Google-Like Word Searches of Medical Records May Boost Patient Safety

Google-like searches of hospital medical records using English words and phrases found more patient injuries than the numerical billing codes used to rate quality of care, pointing toward a better way to highlight safety issues, a study found.

Researchers from the U.S. Department of Veterans Affairs reviewing patient records at six VA medical centers turned up 12 times as many cases of pneumonia and twice as many incidents of kidney failure and sepsis, a blood infection, when they used text searches instead of numeric codes linked to reimbursements.

Configuring commercial electronic medical record systems made by companies such as Cerner Corp. (CERN) for language searching would advance safer medical care, said Harvey Murff, the lead author of the study published today in the Journal of the American Medical Association.

“If you were going to try to develop systems to try to reduce these complications, you really probably would want to know as many as you have,” Murff, a physician at the VA Medical Center in Nashville, Tennessee, said in a telephone interview.

Administrative and billing codes weren’t designed to measure quality of care, said Nancy Foster, the vice president of quality and patient safety policy at the Chicago-based American Hospital Association, a trade group. The most accurate information should be used to hold hospitals accountable, with English text searches being preferable, she said.

Measuring Quality

The U.S. Centers for Medicare and Medicaid Services in Baltimore uses billing codes as a way to evaluate the quality of care at hospitals and publishes rankings online. Payments from Medicare, the U.S. health program for the elderly and disabled, are linked to quality measures under the 2010 health-care law.

Natural language searching isn’t widely available, said David McCallie, vice president of informatics for Kansas City, Missouri-based Cerner. Rewriting programs can destabilize software systems and requires more physician training, he said in a telephone interview yesterday.

Cerner has introduced “semantic search” to systems used by more than 30 of its customers, he said. The feature allows physicians to search patient records by medical keywords, and the company is developing a program that would enable hospital administrators to search across all records, he said.

Natural language searches using medical terms may turn up a higher rate of “false positives,” cases where harm didn’t occur, yet are preferable to billing codes that don’t record actual injuries, Murff said.

Scope of Study

Murff and his colleagues reviewed specially configured records allowing for text searching of 2,974 surgery patients from 1999 to 2006 at six VA centers for six complications including acute renal failure and pneumonia.

The searches turned up patients’ charts, physicians’ notes along with laboratory and imaging notes that weren’t accessible through billing codes, according to the study.

Electronic medical records “may let us get closer to the truth,” said David Classen, a professor at the University of Utah School of Medicine in Salt Lake City, in a telephone interview. Hospitals miss 90 percent of preventable patient injuries and infections when they rely on billing codes and administrative records, Classen reported in the journal Health Affairs in April.

The Global Trigger Tool, developed by the Cambridge, Massachusetts-based Institute for Healthcare Improvement, produces superior results and is being testing in a commercial electronic medical record at a health system for a future study, he said.

To contact the reporter on this story: Jeffrey Young in Washington at jyoung89@bloomberg.net.

To contact the editor responsible for this story: Adriel Bettelheim at abettelheim@bloomberg.net.

Press spacebar to pause and continue. Press esc to stop.

Bloomberg reserves the right to remove comments but is under no obligation to do so, or to explain individual moderation decisions.

Please enable JavaScript to view the comments powered by Disqus.