The twin premises underlying President Obama’s decision to allocate billions of dollars towards national deployment of electronic health records are that they will improve the quality of care and reduce the costs of delivering it.
Hundreds of studies have validated these assumptions, but a new study from the VA Medical Center in Houston reminds us that EHRs do not eliminate all errors, including some they were supposed to catch.
The study was performed by Hardeep Singh and colleagues at an outpatient facility between November 2007 and June 2008. As we have stated in the past, VistA, the VA’s EHR, is among the most comprehensive, respected and widely deployed legacy systems in extant.
The scientists looked at what happened when VA physicians were prompted by an alert that had been triggered by abnormal findings on a medical imaging study, including those which are potentially quite serious.
In VistA, the alert remains on the computer screen of the primary care-giving physician for 2 weeks unless the physician acknowledges that action has been taken. After 2 weeks, the alert disappears.
During the study, 123,638 imaging studies were carried out, and 1,196 alerts were generated.
Remarkably, 217 (18.1%) of the alerts were never opened, and 92 (7.7%) patients did not get timely follow-up, which might include scheduling additional tests or notifying the patient. What is more, the incidence of insufficient follow-up was the same for cases in which the alerts were never opened as it was when the alerts were opened.
In many cases, the lack of timely follow-up was associated with deterioration in the patient’s condition.
A mildly counterintuitive finding was that, when more than one physician was alerted about an abnormality, the alert was less likely to receive attention, presumably because physicians assumed their colleagues would follow-up.
The troubling trends were ameliorated when a radiologist called the primary care-giving physician about the abnormal result. Unfortunately, the study design did not permit investigators to determine whether this subset of cases included all potentially dangerous or progressive abnormalities.
In their write-up, which appears in the Archives of Internal Medicine, the authors drew 4 conclusions:
1) Further thought needs to be given to the frequency with which EHRs generate alerts, and the nature of these alerts. If physicians are bombarded with alerts, they are to going to ignore some of them.
2) In instances where more than one physician receives an alert, there needs to be a protocol that specifies which one should take action.
3) Unopened alerts should remain on physician' screens longer than 2 weeks, "perhaps even indefinitely, and should require the healthcare provider's signature and statement of action before they are allowed to drop off from the screen."
4) The EHR should initiate some kind of super-alert when physicians haven’t responded to alerts.
The humbling study should not change our beliefs and aspirations for the long-term benefits of EHRs, but it does remind us that health care delivery is and will remain a human process. And humans make mistakes.
Glenn Laffel MD, PhD
Sr. VP, Clinical Affairs, Practice Fusion
Friday, October 2, 2009
EHR-Based Alerts Don't Assure High Quality
Subscribe to:
Post Comments (Atom)
Posts
3 comments:
Seems like part of the problem is the delivery of the alert. I like the "super alert" system. Perhaps this means that all forms of communication are utilized: email, automated phone message, skype (for international docs), chat, Twitter (why not, thousand of docs are on Twitter), AND, especially, MOBILE!!
Seems rather embarassing that the authors, after all this hard work, forgot the obvious solution. The answer is not dream up more ways to harass the poor doctors, it's to tell THE PATIENT. Kaiser does this already and it works quite well. Perhaps, if we were not so protective of our privelaged postion of keepers of health information, they would have thought of this too.
FM Wodajo, MD
@OrthoOnc
In managing thousands of servers at once, our solution was a metrics and monitoring system. For each type of "thing" (say memory) on each type of machine (say webservers) we had a threshold.
Those alerts, when they happened, were not spurious. So we knew when they triggered that something was wrong. The problem with these systems often is that they're not tuned, so they generate lots of "white noise" and people ignore what they spit out. Right about the time you chase down your 3rd false positive of the month, you stop trusting the system.
If we know the system is spitting out meaningful stuff AND the right people are alerted (the admin AND for certain kinds of meaningful information, the customer), then everyone can be kept abreast of the problem. That's not to say that, in a pool of redundant servers, the customer has to be aware of a server dropping out. But they should be made aware, automatically, when the entire service is down.
The theories and architectures behind computer monitoring and metrics systems are pretty much the same whether you're monitoring output from servers or from people: you're monitoring numbers being reported. Apply how we handle thousands of machines using minimal staff to how we can handle hundreds of people.
Post a Comment