Clinical decision support (CDS) is meant to improve care quality by providing helpful alerts and advice to the electronic health record (EHR) user. However, too often the result is an annoying proliferation of pop-ups that only frustrate clinicians.

When the CDS system interrupts too much with alerts that are not useful, the result can be counterproductive. Clinicians routinely dismiss alerts. In the process, they may ignore those alerts that are useful, say researchers and hospital leaders.

The detrimental effects of poorly managed CDS are clear, says Ivana Jankovic, MD, an endocrinology fellow at Stanford University and lead author of a study concluding that CDS can contribute to clinician frustration and burnout.1 But CDS can be a valuable tool when managed properly. Jankovic and her co-author determined the negative effects can be mitigated by improving relevance, soliciting feedback, customizing, and measuring outcomes and metrics.

“From the research and from my own personal experience in working on these problems, it is clear that you want to get the end users, the clinicians, involved early in the planning process for an EHR with decision support. They’re going to be positioned to understand the workflow, how clinical decision support actually affects patients, and how it affects their day,” Jankovic says. “Once you have your decision support tool in place, you’re not done. You need to think about whether it’s doing what you think it’s doing and continuously solicit feedback to make sure it’s having the effects you’re hoping for, and not having any unintended consequences.”

Jankovic cautions administrators might not know how frustrated clinicians are with CDS. Clinicians may be aware of the problems and talk among themselves, but they may not feel they have a mechanism for reporting or helping address the problems. That was Jankovic’s experience with house staff. She helped establish a physician council to provide feedback to administrators.

“There is so much going on in the hospital that physicians just accept, like our frustration with electronic health records and decision support. But there is a great need from administration and EHR analysts to have that perspective,” Jankovic says. “You just need to build the [opportunities] for those groups to interact.”

Correcting the problems with CDS might require working with the EHR vendor to adjust the product. Jankovic says hospital leaders also can make significant improvements on their own. Some CDS issues involve integrating it properly into the hospital’s workflow. That is something that can be addressed in-house.

Sometimes, clinicians assume any problems with CDS are just baked into the EHR and cannot be fixed, so they routinely ignore alerts or develop workarounds, Jankovic notes. Hospital leaders should educate clinicians about the ability to change CDS and require ongoing training. Such education should go beyond more than just the initial few hours of training when the system is introduced.

Relevancy Key for Alerts

One of the biggest complaints is the CDS alert is not relevant to the patient, Jankovic says. For example, providers receive sepsis alerts for patients with a low predictive value for sepsis. Some alerts have override rates of 95% because the data behind the alert has not been updated to reflect current thinking.

“As an endocrinologist, sometimes when I’m ordering insulin appropriately with both long-acting and short-acting, I’ll get a duplicate medication order that pops up and I have to acknowledge,” Jankovic says. “The alerts are not doing what they’re intended to do. You’re getting a lot of alerts that are not relevant to your patient and you have to override them, which contributes to your alert fatigue.”

The relative scale of alerts also can be an issue. For a serious issue like prescribing a medication to which a patient is allergic, it is appropriate to include a full-stop, highly visible alert that must be acknowledged.

“But for an alert over a relatively minor issue, after the fourth alert that day, it starts getting really frustrating,” Jankovic says.

“It’s a balance of how much you’re willing to frustrate physicians to help patients. I think we’re all willing to take on a little frustration if it helps patients,” Jankovic continues. “But we’re not seeing the research that all of these alerts actually improve patient care. That’s a big reason for the disconnect.”

In other cases, the alert might be appropriate but it comes at the wrong time in the workflow. Jankovic recalls working in a patient’s prechart section of the EHR, before seeing the patient, and seeing an alert suggest recommending a flu shot to the patient.

“The patient is not here yet, but I can’t do anything else in the chart until I click on ‘order flu shot’ or ‘patient declined,’” she says. “It’s a helpful alert, and I’d be happy to use it at the right time, but it’s coming up at the wrong time and distracting me from the work I’m trying to do.”

Look for Little-Used Alerts

Assessing a CDS for potential problems can take different forms. Some hospitals conduct their own internal analyses, and others use automatic monitoring tools for the assessment, Jankovic notes.

No matter how one analyzes the CDS, the goals are the same. Look for what order sets are never used, what alerts are always overridden, any changes in the use of certain CDS tools after an EHR upgrade, or a change in the workflow, Jankovic suggests.

“If you’re noticing that certain alerts are being overridden 90% of the time or more, you can go to the users and the pharmacy or clinical leaders to do more of the feedback sessions that will tell you why,” she says. “That initial look at the raw data will tell you where the problems are, rather than having to do a lot of one-on-one interviews with the end users to have them tell you what alerts are problematic.”

In her research, Jankovic found patient outcomes are not factored in appropriately when assessing CDS. Some assessments may consider whether clinicians respond to alerts and carry out exactly what was suggested. If many users respond affirmatively, the alert is valid. But the analysis also should consider how that affected patient outcomes.

“We can see that they ordered the blood pressure medication in response to the prompt, but we should be looking further down the line to see if that improved blood pressure, and the next step to see if it reduced heart attacks, kidney disease, and stroke,” Jankovic says.

Contributes to Burnout

Frustration with CDS can contribute to burnout, says Richard Cohan, president of provider solutions with DrFirst, a healthcare software and IT company headquartered in Rockville, MD.

Aside from too many alerts that interfere with care, Cohan notes sometimes-useful CDS functions are not fully integrated into the EHR and workflow. These might be siloed, many clicks away from the way the clinician typically interacts with the system.

“Vendors sometimes forget the S part of CDS — the support. CDS should not tell them how to practice medicine but give them the support and full context around the information they need,” Cohan says. “Anything that is outside the workflow and not integrated causes context switching by the physician. Tools that are several clicks away from the workflow are a real irritant and a real demotivator.”

