Medical data that results from anonymized, centralized Electronic Health Record (EHR) information can provide powerful insights into trends and patterns that might otherwise not be apparent at first-glance. Such data might also provide evidence that can drive EHR development efforts.
Practice Fusion made available a limited set of fully anonymized medical data on the Microsoft Azure Market in order to encourage investigators and data analysts to find patterns. This effort is part of the Analyze This! challenge, and has garnered interest by quite a number of entrants to this challenge – the winner of the challenge will be announced at the Health 2.0 conference in San Diego at the end of March.
We have also looked at patterns in the data, spurred by the kind of imaginativeness demonstrated by the “Analyze This!” challenge. And, in poking around to identify patterns, we have found something interesting.
Case study: analysis of blood pressure
Using completely blinded records (NIST-level de-identification), we can look at patients who have the clinical diagnosis of Hypertension (by the ICD code range documented by physicians in their charts), and we can also look at blood pressures in chart notes. An analysis of medications used for patients with Hypertension is something we will report at a later time. However, looking at blood pressure readings directly (regardless of what their diagnoses are), we can identify a set of patients who have average blood pressures that are elevated (systolics on average greater than 140 or diastolics on average greater than 90). We will call that patient data set “Uncontrolled BP.”
We can also look at all the medications that clinicians have used linked to the diagnosis of hypertension (since the system allows physicians to link a medication to a diagnosis), and can come up with a set of “BP medications” – the list is from actual experience, rather than from reference to an external data listing. One can make arguments for and against this approach.
Of patients with elevated BP, we can see what percentage of them are on “BP medications” and what percentage of such patients are not on any BP medications.
Now, of those patients with uncontrolled BP, and who are not on any medications, we can see how many of them have the clinical diagnosis of Hypertension, and how many have not ever been given the diagnosis of Hypertension – 64% of such patients (uncontrolled BP not on meds) do not have the clinical diagnosis of Hypertension.
Having the clinical diagnosis of Hypertension increases the likelihood that the patient will be given treatment for their blood pressure (we will publish that data at a later time). We also looked at what kinds of specialties might be more likely to “miss” the diagnosis of Hypertension in patients with consistently elevated blood pressures – we looked at the percentage of specialties represented in the “missed opportunity” patient set, compared to the percentage of specialties represented in the overall patient set, and calculated a variance by specialty – the “Relative Risk of Missing the Diagnosis of Hypertension.” Interestingly, Internal Medicine was most likely to catch such missed opportunities; Pain Management was most likely to miss the opportunity to code consistently-elevated blood pressures as Hypertension in their charts.
Implications for product development
We have a product development approach that includes several drivers: (1) elements needed for ONC-ACTB Certification (necessary for Meaningful Use incentive money access), (2) elements internally identified as important, and (3) suggestions by users on features or modifications that would improve the usefulness of the application. These drivers help create our roadmap.
Now, given that we have accumulated significant data that can be analyzed in a fully anonymized fashion, there is a fourth driver: opportunities emerging from data analysis. This is something that might be extremely difficult for a non-web-based EHR to do, since the data exists externally, fragmented into each of the local installations where the product is deployed.
From our analysis of Hypertension, we can see that a simple prompt can be built into the product, such that if a given patient has a consistently elevated blood pressure in prior chart notes, and a diagnosis of Hypertension is not in the record, then the clinician can be prompted with something like this: “This patient has consistently elevated blood pressures. Add the diagnosis of Hypertension to the record? (Y/N)”. When Hypertension is included as a specific diagnosis in a chart, the likelihood that it will be addressed and treated is higher. A simple prompt like this may have an impact on the occurrence of the risks of uncontrolled hypertension – heart attack, stroke, congestive heart failure.
Over time, it will be interesting to see if the introduction of data-derived product features impacts the clinical delivery of care. My sense is that we will be able to demonstrate such changes with more longitudinal data.
Conclusions
Medical data will be a driver of product development for EHR products that are situated to be able to see the data, and make changes that show up “in the field” immediately. Web-based, centralized data is central to being able to identify such trends as they arise. Further, the creation of product features that address patterns as they are identified can be tested longitudinally, so as to make sure that the tools are “making a difference.” This is extraordinarily exciting stuff!
Robert Rowley, MD
Chief Medical Officer
Practice Fusion EMR
