As the chief information officer of a large academic medical center, I oversee four petabytes of data. Is that “big data”? I have little difficulty storing, securing, and accessing it, so I’m not sure it qualifies as big. To me, the bigness of data is not its absolute size, but the task of transforming it into wisdom.
At Beth Israel Deaconess Medical Center (BIDMC), we use big data to create real-world applications that lead to wise clinical decisions for patients. That’s something any forward-thinking provider must aim for in today’s data-driven health care environment. I’d like to discuss three BIDMC big-data applications that required both technical expertise and leadership — and that have helped our patients, including my family and me.
Patient-Generated Health Data
For most of my life, my blood pressure (BP) has been 110/50 — a boring number that is considered medically reasonable. During an annual physical exam this past summer, it was 170/100. My medical workup suggested no medical cause for this elevated BP, such as heart- or kidney-related problems. My clinician was perplexed: Was this sudden increase caused by a stressful commute (the Massachusetts Turnpike is awful), too much tea (caffeine and theobromine), or work-related anxiety? To find out, we needed data.
I have a Withings BP cuff linked to my iPhone 6 via Bluetooth Low Energy. My iPhone is connected to my clinician’s electronic health record via a new app called BIDMC@Home. Using the cuff, I took my BP before and after commuting, drinking tea, and attending anxiety-provoking meetings — nearly 100 measurements in a week. The raw data were just numbers, although they helped reveal interesting information — that none of my life activities (commuting, tea drinking, work) influence my blood pressure. The problem, logged as a discrete data point in my electronic health record (EHR), turned out to be my parents. My mother has essential hypertension, as did my father. So did their parents. It took 53 years for my genome to catch up with me and manifest as hypertension. What wisdom did we glean from this knowledge?
Well, I have glaucoma (elevated eye pressure, also inherited from my family) and an occasional fast heartbeat, called atrial tachycardia. Both can be treated effectively with beta-blocker medication, as can essential hypertension. Drawing on clinical guidelines, my EHR took all three conditions into account and suggested beta-blockers as the ideal medication for me. Today, with 25 mg of the beta-blocker metoprolol taken at bedtime, all of my conditions are completely controlled without any side effects. In short, the BP data I had gathered telemetrically at home, coupled with information in my EHR, helped my clinician and me make a wise choice about my treatment.
How did we make BIDMC@Home happen at an organizational level? Our IT leaders started by showing how patient-generated health data could tie in with pay-for-performance reimbursement. Their IT strategic-planning exercise, with a 24-month outlook, involved 30 stakeholders on the front lines of patient care. This group saw that new mobile apps for patients (which measure and manage health outcomes) could yield clinical and financial benefits. The idea was presented to BIDMC’s senior-management operating council and then to the Board’s IT oversight committee. With support at all levels, the project was funded, new support staff were hired, and the rollout ensued. Our IT leaders also served on Obama Administration committees that wrote requirements for patient-generated health data into 2018 federal health care IT regulations, thereby ensuring the concept’s longevity.
Precision Medicine
In December 2011, my wife was diagnosed with stage IIIA breast cancer with a specific type of tumor. (For medically minded readers, it was estrogen-positive, progesterone-positive, and HER2-negative.) She was 49 at the time, is Korean, and has no other significant medical problems.
Using I2B2, an open-source tool available at all Harvard hospitals, I was able to ask this big-data question: Of the last 10,000 Asian women near age 50 who were treated for the same tumor, what medications were used, was surgery or radiation necessary, and what were the outcomes? The answer: a combination of taxol, adriamycin, and cytoxan was most effective, but the amount of taxol had to be carefully limited to avoid nerve damage. My wife was treated successfully and is now cancer free.
To launch I2B2, the leaders of that effort assembled a coalition of senior IT people, throughout Harvard’s hospitals, who became early adopters. They developed the policies and technologies needed to query large databases across institutions, thereby attracting national and international audiences. With each new adopter, momentum built, implementation risks diminished, and institutions decided they didn’t want to be left behind. I2B2 is now used by more than 60 academic medical centers globally. And the Obama Administration’s Precision Medicine initiative aims to bring this kind of decision support, using data from large numbers of successfully treated people, to every new patient.
Wise Analysis
For my wife’s case and my own, the data analytics were done retrospectively, not in real time. A computer did not constantly mine data and then alert clinicians to a new finding. Someone had to ask the right questions — in short, to glean information from the numbers and then use knowledge to make wise, analytical choices.
