Search is on to cure EHR alert fatigue
■ Researchers, physicians and others are coming up with strategies to ensure that warning alerts generated by health IT systems are less frequent and more meaningful.
- WITH THIS STORY:
- » External links
Almost every physician who has typed orders into an electronic health record or e-prescribing system probably has experienced so-called alert fatigue — the frustration of warning after warning popping up before that order is accepted. The alerts are designed to inform physicians of possible patient safety issues, but their frequency and often lack of necessity make them the electronic equivalent of the boy who cried wolf.
As researchers and health care organizations work to alleviate alert fatigue, it’s clear that the answer is to create systems that take human behavior and supplemental patient data into account when writing rules that decide when and why an alert is fired off. That way, the alerts could have more success in their purpose: protecting patient safety.
“The issue of alert fatigue is that we are not just sending alerts that are considered to be irrelevant, but when you get enough of those, even the alerts that are important are disregarded,” said Jon D. Duke, MD, assistant professor at the Indiana University School of Medicine. Several studies have found the ignore rate for alerts averages between 80% and 90%.
A Dept. of Veterans Affairs-funded study in the April issue of the International Journal of Medical Informatics sought to find the reasons behind a prescriber’s interaction with an alert to see why that person would ignore it and how to ensure more meaningful warnings.
Researchers from the VA and Regenstrief Institute in Indianapolis observed 30 prescribers, which included physicians, clinical pharmacists and nurse practitioners, across five primary care clinics and eight specialty clinics as they ordered prescriptions to determine how they reacted to alerts. Altogether, there were 340 alerts. From the prescribers’ actions, researchers identified 44 emergent themes and nine overarching factors that described the alert interactions, giving some idea as to how complicated it can get to solve the problem of alert fatigue.
However, researchers found a few themes that cut across almost every issue. Often, prescribers were unsure why an alert was generated. Many alerts were pharmacist-designed even though the physician or nurse practitioner was doing the prescribing. The research also found that many alerts were overridden because they weren’t specific to a particular patient.
Alissa Russ, research scientist with the Dept. of Veterans Affairs and co-author of the study, said that by looking at the human factors associated with computerized physician-order entry system use, she’s confident that systems can be designed to work universally well with physicians, pharmacists and nurse practitioners.
Russ said she and her team have created two experimental alert designs that are being evaluated in simulation labs within the VA. Her team is analyzing data on the effectiveness of those experimental designs.
“The paper gave a broad look at the reasons for alert fatigue from a physician’s perspective and gave broad recommendations,” said Dr. Duke, a research scientist of medical informatics at Regenstrief Institute, which is affiliated with the Indiana medical school. “What we are looking at now is how technically do you implement some of these changes.” He did not participate directly in the VA study.
Adding in patient data
To tailor alerts to the physician, patient and situation, many hospitals are looking at physician behavior, then adding discrete patient data to design customized rules for when an alert appears.
Dr. Duke said Regenstrief and the VA are fortunate to have home-grown systems that are easily customized. He said Regenstrief is able to analyze physicians’ behaviors toward certain alerts (if there is an alert a doctor routinely cancels out, for example) and adjusts the alerts according to those behaviors.
Regenstrief also has created a Context-Aware Drug-to-Drug Interaction (CADDI) alert system that uses key data elements in the EHR to produce meaningful alerts.
“For example, if there’s an interaction between a medication like an ACE inhibitor and a potassium supplement, which is a common interaction associated with hyperkalemia, the system goes and checks the [patient’s] last potassium level and is able to then increase or decrease the intensity of the alert based on whether there was any abnormal baseline levels,” he said.
Adventist Health System Chief Medical Informations Officer Phil Smith, MD, said that at his hospital system, which is headquartered in Altamonte Springs, Fla., and has facilities in 10 states, alerts are becoming easier to customize as more data are available in the EHR systems to help refine them. For example, now that 90% of Adventist patients have a problem list in their records, the hospitals in the system can refine the alerts to be relevant to patients with particular conditions.
When Adventist first went live with its CPOE system in 2010, the system produced 84 alerts per every 100 medication orders, “which is horrible,” Dr. Smith said. Six weeks later, they got it down to 34 alerts per 100 medication orders, and doctors were making changes based on the alerts 50% of the time.
When the hospital system could not get the alert frequency below 22 alerts per 100 medication orders, Dr. Smith said, it went back to its vendor for technical tools to help cut the alert frequency even more. The system is now down to 14 alerts per 100 medication orders, and generally eight of those result in physician action.
“By giving less nuisance alerts, you’ll get doctors to pay attention to the ones they should pay attention to,” Dr. Smith said. The impact on patient safety is apparent, he said. Last year, one in 10 of the medication orders had an alert for which the physicians made a change, resulting in 500,000 “patient rescues.” A patient rescue means the doctor changed an order that could have had a bad result.
Lee Lemelson, vice president of clinical applications for Phoenix-based Banner Health, co-wrote a 1998 report that looked at how alerts help prevent injuries. He said the findings are still relevant today. His philosophy is that he rather would have a physician complain about the alert firing too often rather than not alert them and have injury occur.
But, he said, advancements in technology since that study was published have made it easier to customize alerts in a way that patient safety is improved rather than compromised.
At Banner, clinical performance groups meet regularly to discuss rules that can be incorporated, or that need to be refined, to improve patient care. Those modifications are evidence-based and factor in best practices. But in an area as complex as medicine, there always will be a need for more refinement, Lemelson said.
“As we continue to look at medical evidence and best practices throughout our entire medical spectrum, more and more & alerts will be written in the best interest of patient care.”