At the St. John Medical Center in Tulsa, Okla., physicians analyze blood-transfusion records stored in a data warehouse to better understand when transfusions are most effective and why some patients' bodies react badly to the procedure. The knowledge has helped the hospital decrease the number of transfusions that lead to negative reactions by 18% and reduce the number of transfusions performed by 22%, saving $1.4 million annually.

It's an example of how analytics will change the health-care industry. "We believe this is our greatest opportunity to make a difference in how we deliver health care," says Dr. Terry Dolan, president of St. John.

The use of business-intelligence and data-analysis software in the health-care industry isn't new. Such tools have been a mainstay on the financial and administrative side for a decade or more. Florida Hospital in Orlando built a predictive-modeling system based on IBM Intelligent Miner that identifies which patient bad debts the hospital should collect on its own and which ones should be turned over to a collection agency, saving the hospital $200,000 annually in agency fees. Lee Memorial Health System, which operates health-care facilities in the Fort Meyers, Fla., area, analyzes patient admission and discharge data to help administrators adjust staffing for the busiest and slowest times of the day and reduce costs.

Until recently, however, business-intelligence technology hasn't played a big role in delivering care. That's changing, as health-care companies build data repositories with information culled from operational systems used to record patient admissions and discharges, bill patients and insurers, order laboratory and radiological tests, and dispense medications. The data can be analyzed to judge alternative treatments such as heart bypass surgery versus angioplasty by looking at how quickly patients are discharged and whether they're readmitted. They can also study which medicines work best, spot previously unrecognized disease patterns, identify at-risk patients, and even review the performance of individual physicians.

Shands Healthcare, a hospital network in Florida affiliated with the University of Florida Health Science Center, has built a database of all the organ-transplant operations performed by its physicians. Clinicians are studying transplant-patient survival rates and other statistics using reporting and analysis software from Business Objects SA to learn where improvements can be made. BayCare Health System Inc., an alliance of seven hospitals in the Tampa Bay area, is building a database of clinical outcomes of patients treated for acute myocardial infarction, heart failure, and pneumonia. Using data visualization and analysis software from Dimensional Insight Inc., clinicians can compare the results of those medical procedures against benchmarks set by national organizations such as the Joint Commission on Accreditation of Healthcare Organizations. Those comparisons can be done for hospitals, clinical units, and individual doctors.

Already, one BayCare hospital, after analyzing procedures for treating suspected heart attack victims, has cut 20 minutes off the time it takes to diagnose and process a heart patient, from entering the emergency room to reaching the catheterization lab for an angioplasty, says Cindy Righter, clinical outcomes director. BayCare is expanding the database system to cover other procedures, including stroke treatments, joint replacements, and open-heart surgery.

The biggest obstacle remains getting medical data for analysis. Many records, especially patient notes taken by doctors and nurses, are still on paper and useless for analysis. Less than 15% of all clinical data today is digital, Frost & Sullivan analyst Amith Viswanathan estimates. What data is digitized, including orders for lab and radiological tests and pharmaceutical orders, is formatted for billing purposes, not clinical analysis. And clinical IT systems such as lab and pharmacy applications are often standalone, incompatible with other clinical apps, which makes data interchange difficult, Forrester Research analyst Eric Brown says.