At BIDMC, we now use a tool called “screening sheets” to support continuous data analysis. Experts decide what data elements and what questions are important for common diseases — and that information is built into the screening-sheets tool. As patients receive new medications, lab results, and diagnoses, the electronic health record alerts clinicians when to take action. For example, a patient with newly diagnosed diabetes is automatically enrolled in a protocol that includes eye exams, foot exams, and pneumonia vaccines. Any gaps in care for the patient are coupled with information about best practices, and the clinician is proactively informed about both so that he or she can make a wise clinical choice.
We will soon incorporate big data from the genome into our screening sheets. As the second human sequenced in the personal genome project, I know the disease risks identified in my DNA. My clinicians, aided by computer-based clinical decision support, can analyze my new lab results and symptoms according to the likelihood that, given my genetic profile, I will experience a particular disease in my lifetime.
To make screening sheets a success, our IT leaders articulated a vision that was consistent with clinicians’ need to improve quality, safety, and efficiency for better population health. Although the clinicians could not pinpoint the specifications they desired, they enthusiastically embraced the resulting functionality — a “list manager” of patients identified by the screening sheets. The list manager might, for example, generate this message: “Of 4,000 patients seen, 372 (9.3%) have a diagnosis of diabetes; 50 of the 372 have not had their recommended pneumonia vaccine this year. Your care manager should contact these 50 patients (listed here) to schedule vaccination.” It’s easy to act on that type of message.
At BIDMC, we don’t overwhelm clinicians with big data but instead reduce their burden by staying one step ahead of what they need to make wise clinical decisions. Our IT leaders “skate where the puck will be” and score important goals for the entire organization and the patients it serves.
Sorry, comments are closed for this item.
Bob Kelley, Senior Research Fellow, Truven Health Analytics
The example of using a large sample of similar cases to determine the appropriate course of treatment for Dr. Halamka's wife seems to minimize the significance of an extraordinary shift in the method of creating medical 'evidence.' For years we have focused efforts to reduce unwarranted variability in treatment decisions by promoting 'evidence-based medicine.' The only acceptable source of this evidence has been the RCT and, in limited cases where these trials aren't possible, epidemiological studies. By its very nature, the RCT is a 'small data' methodology. Its deficiencies result from its perceived advantage in controlling for random patient characteristics. As a result of this 'control', variables that might be significant (e.g., Korean descent) cannot be revealed. Maturing 'big data' analytic methods are becoming an important supplement (if not substitute) for many of the traditional methods of creating evidence. I applaud the efforts at Beth Israel Deaconess to begin to tap into the power of this data, but would add a caution. Dr. Halamaka's example suggests a tendency to use 'big data' to simply expand the number of anecdotal cases exposed to the clinician's own unique set of decision-making heuristics. This does not raise to an acceptable level of 'evidence.' We could ask, "Would all clinicians reach the same conclusion, given the same set of observations?" "What is the limit on the number of variables that a clinician can actually deal with effectively?" I still believe that the discovery of important evidence is a continuous process of building on previous discoveries. We need to create a new mechanism for measuring the strength, or more importantly the usefulness, of this new type of evidence. The availability of this 'big data'creates the opportunity for a new kind of continuous, real-world experiment, in which we are all participants. We need to redesign clinical practice to incorporate this real-time, up-to-date evidence, into clinical practice. I don't think the best approach is to simply enable each clinician to access the data and apply his or her own set of heuristics and criteria in making a treatment decision. New analytic techniques will enable better rates of discovery of new and useful evidence.
January 05, 2016 at 10:18 am
Cathy
This sounds fabulous! However, in the real world of health care, most EMR systems are so cumbersome to use, and so many end users are either too rushed, or not well-enough trained to enter every single bit of data, that it seems like a long shot. I've been around since long before the inception of the EMR (as an end user - not an IT professional), and while your news is so exciting it makes my mouth water, it is doubtful I will ever see it used routinely in my lifetime. I and my team would be happy with a more efficient platform to perform our daily tasks, and the ability to cross-reference other systems. Keep up the good work, and maybe someday this awesome tool will save many more lives! Just never forget that the integrity of the EMR, no matter how awesome it is, depends on the real people entering data, so training and facilitating at that level is a very large piece of the pie!
CathyB
*30 years Medical Transcriptionist - from noncorrecting Selectric typewriters to Word Processer, to EMR
*Meditech Superuser and Trainer
*Current Medical Scribe using Centricity platform
January 16, 2016 at 10:04 am