CDS also can integrate issues such as price transparency and prior authorizations, which are increasingly important to patients and government regulators. “That can be beneficial to providers, patients, payers, the health system. Patients have improved satisfaction, which is important to these health systems that are focused on Net Promoter Scores and quality measures,” Cohan says. “Everyone benefits by taking that friction out of the system, especially now that the effects of COVID-19 have made patients more sensitive to pricing issues.”

Cohan says a good CDS system will make data actionable, rather than just throwing up a lot of data for the clinician to wade through and determine how to use. The information also must be concise, visible, and intuitive. Screens are crowded, so important CDS guidance should be prominent and suggest the proper action to take.

Vendors should be pushed to adapt CDS to real-world conditions, making them more useful and less frustrating for clinicians, Cohan says. Many large vendors will respond to their client requests to improve their products. “By doing so, we improve everyone’s satisfaction with the process and the encounter,” Cohan says. “Hopefully, that impacts quality of care, improves stars and HEDIS measures, and outcomes for patients.”

CDS Systems Are ‘Tired’

KLAS Research, a healthcare IT data and insights company based in Orem, UT, recently issued a report that assessed CDS in seven different products.2 The co-authors say their research revealed a lot of room for improvement in CDS.

Joe Van De Graaff, vice president of digital health and security for KLAS, says traditional CDS is “a little tired.” That stems from good intentions but often unwise use of alerts.

“If my alarm clock goes off six times every morning before I need to get up, I eventually just don’t listen to the alarm clock anymore,” he says. “Having the right information at the right time, or even half of the information at the right time, is far more impactful than having all of the information at the wrong time.”

In addition, some elements of CDS can be too cumbersome to use in the moment. For instance, static lookup of information needed for diagnoses may be helpful but not as much as a prompt that is tailored to the patient information in the EHR.

Clinicians report they want that kind of information to pop up in useful ways derived from the specific patient’s data, says Jennifer Despain, director of market analysis for KLAS. They do not want to have to input data again to use the CDS resource.

With any CDS input, clinicians dislike jumping from window to window or input data that are in the record already. Truly useful CDS prompts will act on the patient’s specific data and provide meaningful options or suggestions to the clinician, all in one alert window.

“There is a trend toward making progress in this area, not as much as we want or hoped for, but not necessarily for lack of trying,” Van De Graaff says. “Much of that relates to the nuanced clinical workflow and the demanding needs of caregivers, and the limits of machine learning and artificial intelligence. We’re making progress, but we haven’t broken through the ice yet. There is some reluctance to that machine learning and how you can trust that without a clear decision tree or pathway.”

Work with Vendors

Hospital leaders can work to tailor a vendor’s product to their own needs. In most cases, they will need support from the vendor, Van De Graaff says. For example, vendors can help with true integration so a clinician does not have to go from one system to another to look something up.

Van De Graaff recalls one chief medical informatics officer at a hospital saying the best thing the vendor did for him was to provide all the available alerts and let his team decide which ones were best to include.

“Often, the assumption is that the EHR is built as is and it’s just going to do its job. But I think of how the real value in your smartphone is when you personalize it,” Van De Graaff says. “An iPhone or Android is good right out of the box, but you get really productive with it when you tailor it to your own needs and what you want from that device.”

Clinicians who are unhappy with the CDS probably will make their feelings known to hospital leaders. Van De Graaff cautions people can be satisfied with an existing system simply because they are familiar with it. They are aware of its faults and have come to accept them, using workarounds and just resigning themselves to clicking past the irrelevant alerts. That does not mean the system is working effectively and contributing to better patient outcomes.

Address Reluctance

Involving physicians and nurses in the CDS development process is key to success, says JD Tyler, MD, chief medical officer at Tissue Analytics, a company in Baltimore providing technology to address wound healing. He is a former hospitalist and chief medical officer for a company that provides a range of specialty EHRs, and other technologies, including CDS, to hospitals and other facilities nationwide.

“With that background, I can state unequivocally that the most important thing we can do to tailor CDS for physicians is to make sure it helps physicians work instead of adding their workload,” he says. “The philosophy of most physicians is that ‘if it’s already in the medical record, don’t make me have to input it again.’”

Another important issue is the inherent resistance of some clinicians to using CDS. There remains considerable concern among many clinicians about CDS, specifically that it takes the “art” out of medicine, Tyler says. “I disagree. The art of medicine is how you talk to patients, gather information, interpret data. Your decision-making should be scientific and objective to drive optimal clinical decisions. That’s what CDS brings,” he says. “CDS isn’t replacing you. You are still behind the wheel, and you can disagree. When well-designed, supported, and fully utilized, CDS is an important tool for improving the quality of care for all patients and helping to ease the workload for busy physicians.”


  1. Jankovic I, Chen JH. Clinical decision support and implications for the clinician burnout crisis. Yearb Med Inform 2020;29:145-154.
  2. VanDeGraaff J, Despain J. Clinical Decision Support 2020: Progress and Innovations in CDS. KLAS Research. May 5, 2020.


  • Richard Cohan, President, Provider Solutions, DrFirst, Rockville, MD. Phone: (888) 271-9898.
  • Jennifer Despain, Director, Market Analysis, KLAS Research, Orem, UT. Email:
  • Ivana Jankovic, MD, Endocrinology Fellow, Stanford University School of Medicine. Email:
  • JD Tyler, MD, Chief Medical Officer, Tissue Analytics, Baltimore. Phone: (647) 362-4684.
  • Joe Van De Graaff, Vice President, Digital Health and Security, KLAS Research, Orem, UT. Email: