Clinical alerts may help reduce dangerous drug interactions, but sometimes they cause more problems than they solve. Experts describe best practices to make them work better.
Clinicians, whether young or old, technophobes or technophiles, continue to complain about the avalanche of unnecessary alerts they get when walking through a clinical decision support or e-prescribing system. It's not uncommon to hear a cardiologist, for instance, complain: "I've been practicing for 15 years. I don't need to be cautioned about ordering aspirin for a patient at risk of hemorrhagic stroke."
On the flip side, IT leaders and clinicians worry that these systems miss needed alerts because they're incapable of taking into account important free text data from clinicians' notes. Allison McCoy and her colleagues at the Department of Biomedical Informatics, Vanderbilt University School of Medicine, give a good example in a recent Journal of the American Medical Informatics Association (JAMIA) report.. They give the example of an infectious disease specialist who orders a powerful antibiotic for a life-threatening infection in a patient with compromised kidney function, and who inserts that fact into the clinician's notes.
In a situation like this one, it's likely that clinicians would by default get a warning to discontinue such a drug because it can be toxic to the kidneys. But instead, the clinical decision support system should alert doctors to monitor renal function closely. The risk of dying from the infection is far greater than the risk of renal damage from the antibiotic. But most CDSSs aren't that smart, yet.
It's also important to have an alert system that puts more emphasis on evidence-based rules and algorithms than it does on expert medical opinion. As MCoy and her colleagues explain, some alerts are just so well documented and too important to ignore. Their JAMIA report points out, for instance, that "an alert to discontinue ongoing oral potassium supplements and to monitor serum potassium levels for a patient whose potassium level had gradually increased to 6.0 mEq/l should rarely be overridden."
Of course, any alert system tweaking is going to involve tradeoffs between false positives and false negatives. If an alert is too sensitive, it may warn caregivers about threats to patient safety that are very unlikely to occur in the real world--a false positive. On the other hand, if the system only warns clinicians about the most severe, life-threatening dangers, they're going to have a lot of living but very sick patients on their hands as a result.
Similarly, every system needs to strike a balance between sensitivity and specificity. Typically, the more sensitive the system is to detecting dangers that are real, considered true positives, the less specific it is--that is, the less capable it is of ruling out those dangers that aren't, also called true negatives.
What Do The Experts Say?
At our recent InformationWeek Healthcare IT Leadership Forum, Johns Hopkins Hospital's CIO and CMIO talked about how their organization is solving the alert fatigue problem.
CIO Stephanie Reel said the hospital's alert technologies, tools, and protocols "can identify the fact that someone has seen an alert before for a specific patient and doesn't need to see it again." The hospital also tailors certain alerts to different categories of clinicians with different skill sets. "An experienced cardiologist doesn't need to be alerted about an interaction between Coumadin and aspirin," Reel said, because it's assumed that a practitioner with this level of experience already knows that fact.
Another way Johns Hopkins deals with too many alerts is to devise treatment protocols that reduce the need for alerts in the first place. The hospital does this by creating "smart order sets."Peter Greene, MD, Johns Hopkins' CMIO, used the example of ordering postoperative insulin for a patient. There are now several hundred different options on how to order a diabetic regimen in such a situation.
So instead of having the physician walk through countless screens and alerts, a lot of the programming is done behind the scenes, inserting several key pieces of information about the patient that are already in his e-chart to help find the right insulin regimen. So in the end the ordering physician needs only to answer a few questions before the CDSS makes a very specific recommendation.
At University of Pittsburgh Medical Center, the CDS and e-prescribing systems are also getting smarter. Daniel Martich, MD, the center's CMIO, explained in an email that UPMC also directs specific alerts to specific provider groups. In certain inpatient situations, alerts about ordering or canceling telemetry will go only to the attending physician and his or her team, not to consultants.
Similarly, Martich explained that "we fire certain specific medication interactions on the inpatient side only for the prescribing pharmacist, and not the ordering physician in CPOE."
Other specific, evidence-based alerts that UPMC pushes out include one about the risks of anticoagulants (sent to clinicians only if these drugs are ordered for a patient with an epidural catheter in place), and one that alerts the clinician about the need to adjust the dosage when placing NPO (nothing by mouth) orders for a patient receiving insulin or oral anti-hyperglycemic agents.
Of course, as your IT department thinks about ways to make its alert system smarter, you have to maintain reasonable expectations. Not every healthcare provider has the deep pockets of a Johns Hopkins or UPMC. Customizing these programs can get expensive, requiring clinical experts to review the algorithms and IT staff to rewrite the code.
But as your organization balances costs and benefits, keep in mind that smarter decision support systems will help clinicians concentrate on the most important issues and reduce expensive medical errors. It's not too hard to believe that more dollars spent on such IT initiatives will translate into fewer dollars going out the door in malpractice settlements.
Find out how health IT leaders are dealing with the industry's pain points, from allowing unfettered patient data access to sharing electronic records. Also in the new, all-digital issue of InformationWeek Healthcare: There needs to be better e-communication between technologists and clinicians. Download the issue now. (Free registration required.)
InformationWeek Tech Digest, Nov. 10, 2014Just 30% of respondents to our new survey say their companies are very or extremely effective at identifying critical data and analyzing it to make decisions, down from 42% in 2013. What gives?