The kind of medical data that can be garnered from anonymized, centralized Electronic Health Record (EHR) information is staggering. Recognizing this, and recognizing that the possibilities considerably outstrip our own imaginations, we uploaded a limited set of de-identified medical data to the Microsoft Azure Market in order to encourage coders and investigators to look at the data and find patterns and correlations. This is the thrust of the Analyze This! coder’s challenge.
Some of the power of the data is seen within the specific Practice Fusion Medical Research Data itself. More powerful correlations can be seen when this data is compared to other data from other public sources – an example would be the finding that average Body Mass Index (BMI) is correlated with the prevalence of farmer’s markets and fruit consumption, weakly correlated with median household income (for females, but not for males), and not correlated with other factors such as the density of fast food restaurants, or educational level.
One simple finding from the Practice Fusion experience is the prevalence of various diagnoses seen by our users. Since Practice Fusion is a web-based EHR that has been actively adopted by smaller practices (group sizes less than 5 are the bulk of our users, though larger groups can and do use the system), and since many of our users are Primary Care Physicians (PCPs), the findings of what constitute the “25 most common diagnoses” represent the experience of PCPs in small practices across the country.
The list of common diagnoses, therefore, represent a combination of the prevalence of certain conditions in the U.S. population, as well as the kinds of practices using the EHR. Not being a specialty-focused system, the findings are what PCPs see in their offices. Recognize also that the data set is quite large (there are now over 7,000,000 patients that have been uploaded into the system, making it one of the largest medical data repositories in the country).
The findings should not be too surprising: hypertension, hyperlipidemia and diabetes are the most common diagnoses among patients seen by Practice Fusion-using physicians. Back pain, anxiety and obesity follow closely behind. The details can be seen here:
Identifying the kinds of conditions commonly encountered in clinical practice – among smaller practices not affiliated with any specific hospital or institution, particularly Primary Care practices – is important in setting the context for where disease intervention efforts should be focused. Clinical Decision Support and multi-disciplinary outreach – collaborative decision making that includes the patient in the discussion – can be prioritized as a result of this kind of data. This small example of the use of de-identified data shows how such data can be used to fashion policy and prioritization in order to improve health for the common good.
Robert Rowley, MD
Chief Medical Officer
Practice Fusion EMR