This was the final event of the meeting for me, and I apologize for not getting notes up sooner. It’s harder when you’re back at your actual job…
I believe a video of this lecture will be posted at some point on this NIH site, and you can get Dr. Halamka’s slides at his blog, Life as a Healthcare CIO. But if you just can’t wait, following are my typically lengthy notes.
[Update] The video is here: Thanks to Taneya for the link!
Knowledge Services and the Role of Medical Libraries in Healthcare IT
John Halamka noted that he was an NLM fellow in 1997, when he presented a paper on the use of the internet to exchange health records. He was widely considered to be insane as a result.
The recent stimulus bill provided 787 billion: we are responsible for spending 30 billion of that wisely.
Halamka sees five grand challenges that will require work from NLM, medical librarians, and informationists.
Background: if we’re going to exchange health records, we need to protect privacy. Everyone has different ideas of privacy. Halamka’s genome has been published, and you can buy his stem cells for $49, so he has one idea about privacy. Other people may have other ideas, and we need to work with that as well.
Take a cold: it’s not particularly confidential, so we probably don’t care if it shows up in our record. Mental health visits, or stays at the Betty Ford clinic–these are more confidential.
So it would be good if we could segment records to control who sees what: have all the information stored, but not all viewable by everyone. This is very complicated to achieve.
Right now, hospitals can’t share records: if you get an MRI at one hospital and then cross the street to be treated at another, you need a new MRI. This is clearly inefficient: we want to be able to share records where appropriate.
But who do you trust with your information?
Your doctor? Yes, maybe.
NLM? Of course!
Your insurance provider? Merciful heavens no.
Your employer? Well, you really can’t, can you?
So we want everyone to have an electronic health record (EHR) but there are currently very different standards and types depending on who creates it. Currently we can get a log of who looks at it, but not what happens with the information after they’ve accessed it–who they show it to. We need to better log where info goes and for what it’s used.
We hope that recent healthcare reform will lead to paying for quality, not quantity. To ensure this, we need to be able to measure quality, and to provide feedback for improvement.
We also need to assure device security, in case someone loses a Blackberry on the train.
Story: A Beth Israel Deaconess Medical Center (BIDMC) employee did some analysis, copied it to a flash drive, moved to San Francisco, and copied the drive to a laptop, which was stolen and pawned. If this kind of thing happens, regulations state the event needs to be disclosed to patients, and to ‘prominent media,’ which was done. BIDMC was fortunate that the State of the Union address and the introduction of the iPad were both that same day, minimizing the bad press; but we want to make sure this sort of thing doesn’t happen in the first place.
We want to make sure every patient gets the right care: we need to protect the vulnerable. Make sure people get the right meds. Halamka’s grandmother died from medication combinations that needn’t have happened if her complete health record had been accessible by her physicians.
Grand Challenge #1: Consent to share data
Principle: you should be able to decide who sees what about you, and when. Presently, state law trumps HIPAA, and you may not even have the same standards from one hospital to another about what can be disclosed.
- Opt-out (info is shared unless you say no)
- Opt-in (info is not share unless you request it)
- Either one with various restrictions
- No consent needed (share everything freely)
So we might say we’ll have everything shared, except if it involves drug or alcohol use. But this is very difficult to automate: natural language searches would seize on and redact parts of a record where it mentioned that a person ‘uses alcohol’ for disinfectant, for example.
Institutional level options: BIDMC says ‘we won’t share anything unless you say so.’ Another model could be: ‘share anything for research’ purposes (Halamka is part of the Personal Genome Project: committed to research, so use any of his info for any study you want!) Or: “share everything except my mental health info.”
Situational level options: We might share all data for emergency care, but not in general: dermatologist doesn’t need to know entire history, but EMT should have access
What if we could use a controlled vocabulary to codify these consent options?–put together a record in xml-style format that states a person’s specific opt-in and opt-out choices
We could codify a format for the record itself, then for a consent section. it would be great to have research data, for example, along with details on exactly what it was cleared to be used for; wouldn’t have to seek new permissions every time.
