Moving the physicians of any organization forward to embrace clinical decision support for the ordering of diagnostic imaging procedures is inevitably a journey, healthcare industry leaders agree. That certainly has been the case at Froedtert Health, the Milwaukee-based integrated health system that encompasses three hospitals and more than 25 primary and specialty care health centers and clinics, in southeastern Wisconsin, as well as the Medical College of Wisconsin. The three inpatient facilities that are part of the regional health network are Froedtert Hospital (Milwaukee), Community Memorial Hospital (Menomonee Falls), and St. Joseph Hospital (West Bend). According to its website, the network’s three hospitals encompass 784 staffed beds, nearly 40,000 annual admissions, and more than 900,000 annual outpatient visits.
Helping to lead change around clinical decision support at Froedtert Health has been CMIO Robert Donnell, M.D., a physician who has been in the CMIO for four years, and who continues to spend 20 percent of his time as a practicing urologist. About three years ago, Dr. Donnell and his colleagues began the current phase of their journey around CDS for diagnostic imaging; a year ago, they went live with a solution from the New York City-based MedCPU.
Dr. Donnell spoke recently with HCI Editor-in-Chief Mark Hagland about the CDS initiative. Below are excerpts from that interview.
Tell me about your organization’s journey around clinical decision support for diagnostic imaging.
It has been one of the most interesting projects and studies, all at the same time. It really has been a lesson to me, and always a positive reinforcement as a lesson, on the value of physician and clinical leader engagement. When you have an idea, getting buy-in and sponsorship, actually working something through the system, is extremely important.
We began with the recognized need for clinical decision support, but also recognized the gap—current clinical decision supports failed to incorporate information buried in the progress note—and that’s where 80 percent of the information is. And it’s buried in freetext, much of it. Beginning with the paper by Bates [the Nov-Dec 2003 Journal of the American Medical Informatics Association (JAMIA) article, “Ten Commandments for Effective Clinical Decision Support: Making the Practice of Evidence-based Medicine Reality,” by David W. Bates, M.D. et al], we presented to them a list of what it would take to invoke or employ a clinical decision support system that would help them but not get in the way. We do have smart sets and order sets and basic alerts. And they’re very good. But as we know, the importance there is that I complete all the discrete elements in order to trigger those alerts, or I work in a very specific order in order to have those systems function at their best.
Robert Donnell, M.D.
But I was trying to get our organization to understand that we want to actually engage our clinical decision support at the point where our providers are actually thinking, not once a decision has already been made—helping them at the point of decision, rather than forcing them to back up or make a right turn. And if I could get them to be satisfied and be engaged, we felt that we could improve care. I’m a firm believer in CDS: my original background was as an engineer, and it was a cultural shock to enter medicine and find that we didn’t have decision analysis readily available.
So when did you begin this process?
We started the selection process about three years ago. We selected a vendor about two years ago—MedCPU—and we went live just a year and a half ago. Going live first on our academic campus for radiology, and subsequently rolling it out to our community divisions, and then adding more features as we went.
What are the core features you’re using from that solution?
We are currently using the radiology components, both ambulatory and acute-care and emergency room; we have value-based purchasing; we have inpatient-only surgeries. There’s a list of 1,700 surgeries that CMS [the federal Centers for Medicare & Medicaid Services] will only reimburse for if the surgery is performed while the patient’s an inpatient. That’s largely an administrative issue. We also have CDS for back pain, for cancer care, particularly in breast and gynecologic oncology patients who may present to the ER and other non-oncology services; we have sepsis. And we’re getting into our high-volume care for diabetes, ensuring colonoscopies are done, breast cancer screening, and cholesterol testing, and those ambulatory measures that are publicly reported.
This set of clinical decision support tools is available to practically all the physicians, then?
Correct. One of the tenets I put forward to the organization was that if we decided that there was a need for clinical decision support, it needed to be translatable to every area of our practice, to unify care.
