A few weeks ago, I was asked to describe the way we do performance measurement at BIDMC.
Our strategy is simple - all our clinical systems are built on hierarchical databases and our decision support databases are relational, updated by nightly extracts of our clinical systems.
Clinical data is inherently hierarchical - patients have many visits, with many labs, with many observations. This creates a tree of data - a hierarchy that does not fit nicely into columns and rows. Doctors typically ask questions about individual patients, such as their most current results. Retrieving data the way it is stored, hierarchically, is blazingly fast. We retrieve patient specific date from terabytes of historical information in milliseconds.
Hierarchical data is great for clinical care, but not so wonderful for decision support. Asking a question such as "how many patients in the past 10 years have had a creatinine of 2.0 and a Hemoglobin A1c greater than 9.0" would require that every lab result ever done be examined one at a time.
For population health and performance measurement, querying an indexed relational database that is optimized for reporting makes the most sense. To enable this, we've created numerous data marts based on nightly extracts of our clinical and financial systems. Our current data marts include admissions, ED, outpatient appointments, OR, laboratory, microbiology, blood bank, radiology, cardiology procedures, inpatient pharmacy, outpatient medications, billing and payroll.
We use these data marts for 3 kinds of queries:
1. Expert analysis using relational query tools such as SAS, Access, and SQL Reporting services. We have a team of decision support professionals reporting to the CFO and a team of dedicated IS analysts performing these queries. Creating such queries requires an understanding of the quality of the underlying data, its source, and its meaning. Years ago, I hired a new analyst who noted that the average length of stay in the operating room was 120 days. The person did not know that the length of stay in ORs is measured in minutes
2. Parameterized web-based reports that can be run by anyone. When a new report has been developed by an analyst and is ready for broader distribution, we go through a process to make it available via our web-based Performance Manager tool. This typically involves developing result rollups to enhance performance, writing parameterized stored procedures, and developing a web page interface that allows users to select parameters via dropdowns and checkboxes and to easily navigate drilldowns and trending. Performance manager has over 150 web-based reports which allow untrained users to create accurate reports and explore results at the touch of a button. Reports include financial performance ( discharged not final billed , ED and inpatient volumes , gross patient services revenue by cost center ), clinical performance ( antibiotic resistance and sensitivity , ED throughput ) and even IS uptime .
3. Self service queries via a drag and drop tool. In 2008, we're launching a graphical tool which enables user defined queries of our datamarts for clinical research. This tool will returns counts of patients that can be used for pre-research investigation such as ensuring enough data is available to conduct an actual study. IRB approval would be required for any more detailed information. The drag and drop interface includes enough metadata to limit the queries to those data elements which make logical sense, building in the expertise of our analysts but allowing untrained researchers to do patient de-identified data mining, protecting patient confidentiality.
Using these three methods, we provide the data needed to empower our quality reviews, our process improvement efforts, and clinical research. Of course, we do not sell data or ever release data to third parties without patient consent. All our secondary uses of data are reviewed by our privacy officer, our IRB, and our security professionals. In this way, we ensure that all data in our enterprise is used on a need to know basis, following HIPAA to the letter of the law.