No matter what you call it, the Clinical Looking Glass project, headed by Eran Bellin, M.D., is taking data mining to the next level. The application, 10 years in the making, is enabling some 250 physicians to conduct their own ad hoc research studies. Some are as simple as identifying all patients taking a drug that has been recalled. Others are far more complex, such as assessing whether a certain type of filter is beneficial to patients with blood clots.
Bellin heads the project in his role as vice president, clinical I.T. research and development, at Emerging Health Information Technology, a for-profit unit of Montefiore. The unit coordinates I.T. for Montefiore's four hospitals serving The Bronx and nearby Westchester County as well as its 21 community health centers, plus other area hospitals.
How It Began
The project began when Bellin and his staff determined that it was difficult, if not impossible, to extract research data from its core clinical information system, from GE Healthcare, Waukesha, Wis. The I.T. team searched for an adequate data mining tool to aid with the effort, but couldn't find one. So they built their own.
Now Emerging Health Information Technology is continuing development in a collaborative project with the U.S. Department of Defense at a yet-to-be-named Washington-area military hospital. Eventually, the hospital's technology development arm hopes to market the data mining software to other hospitals, Bellin acknowledges.
Montefiore's lack of a comprehensive electronic health record is not impeding its outcomes research efforts. The organization has yet to automate the clinical notes of nurses and physicians at the hospitals. And it's only in the very early stages of deploying electronic health records software from GE at its clinics. Nevertheless, it's surging ahead with outcomes research based on the wealth of data it has already gathered.
"To improve the quality of care, you need to be able to find patients, determine if they are doing well, intervene as needed, and then follow up to see how they are doing," Bellin says. To accomplish those steps requires powerful data mining software, he argues. Equally important, he says, is a culture "that's open to the possibilities."
So far, the organization has spent $15 million on the project, says Steven M. Safyer, M.D., Montefiore's president and CEO. One motivation for the major investment, he says, is that the provider needs to better manage patient care under its $650 million in annual at-risk, or capitated, contracts with multiple payers, he explains. "What we aim to do is to manage the care for our patient population long before they need hospitalization," he says. "And that requires data to design programs that improve their health."
A key to the success of the project so far, Safyer says, is that physicians can use the technology on their own. "We knew that if we couldn't involve physicians in this project, the project would fall flat on its face. So we demonstrated to physicians that this product brought real value to the care they are providing."
How It Works
Montefiore's clinical information system gathers a wide variety of data, including laboratory, admission/discharge/transfer, radiology and medication information. To support the research efforts without slowing down transactions related to patient care, the organization creates a duplicate database to fuel research using the Clinical Looking Glass application. "We reorganize the data to fit our data model and to get the speed we need to allow ad hoc studies," Bellin says.
To create the searchable database, Montefiore uses a mix of standards, including ICD-9 billing codes and a proprietary disease classification method for problem lists from the University of Nebraska.The data mining system, which uses business intelligence, enables doctors to ask complex, "clinically meaningful" questions rather than relying on a limited set of pre-determined queries, Bellin explains.
Accessing the Data
Physicians access the data mining software via Montefiore's intranet using a Web browser. They receive about two to three hours of training to get started. Doctors can initiate a study by answering a series of questions selected from checkboxes.
To initiate a study, a doctor defines a "cohort" of patients, based on such factors as age, race, gender, medications, diagnosis and physician, Bellin explains. Then the doctor uses the application to conduct a study of what happened to those patients over a defined period of time, such as, for example, how often they were re-hospitalized.
In addition to tracking the activity for individual patients, the software conducts aggregate studies for the entire cohort. It can compare those results to a control group of other Montefiore patients.
Before the data mining application was available, a study to analyze lengths of stays and readmission rates would take as long as six months to complete, Bellin says. The application can crunch the necessary numbers for such a study in about 15 seconds.
"In most data mining systems, you have only a priesthood that can use it" because of the need for expert training, Bellin contents. "In our system, every smart doctor or nurse or administrator who takes the training can ask their own questions."
Montefiore has used the data mining software to support an effort to reward some of its 2,000 physicians, most of whom are salaried staff members, for their performance.
Since 1996, the organization has paid certain physicians annual bonuses of up to $3,200 for following diabetes protocols, explains Arthur Hopkins, M.D., a medical director for Montefiore Medical Group. For years, that required tedious manual chart audits. But now the data mining software greatly eases the analysis, enabling Hopkins to "peel back layers of the onion" to carefully analyze how each clinic and each doctor scores on the control of diabetes symptoms.
"Within each of our divisions, some of our doctors are doing extraordinarily well," he notes. "So we're taking a high-performing physician on the road to give case presentations to offices that are not doing as well."
Be the first to comment on this post using the section below.