So it’s available to all affiliated physicians and all physicians working inpatient and ED, correct?
Yes, and we will fire alerts for nurses and medical assistants.
How many physicians does this encompass, then?
Just over 2,000 physicians.
So they’re using this in the inpatient, ED, and in the outpatient setting?
And you went live a year ago?
We went live in stages, but were live across the entire platform just about a year ago.
Do you have any metrics to share yet?
Yes, we do. We are one of eight American College of Radiology centers of excellence. We do a phenomenal job on ordering images. But we knew there was a large gap between how we order and compliance with ACR guidelines. We are now at 92 percent compliance with the appropriateness criteria guidelines modified to some of our specific needs; that is a rise from 58 percent, in just four months.
So, since this initiative began, ordering physicians at Froedtert are being more cautious or careful in ordering imaging studies?
Yes, they are being more compliant in their ordering patterns, but that also means avoiding duplicates. And that’s very unique. When we look at the literature for radiology, which is a good example for CDS, one of the challenges is that they’re all retrospective studies, and it doesn’t provide the opportunity to go back to the provider and ask, was there additional information that might have made this more appropriate? I could put up an alert that would say that this particular clinical condition should be present. But we’re finding that a significant percentage of providers do go back and do things, but that might not have been documented. So for example, if I order a CT scan for a headache that’s been present for a week, the guidelines might say that an MR would be better; or that no study would be appropriate. But because I can work with the provider while they’re still in their progress note or in order entry, I can get to them at the cognitive stage and also at the action stage.
Are there any other metrics to discuss?
Yes—so, for example, in back pain, in ordering therapies and plan of care for back pain, we are firing roughly 60 alerts per month across our enterprise. And out of that, 56 alerts out of 60 actually have changed the course of care that the provider recommended. We have early metrics for sepsis, where we are alerting much earlier. We are able to drive the ordering of serum lactates on a very consistent basis; doing so has helped us to ensure that blood cultures are drawn prior to antibiotics, and that antibiotics are given more rapidly. We had always done a good job administering antibiotics, but we’re getting to patients faster. We have alerts, and we are driving compliance with VTE [venous thromboembolism] prophylaxis. With regard to our compliance with VTE prophylaxis in post-acute surgery patients, where we’re focusing tremendously, we have data showing that we’ve increased that level to nearly 98-percent compliance with guidelines in that area; it was probably in the 60-percent range when we started.
Have there been any major challenges of any kind, in this initiative, so far?
We continue to work through the sepsis model; it’s under design right now. It’s very complex, and we’re fine-tuning it. That’s why [as we move forward with various components] we’re live in the silent mode and fire our alerts to the developers, and make sure the alerts match our processes. We can turn on parts of this as we are assured that physicians are following our processes. And that’s helped us avoid major problems, because we’re very much of the mindset that if we present an alert, we have a rule that if providers aren’t following a particular suggestion 93 percent of the time, we go back and look at the rule and investigate to make sure that alerts are followed 93 percent of the time, and that provides a high degree of adherence on the part of providers.
What have been the biggest lessons learned so far in this initiative?
The biggest lessons learned would be the following: that we have to work within the workspace and within the mindspace of the people providing the healthcare; and that the technology has to meet the needs of people, not the people meeting the technology.
What would your advice be to fellow CMIOs, regarding moving forward with initiatives around CDS, like this one?
There are some extremely smart individuals in our field who have put out seminal papers—the five rights of CDS, and the Ten Commandments from Bates [“Ten Commandments for Effective Clinical Decision Support: Making the Practice of Evidence-based Medicine a Reality,” Nov-Dec 2003, Journal of the American Medical Informatics Association (JAMIA), David W. Bates, M.D., et al]. Those pioneers really put out useful information and insights. And we as CMIOs really need to hold the developers accountable for that, because we’re speaking on behalf of those providing the care. And if we can hold them to that, we will all benefit. With provider engagement and the right technology, CMIOs will make their organizations very successful.