Healthcare Analytics, Population Health Management, Healthcare Big Data

Analytics in Action News

Predictive, clinical analytics at MGH turn data into insights

By Jennifer Bresnick

- Massachusetts General Hospital (MGH) has a reputation as one of the top epicenters for healthcare in the nation, bolstered in no small part by its deeply integrated and robust clinical analytics programs.  Spearheaded by MGH radiologist Dr. Michael Zalis and software architect Mitch Harris, PhD, the Queriable Patient Interface Dossier (QPID) began in 2007 as a way to pull pertinent information from the EHR, but quickly evolved into a fully-fledged clinical intelligence platform based on machine learning that provides predictive analytics, patient risk scores, and key data insights to make clinical decisions more efficient and effective.

Dr. David Ting, Associate Medical Director for Information Systems at the Massachusetts General Physicians Organization, sat down with HealthITAnalytics to explain how MGH leverages their vast pools of patient data through the QPID engine to deliver important details about patients at the point of care.

What are some of the uses cases for the QPID analytics suite?

Dr. Zalis and Dr. Harris created what a lot of people initially looked at as a Google for patient data. The first iterations of QPID were simply about entering a search term like “malignancy,” and then QPID would scan not only your primary EHR, but really any data repository where there’s electronic patient data.  It would look not only in the coded, discrete fields, but also look through text reports, lab reports, radiology reports, problem lists, and medication lists – really anything that contained patient data.

It wouldn’t just look for the term “malignancy,” but it would look for “cancer,” “colon cancer,” “breast cancer,” “tumors,” or any sort of adjectives that go along with the term malignancy.  And it goes a step further. As users enter in searches and try to refine searches, QPID learns from that, and it gets better and better at figuring out what a user is really looking for.

READ MORE: Population Health is Top Data Analytics Challenge for Providers, Payers

Now, we have several different uses for QPID as it has evolved and grown.  Whenever a medical resident admits a patient, they use a functionality called the QPID portal, which is a web-based tool that is a collection of standard searches that pull up data we typically need when admitting a new patient.  Things like, has this patient ever had a history of coronary artery disease?  Has the patient ever had pulmonary disease?  Has the patient ever had a blood clot?  Has the patient ever been on blood thinners?  Does the patient have a family history of heart disease?

You can either go into the patient’s chart and do it manually and take hours and hours and hours of looking through the chart while you probably miss a lot of things, or you can go to this portal where everything just shows up on your screen like a giant radar screen, and it’s a beautiful display where all of the questions that you want to know around a patient medical admission are queued up automatically.

So, in a way, you can imagine that these portals are something like a special EHR viewer.  We have created standard search sets to answer questions related to different departments and activities.  You would ask slightly different questions if the patient is being admitted to gastroenterology as opposed to orthopedics, and we’ve designed the program around that.  They’ve got a portal for the emergency room.  They’re creating a portal for end of life care, and so on.  It’s really extremely efficient.

How does MGH use predictive analytics to identify patient risks?

We recognized that insurers and our surgeons are concerned about patients that come to Mass General, which is a quaternary referral center.  Patients come in with the most difficult cases for vascular surgery or for orthopedic joint replacements.  These may be patients that have sought care in other institutions and were told that, “Gee, your condition is so difficult, you should go to another doctor to check this out.”

READ MORE: Ochsner Adopts Artificial Intelligence Tool from Epic Systems, Microsoft

Historically, our physicians have been taking on these cases without the benefit of really, truly knowing the surgical risk that this patient is incurring, or if this is even an appropriate procedure for the patient to have.  You can imagine where insurers come into this, because they would like very much not to be paying for inappropriate or unnecessary joint replacements and back surgery or vascular bypass grafting.  It’s expensive and it can be dangerous for the patient.  But how do you know?  How do you know whether something’s appropriate or not when this is a patient that comes to you and you have 15 minutes to go through the chart or talk to the patient?

Well, that’s how we use QPID.  For any given high cost or high risk procedure, they’ve worked with experts in the field to say, “What are the questions that you need to have answered to allow you to decide what the surgical risk is?”  The system automates those searches using national guidelines, and then it essentially shows the results in a dashboard with a red, yellow, or green risk indicator for the surgeon or proceduralist to see.

And it’s very interesting.  Initially, the leadership was very excited about this, but we were a little nervous about how our surgeons would respond.  Would they think that now a robot is telling them what is appropriate to do?  As it turns out the surgeons love it, and the American College of Surgery loves it so much that they want to make this a standard of practice across the country.

Surgeons, even the world-renown surgeons, do not want to operate on a patient who’s going to die on the table.  The last thing they want to do is do harm to a patient or do something inappropriately.  So, if by using this tool, you can have a more informed discussion with the patient and say, “Mrs. Jones, as you can see here, the tool I’ve used is supposed to look through your entire medical records shows that you are at extremely high risk for having a bad surgical outcome. You would be much better suited to have this other kind of surgery.” It turns out that our surgeons are relieved that they have that kind of backup at their side, so rather than feeling insulted that the computer is kind of guiding their decisions, if anything, this is kind of the decision support that the surgeons are very, very interested in.

It also helps the patient make much more informed decisions and consent to procedures with a greater understanding of their potential outcomes.  In the past, if a patient came in for a knee replacement, and they would all hear the same thing.  They would hear the orthopedist saying, “Well, you know, there’s a one percent chance that you might die from anesthesia complications, there’s a two percent chance that you might get infections.”  And so on and so forth.  But that’s all just on average.

READ MORE: Machine Learning, Genomics Predict Risk of Aortic Aneurysm

Using analytics, we can specifically tell the patient, “Well, based on your profile, you have a 15 percent chance of an infection.  You have a 90 percent chance of a good outcome from this type of surgery or this type of joint.”  And so, you can tell immediately from a patient perspective a risk management perspective, and the physician’s perspective that this is a win, win, win.

Are there any blind spots when it comes to the data that QPID uses to make its determinations?

There’s no way that QPID can scan paper charts. If those papers were actually scanned, literally put into the record, then we could recognize from that.  But as long as it stays in a paper chart, it’s not available as a data point.  But we ca use a kind of surrogate data, so even though a physician’s notes may be entirely on paper, chances are the labs are available in coded form.  You can tell a lot from a patient just by looking at their lab data, or just looking at their radiology reports even if you’re missing the actual paper chart.  As long as QPID has fingers into a data source, then we’ve got something to go on.

It does beg the question of whether or not we putting too much trust in systems like index and search functions.  But we might be lulled into a false sense of security because we don’t think that, “Hey we’re only looking at data sources where we have the data and we might forget that, gee this patient coming from the other side of Massachusetts may be not connected to our data sources.”

And so that’s where I think you always have to be careful and say that you should definitely do the automated chart biopsy as step one, but then you always have to talk to the patient and the family to supplement your information.

X

Join 25,000 of your peers

Register for free to get access to all our articles, webcasts, white papers and exclusive interviews.

Our privacy policy

no, thanks

Continue to site...