When it comes to getting clinicians to buy into using electronic prescribing systems, it's important to focus on quality over quantity of alerts, according to healthcare researcher Nidhi Shah, M.D.
Shah was the lead author of a recent journal article that examined the positives and negatives of an e-prescribing system in place at 735-bed Brigham and Women's Hospital in Boston, where Shah used to be a general medicine research fellow. While there, Shah and colleagues designed a system for Brigham and Women's with a selective set of alerts designed to improve workflow by allowing only high-severity alerts to interrupt prescribing. Shah now is an attending physician at 830-bed Yale-New Haven (Conn.) Hospital.
Doctors have shown resistance to electronic prescribing because of workflow interruptions caused by clinically irrelevant medication alerts that are either overridden or ignored. Since these actions defeat one of the main purposes of using such systems in the first place-decision support-Shah said a balance needs to be found between safety and efficiency concerns.
"That was one of the major objectives in designing our alert system, and one major criticism physicians who use electronic prescribing systems have is that they are overburdened by alerts that are not useful to them," Shah said. "I definitely think more work needs to be done to find the perfect balance."
At Brigham, there were two types of alerts: noninterruptive, which generated a warning in the upper-left corner of the clinician's computer screen; and interruptive, which wouldn't allow clinicians to finish a prescription until they responded to a pop-up box on their screen-either by changing their order or listing a reason why they would prescribe a medication despite the warning presented in the alert.
The system was used by 701 clinicians in 31 Boston-area primary-care practices affiliated with Massachusetts General and Brigham and Women's hospitals. Prescriptions from those clinicians generated 18,115 drug alerts from Aug. 5, 2004, to Jan. 5, 2005.
Of the 5,182 interruptive alerts, 67% were heeded and not overridden, according to Shah's report, which was published in the January-February issue of the Journal of the American Medical Informatics Association.
"Most studies have been showing low user acceptance," Shah said. She said that previous studies have shown that clinicians would only accept the recommendation given in an alert 8% to 10% of the time. "This is one of the first studies that shows systems can be developed with high rates of acceptance," she said.
"I think one of our major findings is that it is possible-despite all of the criticism of computer-prescribing and support systems-to develop decision-support systems that clinicians will be supportive of, and that's a lesson learned to be taken from this study," she said. "We also found that it is very important to collect the reasons why clinicians override alerts. These are things we took back to revise our alerts in the future."
Common reasons for overrides included: alerts were sent for a negative drug-drug interaction for medications patients were no longer taking-this usually happened when a patient's drug list had not been updated; new evidence had been found to support certain drug combinations but the alert system hadn't been modified to reflect this; or there was a lack of an alternative medication.
Inapplicable alerts can cause problems, too. Dean Sittig, director of applied research and informatics for Kaiser Permanente Northwest in Portland, Ore., said an alert may warn a physician entering a prescription for Valium that the drug should not be prescribed for an elderly patient, but 75% of the time, the prescription will not be for an elderly patient. This causes the doctor to get in the habit of ignoring their system's decision-support tools.
"Doctors are not ignoring good advice, they're ignoring bad advice and it's appropriate to ignore bad advice," said Sittig, one of the authors of the book Improving Outcomes with Clinical Decision Support. "I've been saying for a long time that, if you increase the specificity of your alerts, people would follow them. This report is the first proof that it can be done."
Because the system was implemented as part of an overall quality-improvement project and not just for the study, Shah said the researchers were not able to come up with a pre-study set of statistics measuring clinician acceptance. She also said that instead of using a commercially available prescribing program, a "homegrown" system was used.
The results of the study may modify the current more-alerts-are-better thinking that guides many commercial systems, Shah said. At the very least, she said it shows the importance of giving customers the ability to modify systems to meet their specific practices.
Sittig said vendors are too wary of liability concerns and will warn physicians of hazards that have the remotest chances of occurring, which leads some healthcare organizations to turn off alert systems altogether.
"Vendors don't want to turn off alerts because they're afraid of liability, but healthcare providers take risks every day," he said. "They know about the potential interactions, but they also know the risks are small and their patient needs the medicine."