codes that seem to exaggerate the condition may be included for a variety of reasons: to raise reimbursement rates to better reflect the level of effort the doctor has put into the patient, or to establish a history on which to base a later referral, for example.

    I heartily agree that ICD-9 is an entirely inadequate structure for modern health data, and hope that ICD-10 brings some relief... but many of the problems e-patient Dave had seem to also be related to the world of insurance reimbursement, and how coding affects that. Our records are, in effect, falsified by billing departments in an effort to get paid for what was actually done.

    And there is definitely value in the patient having that information. What if you're denied insurance down the line based on your "anxiety disorder" or "brain tumor" that never existed? What if you catch outright fraudulent billing on the part of your doctor's office? What if a medical error trickles into the billing codes and allows you to correct the *actual* record?

    The information, as it stands, is useless for some purposes, but is still valuable from many perspectives... and knowing what's there can guide the process of making the information more accurate and useful for all purposes.

    ReplyDelete
  22. Could someone be so kind as to post references to articles about the issues with using billing data for clinical information? The links here touch the surface of the topic but don't provide a full scope.

    ReplyDelete
  23. In a blog post announcing GoogleHealth interoperability with BIDMC EHR John Halamka wrote that an initial group of 5,000 patients with Gmail addresses were offered to transmit their BIDMC data to GoogleHealth.

    I am now asking, in this age of openess and transparency, that John Halamka release the data and let all of us know how many of these 5,000 have used the opportunity.

    There is just something that doesn't make any sense and I sincerely hope that someone in the Obama administration is paying attention to what is going on here!

    ReplyDelete
  24. We may not be able to get away from the use of ICDx codes for billing in the US healthcare system anytime soon, but we must not assume that these "diagnoses" align closely with clinical reality.

    Caveat emptor. It is up to the people using this information to understand when and why and from where it was collected. If health care providers and health information professionals believe that this "administrative data" represents clinical reality then shame on us.

    Recep Deniz MD

    DoktorTR.Net

    ReplyDelete