Grand Challenge #2: Engaging the patient
ARRA want us to share with the patient—but there’s no Unique ID for patients, so it’s very hard for an institution to be sure that you’re the right person to access a record.
The idea of a National Health ID number is scary to people—we don’t want to have something that could potentially be linked with employment and tax records.
One example: BIDMC gives everyone a site-specific number so they have all their records from that institution linked and accessible–but that doesn’t help if you go to another hospital.
Another model: MS Health or Google Health can get your permission to pull information from pharmacies, etc. to compile a record. And then, you could get it in an RFID implanted in your triceps, as Halamka did! Or, you could just let people access it, like a “health URL.”
You could get the address from anyone you trusted (Google, a healthcare vendor, whatever) and could tell your doctor “post this visit record to…” the URL, which would then cause the record to update. If you wanted to change URL providers, download your info, whatever, you could.
A concern: when you share info with patients, it tends to be confusing. What does this term mean? Is 100 cholesterol good or bad? So it’s important to link the health record with contextual info to make it actionable to patients. Otherwise they can have pages and pages of detailed information about their health that actually doesn’t really mean anything to them.
Reassurance: it’s unlikely that we’ll wind up with a giant database in the White House basement managed by Sarah Palin. If we can leverage the current web, with social networking and other tools, we will likely wind up with a lot of options that don’t require this scary centralized control.
Grand Challenge #3: Standards enablers
It’s hard for doctors to get things in standard formats. They use all kinds of terms, scrawled handwritten notes, etc. They don’t think in terms of hierarchical vocabulary and the correct way to phrase things for the system.
We need standards for three things: content, vocabulary, and ways of moving data.
Content: We actually do have existing standards that are basically good enough on the content side. There’s lack of detail in some places, choices of two standards in others. There’s a lot of learning called for: a doctor-designed standard works to be very simple and easy to use, while standards organization folks designed a much more all-inclusive but difficult to learn/use one. ‘Green CDA’ makes an xml-style sheet, showing patient info, problem, etc. We can combine two things, for aspects that work for both doctors and standards people.
Vocabulary: We need a common terminology. There are currently standards for several different uses, which is fine since not everyone needs the same functions—we just need to get them standardized and mapped to each other. This will allow for implementation of the good stuff in the PHR.
Moving data: Finally, there’s a need for standards in getting data from place to place without it being changed, intercepted, or sent to the wrong place. Again, there are several existing ways to do this, so what we really need to do is just pick something that does the job well enough, and go with it. This will probably be REST—used by Amazon, Google, and others that send a lot of data around.
Grand Challenge #4: Aggregating data
It would be good to be able to look at data from many peoples’ records to see if there are indications that a new drug is dangerous, catch early signs of outbreaks, etc. In the US, we just don’t trust government centralization of data, so we shouldn’t really count on that. So instead we find out who is trusted, and delegate this record keeping to them.
We will need to work out some things like, how do you record that–one document per episode/visit? Or keep all the data from a person’s EHR all together?
Where do we put the complexity? Who has to do the structuring, mapping from one thing to another? If we codify the data at starting point, there are fewer chances for things to go wrong. Here’s where standards come in: if we can make sure that data is structured at the point of entry, there’s no need to translate or map it later.
As noted, we don’t trust central aggregators, so instead we should keep this work close to the hospitals. They can de-identify as much as possible. Often you don’t need to know names/addresses for things—so it’s OK to keep the data anonymous, but with ability to re-link to personal data in case it’s needed for epidemiology later.
Example: Imagine Dr. A orders a test for a patient. Patient doesn’t get it right away, but later goes to Dr. B and gets it done there. We would need to link this information, but how?
Halamka suggests looking to Las Vegas. They keeps lots of info, all de-identified, through a process in which they standardize all names, and use this standardized form of a name, a date of birth, and a person’s gender to create a unique ID with it. So your name is not in the database, but the record is uniquely you. We could do something like this.
