Decisions Made Easy

Radiology decision support systems, believe Dr. Keith Dreyer and Dr. Jeffrey B. Weilburg, help physicians to improve quality and reduce cost of patient care. In the future, Dreyer and Weilburg predict, the technology will become more and more commercialized and accepted throughout the 5747 hospitals in the US. But what are some of the implementation challenges? EHM found out more.

EHM. What are commonly used clinical decision support systems (CDSS)?

KD. The first real decision support systems that came into practical use were around drugs. They proved to be able to control drug interactions and dosing errors. These systems came into place at our institution about five years ago.

The other general systems that are available around clinical decision support do analysis on charts. They use natural language processing to find differences in practice patterns and alert the physician of specific circumstances in general clinical settings.

But the real value today in clinical decision support is really when it’s focused in a specific area. Our specific focus is in radiology decision support, which is an area that has very little deployment throughout the world. The system that we created in 2004 was one of the first systems that could be used for it.

We have two types of radiology clinical decision support. One is for physicians to determine what is the most appropriate exam to order, and the second system is for the radiologist who’s interpreting the exam to aid them in the interpretation process.

EHM. What impact has the system had on patient care?

KD. What this system has done since it’s been deployed in September of 2004 is that it has decreased the variability between ordering physicians. The radiology interpretation system was deployed in January 2007, and we’re using it to have advanced ways to get to patient’s clinical histories so we can display, depending upon the indication of the exam and what type of exam, specific information to the radiologist as well as give them additional information from medical knowledge, text books, online et cetera. We have identified other areas where there are inconsistencies in radiology reporting, and we’re just about to take the same method of decreasing that variability.

EHM. How do hospital administrators benefit from radiology interpretation systems?

JW. They view it as a positive in that it improves quality. It has been largely neutral to positive for physician satisfaction, and neutral largely for patient satisfaction, but it improves the efficiency of the ordering process. You could also say that it represents an attempt by the clinician community to do their part in an appropriate way in cost control.

One of the great things about an internal system rather than an external company is that we are able to gather extensive and finely-grained data on each individual’s ordering patterns, and begin to connect those ordering patterns with outcomes. We are also working on ways to determine how to age-, sex- and severity-adjust utilization so we can spot outliers and then respond to those outliers. What we tend to do is share the data with the clinicians and try and engage them in an understanding of whether or not their being an outlier has anything to do with better or worse clinical outcomes. We also pay a lot of attention in engaging clinicians to work flow issues.

KD. They were really the driving force behind this because of the relationships that we have with payors. The alternative was to decrease our reimbursement or to use radiology benefits companies, and all of that adds cost to the system. So deploying the system has reduced cost in many ways. And then, in having order entry and decision support all electronic, it allows us a way to communicate with our physicians. Some of the physicians are employees, some of those physicians are not employees. We have about 5000 physicians in our plan and now we have a way to evaluate them on a regular basis to determine who needs additional education and who needs additional information right at the time of ordering.

EHM. What were the challenges you came across when you implemented the new technology?

JW. The implementation of the radiology order entry (ROE) system was initially resisted by the physicians because it was a change in their practice behavior, and it required them to sit in front of a computer screen rather than to fill out a paper requisition form and hand it to a practice assistant. It took more physician time, but over a period of years it has gained reasonably good acceptance by the physicians because it reduces the errors when you have paper forms and multiple transcriptions. It has improved quality, and it has turned out to be about as time-intensive for the primary care doctors and specialists as paper forms. Some of the surgeons still don’t like it because it interferes with their workflow, but by and large they’re using it because they have recognized that it prevents them from having to call external utilization management companies.

KD. Deciding the method by which we will manage the system is a challenge. We have a couple of hospitals using the system. One has decided that if you go below four for score, you have to call a radiologist to discuss the exam to order it. Another hospital has decided to let anybody order anything. But then those abusers of the system that have consistently low scores will have to meet with their superiors and administrators to have discussions where they’re at and what additional education they might need. Because it’s such new territory, developing the software and creating all the rules took years and was pretty complicated, but from a deployment standpoint, the biggest challenge is to determine which is the best method to deploy and manage the system.

EHM. How user friendly is the technology?

KD. There are two things that we’re looking at doing. One is to be able to record all of the information from the electronic medical record and have that in the system prior to the physician ordering a radiology procedure. That requires everything to be quantified so that we can know exact information before the scores get calculated.

The other approach, because the sub-specialist who see complex patients often see patients with similar circumstances, is to create a favorites list for the physicians to say, ‘These are the class of indications of this type of patient, and these are the typical examinations that I want to order’. It gives them a shortcut to be able to circumvent some of those long fill-ins that are required.

JW. User-friendliness is a challenge because Mass General is a tertiary care institution, so we see patients with severe and complex medical and surgical problems locally and from around the world. It’s very difficult to have any sort of user-friendly system that allows you to handle patients of such wide complexity and variety. It’s also very difficult to have a system that’s suitable for both primary care doctors who see more primary care, and specialists who see more tertiary coronary care. So there are a lot of challenges in implementing something that’s acceptable to physicians that doesn’t take too much of their time and doesn’t interfere with their workflow and clinical process. We’re gaining experience with and working on that as we go.

KD. The thing that is unique about our system, and it’s really added a lot of advantages, is we’ve used natural language processing on the radiology reports so we can look at millions and millions of radiology reports to study the rules, to see when we do deploy the rules in this fashion, how did it change the outcome of detection of certain diseases, and we’ve been able to modify the rules in that way. And further, we can use the natural language processing to be able to evaluate the outcomes of patients as well as the outcomes that are received by ordering physicians. So we can compare any individual ordering physician to a group of their peers or to another individual. This is the first time in radiology domain that anyone’s been able to do this and it’s really changed our ability to understand the rules that we’re given as well as the individual physicians.

EHM. You’re continuously modifying the system. How are you hoping to improve it?

KD. We’ll continue to expand the information, but in the next one to two years, in addition to the ability for a physician to order an exam online, they will also have a scorecard for themselves of all the previous exams that they’ve ordered of this type for these indications, what the outcomes were, how successful they were compared to their peers, if they order the alternative exam what’s the probability of them finding disease versus using this exam, et cetera. Physicians, from what we’ve seen, really want to have the facts in front of them. We have millions and millions of historical data points – now we can collect the information and give them a graphical representation of that history.

EHM. Are other healthcare institutions utilizing radiology decision support systems?

KD. We’ve seen a rapid growth in the US in the last 18 months. Our system that we license out is now in about 15 hospitals. Other major organizations have licensed it and I think it’s going to be like PACS or some of the other systems that were out there in the leading institutions first. From there, the product will become more and more commercialized, be more accepted by the health plans and it’ll start to deploy throughout the 5000 hospitals in the US.

About Keith Dreyer

Dr. Keith J. Dreyer is a radiologist with a PhD in computer science. He has been a professor at Harvard for 12 years. His current role is Vice Chairman of the Department of Radiology at Massachusetts General Hospital and Assistant Professor of Radiology at Harvard Medical School. His current interests are in decision support, natural language processing, data mining and computer aided diagnostics.

About Jeffrey B. Weilburg

Dr. Jeffrey B. Weilburg is the Associate Medical Director of the Massachusetts General Physician’s Organization, a group specifically representing the physicians at Massachusetts General Hospital. His clinical background is that of a psychiatrist. He became interested in radiology utilization management because of his experience with carve-outs in psychiatry.