Tuesday, January 26, 2010

BIDMC Data Marts

At BIDMC, our clinical systems are written in a hierarchical database called Cache - a very fast transactional system with great reliability and disaster recovery features.

However, for population health, quality, and performance analysis, we export our clinical care data into over 80 data marts build with SQL Server 2008.

These data marts are focused on specific reporting areas such as pharmacy, radiology, lab, and O.R. They are designed and maintained by an IS team within Clinical Information Systems. Updates generally occur daily and are managed via SSIS packages.

BIDMC data marts are used to support ad hoc queries by analysts as well as standard reporting via Performance Manager, a web-based, self-service reporting application developed by BIDMC IS. Some of the key content areas and uses for the data are shown in the graphic above.

One of our most powerful data marts is the BIDMC/Joslin Diabetes registry, which uses a Master Patient Index to link all the records of the two institutions together into one reporting infrastructure. It identifies all diabetic patients and consolidates a variety of relevant clinical and operational data into a single data mart optimized for tracking and reporting. Data elements include laboratory results Hemoglobin A1C and cholesterol, blood pressure, and outpatient medical and vision care appointments. In addition to data from BIDMC, the registry includes laboratory and vision care data from Joslin, providing a complete picture for BIDMC patients who also receive diabetic care at Joslin.

We've identified the person whom we believe is the Primary Care Physician and generated web-based reports for PCPs to validate that patients in the registry actually have diabetes. Surprisingly large number don't have diabetes. Diabetes was coded, for example, in ordering an A1C as a "rule out". We've also designed management reports that measure compliance with guidelines for diabetes care. We hope to leverage this infrastructure as part of our Beacon Communities grant application and make it available as a model for diabetic care throughout the community.

Our data marts, combined with the community quality data center hosted by MAeHC, provide us all the tools we need to improve quality and efficiency in our inpatient, ED, and ambulatory practices throughout greater Boston.

4 comments:

Been Porked said...

In the January 19th HealthLeaders Media, there is an article “Quality Reporting May Prove Challenging Under Meaningful Use”, that discusses the challenges that hospitals face.

Yet it would seem that your organization’s extensive data mart and reporting capabilities will enable you to meet the reporting requirement on the 35 quality measures to meet the 2011 meaningful use requirements, assuming complete and correct discrete clinical data is captured and stored somewhere. I am probably oversimplifying but is that a reasonable conclusion?

If hospitals do collect and store discrete clinical data somewhere, is the challenge integrating that data into a data mart or data warehouse? Is it the ease of reporting? The article says “Tripling the number of quality measures that Texas Health Resource's reports on in less than a year is going to be difficult, acknowledges Ed Marx, senior vice president and chief information officer. One of the challenges is the maturity of the technology to produce the types of reports required in just a few key strokes, Marx says.”

Thanks for the informative blogging.

Michael Kovner said...

John - Regarding the BIDMC/Joslin MPI, do you share a common registration system or do you use a patient matching algorithm like Initiate or NextGate to link the patient records?

John Halamka said...

We use Initiate as part of our EMPI. Joslin uses Nextgen and BIDMC uses a home built registration system.

WDavidStephenson said...

John: do you foresee a day in which all of the data mart results will be merged in real time? If so, what are the necessary steps to realize that? (my hidden agenda in asking that is I'm more and more convinced that designing an interactive data entry interface that would both flag erroneous data (or at least egregiously erroneous data) through a highly interactive process combined with business rules and that would automatically structure the data might allow this. But what do I know?