Another option is that we could just point to the location of the data, and not actually put anything into the net: just say “[your name here] has a record at [institution name]. But what if you’re at a mental health clinic? If it’s known that you have a record there, that’s disclosing something about your health.
Halamka likes SHRINE: Shared Health Research Information NEtwork. It leaves the data near your doctor, but allows for querying of multiple hospital databases. So you can run a query on obesity and lifespan, and get data from patient records at the various institutions participating, but never see data on any individual.
For this to work, we’ll need libraries to make sure data is structured.
Grand Challenge #5: Decision support
More literature is published every year than clinicians can read in their whole lives. We have a ton of data: now what we need is ways to know what to do with it. We establish rules: take this action for this condition. But these rules may be different from one organization to another. If only we had nationally standardized rules. We need a knowledge base for hundreds of common conditions that will set up the rules for treatment so that it’s easy to know the best thing to do.
NLM is great at codifying knowledge, and we need this. We could have an evidence base that lets you put in patient condition info and it returns the answer: do this, and here’s why, and here are safety considerations. It could be consistent across country, so you’d always get the best evidence-based treatment at every hospital. It would of course be reviewed regularly.
The role of medical librarians now becomes empowering clinicians to access this information. Doctors will still need to be able to consult experts. Catalyst.harvard.edu codifies all the expertise of all 18,000 faculty members at Harvard using NLM terms—it’s like social networking for science. Something like this, widely applied, could be a great resource.
Final Thoughts: The genome
He is one of the first to be completely sequenced. His genes are available to the public, along with some possible predispositions that can be deduced from the sequence. This has potential for health: Halamka sought early treatment for glaucoma as a result of what he learned.
But sequencing is easy, since we can now sequence a full genome for $500: the interpretation/annotation is the hard part. So you have this gene: what do you do about it? That’s what will take time and money now.
From the Q&A
First four challenges are more medical records librarian work than medical librarian. Once we can aggregate data about populations, will collective literature be irrelevant? — No. We will still need interpretation, studies to show why the answers to some queries may be what they are. Easy to pull facts–expertise is still needed to say what it means.
About that chip implant? — The idea was that maybe we should just all carry our medical records around with us. Naturally he went out and had it done to try it out: the RFID chip points to a website with his info. Doesn’t hurt that much to get implanted, but pretty much won’t ever be removed, and can’t go into Best Buy because their DVDs use chips with the same frequency. Reportable results from his experiment: probably won’t catch on. Easier to put that information somewhere else.
We know that interpreting for patients is absolutely crucial. Critical need, place for medical librarians.
About E-patients — Halamka used to share all info with patients, including administrative data. Problem is that this tends to show a lot of ‘reasons for test’ that show up as conditions a person has. Disconnect between clinical observations and billing codes. Now, only SNOMED info goes to patient, so they only see the information about conditions they actually have.
As we know, there’s no curation on the internet, so you have to be careful and wary. The Bulgarian Journal of Irreproducible Results is as likely to come up as NEJM. Sometimes, patient data is what you need, and that’s a helpful use for the internet, but curated databases are much more likely to be useful in general.
Also, beware: patients come in very different levels of sophistication. E-patient Dave is not necessarily typical. Never discount crowdsourcing, but set a floor; minimum standards for reliable info.
How to provide continuity of care is a concern for long-term care. Record should follow you all your life, and include more than prob/med list.
Concern: doctor looks up the right way to do something the first time, and then relies on ‘ever-failing memory’ for the next several times, while the known best practice may change.
We need to stop ordering “heparin” and start ordering “anticoagulation”—so someone looks up the procedure every time. Also with chemotherapy: order therapy, and let the computer work out the dose. Problem: what if the computers go down? If the system isn’t there, you can’t do anything. Multiple backup systems are crucial. But basically we never want to rely on physical memory, or the apprenticeship method of learning. We want solid evidence, and we want it to be consulted every time.