But bar codes make new problems and aren't the panacea that safety advocates expected, a research team lead by Ross Koppel of the University of Pennsylvania School of Medicine has concluded.

In a first-of-its-kind study, Koppel and colleagues from the University of Wisconsin and the Main Line Health System spent several years observing the use of bar-code technology in five hospitals. They also analyzed half a million medication scans to examine how well such systems worked to reduce medication errors.

Bar coding has not yet been proven to reduce medication errors, and often the shortcuts that caregivers develop undermine its effectiveness, the researchers found.

"A lot of the time, the people who develop these systems think of them in the same way they think about stocking a Sam's Club," Koppel said in an interview yesterday, "but hospitals are very active places where serious things are happening all the time."

David B. Nash, chairman of the department of health policy at Jefferson Medical College, said, "A bar code is swell, but without a culture of safety, it is not nearly as valuable as everybody believes."

The issue is critical because medication errors are believed to be the most common mistake in U.S. hospitals. In 2006, the Institute of Medicine, a nonprofit group that advises the federal government, estimated that at least 400,000 preventable medication errors occurred each year in U.S. hospitals, resulting in about 9,000 deaths and $3.5 billion in extra cost.

About one-third of the nation's hospitals currently have a bar-code medication-dispensing system, but most if not all hospitals are expected to install such a system in the next few years, Koppel said.

The software costs several hundred thousand dollars both to install and connect to many hospital systems, such as the pharmacy, he said.

In hospitals that use bar codes, each patient gets a unique identifier that must be scanned when he or she is given medication. Each drug is also bar-coded and scanned. The computer then confirms that the proper medication and dose are being given to the right patient.

The study - released online yesterday in the Journal of the American Medical Informatics Association - found practical problems large and small, leading nurses to develop work-arounds that often undermined the system's safeguards.

Koppel cited an example where four patients need a medication that requires refrigeration, such as insulin. If the fridge with the drug is on a different floor, a nurse might copy each patient's bar code to pick up the insulin doses and do the scans far from the patients' bedsides.

By carrying four drugs with different doses on the same tray, "you have now completely obviated the protections of the bar code," Koppel said.

The bar-code system consists of a handheld device for scanning and a mobile computer that can be taken into rooms.

But sometimes the computer can't make it into patient rooms crowded with lifesaving equipment or because the patient is isolated with a contagious infection.

So the nurse takes just the portable scanner into the room - covered with a plastic bag if the patient is contagious, scans the wrist-band bar code and the medication codes, and gives the drugs without hearing the alarm on the computer in the hallway indicating that it is the wrong patient.

The researchers found all kinds of improvisation occurring. Patient bar-code ID bands were taped to doorjambs, nurses' desks, scanners, drug-dispensing machines and clipboards, and were also worn as key chains on belts, among other places.

There were times when the battery on the computer or the scanner died. Or there was a wireless dead-spot in the hospital room.

"In 99 percent of the cases, it ain't the nurse, it is the system," Koppel said, "and sometimes it is easily fixed by reprogramming."

Nurses overrode patient ID scans 4.2 percent of the time, the study reported, often because bar codes were unreadable for reasons ranging from the child who chewed on hers to the patient with dementia who tore his off to the ones smudged and soggy from blood, urine or feces.

Despite the problems with bar coding, Koppel said the answer wasn't to scrap the idea but to figure out ways to make it work.

Koppel said that by following eight key recommendations, four of the hospitals where the study was conducted were able to greatly reduce errors.

He said the two most important were for hospital executives to spend time actually watching how the systems are used and to demand that the vendors design systems for the real world of care.

"Nothing beats looking at what actually happens on the hospital floors," Koppel said.