Information technology (IT) is the key to improving the quality and efficiency of radiology -- and to solidifying the role of radiologists in the era of value-based healthcare, according to Dr. Paul Chang, who spoke at a September 21 webinar hosted by the Society for Imaging Informatics in Medicine (SIIM).
Achieving these goals will require going beyond the use of simple analytics tools such as scorecards and dashboards, he said. New IT tools such as artificial intelligence (AI) should be used proactively to enhance all aspects of radiology workflow, augmenting and assisting radiologists as well as improving collaboration and communication with referring physicians.
"We need to leapfrog and to do what other business verticals are doing, and that is embracing IT to achieve evidence-driven workflow orchestration," Chang said. "And [we also need] to embrace these newer technologies [such as AI] and not be afraid of them or overly hype them, but use these tools appropriately to achieve data-driven, human-machine collaboration -- what we call cybernetics collaboration -- to achieve improved quality and efficiency."
Radiologists must show evidence
Dr. Paul Chang from the University of Chicago.As healthcare shifts from a focus on volume to a value-based model, the radiology department will also transition from being a revenue generator to a cost center. As a result, radiologists will no longer be valued just for interpreting images; they will be valued for managing the role of imaging in a complex, capitated/shared-risk system.
Radiologists must also provide quantitative evidence that they are irreplaceable in this aligned model, according to Chang, a professor of radiology and vice chair of radiology informatics at the University of Chicago.
"We in radiology have to achieve quality, safety, and efficiency simultaneously, and that's very hard to do," he said. "I know of no other way of doing that than leveraging IT and informatics solutions that provide meaningful innovation to allow us to maximize quality and value to our patients throughout the enterprise."
The goal, then, is to design IT systems that can improve efficiency and productivity, as well as quality. Fortunately, it's frequently possible to design systems that can attack these goals at the same time; both efficiency and quality loathe the same enemy: variability. And the greatest source of variability in the hospital is people, Chang said.
IT principles for quality
Other industries bend over backwards trying to minimize workflow's dependency on humans who must remember to perform tasks, according to Chang.
"We should be leveraging -- like other industries -- electronic-based workflow orchestration instead of relying on humans to remember to do the right thing," he said. "Evidence-driven workflow is key."
Chang said he's not a fan of scorecards, which are retrospective in nature and aren't useful for understanding in real-time what's going on in the department.
"I'm a bigger fan of operationally embedded dashboards integrated into evidence-driven workflow," he said. "But it's more important to go beyond just the 'tools' such as the dashboards and workflow engines. We need to think about a complete quality-driven workflow model."
This requires much more flexible IT architecture and methodology, Chang said.
"This is a team sport and requires close and continuous agile collaboration amongst all of the stakeholders, including users, IT, and administration," he said.
Universal protocol
The University of Chicago's radiology department employs the quality concept of the universal protocol: the correct patient, the correct exam, the correct interpretation by the radiologist, the correct communication/collaboration with the referring physician, and the correct incorporation of the results by the clinician.
In terms of the correct patient, electronic-based workflow has yielded significant improvements over film-based operation, Chang said. However, patient identification mistakes can still occur due to human error, and those can be challenging to identify and correct.
Advanced patient identification technology such as radiofrequency identification (RFID) can minimize the need, for example, for technologists to manually select the patient's name from a list at the imaging modality, he said.
Computerized physician order entry (CPOE) and clinical decision support (CDS) can help ensure that the correct exam will performed. However, Chang isn't convinced yet that CDS should be integrated with CPOE.
"I'm a big believer in decision support; I'm just not a fan of the relative immature nature of how we've integrated decision support into CPOE," he said. "As a radiologist, I kind of feel out of the loop [due to how CDS is integrated into CPOE]."
Linking CPOE with CDS can come with unintended consequences, such as sparse clinical context for the imaging order, Chang said. For example, a 2015 study from the University of Chicago found that 60% of CPOE imaging orders lacked significant clinical history.
"Clinicians give just enough [information] to get a clean order, but not enough for me to do my job," he said.
In addition, a correct order does not guarantee that the imaging exam will be performed correctly. There is a huge opportunity for advanced IT to provide "point-of-presence" clinical context from the electronic medical record (EMR) to support assignment of exam protocols.
"If [an exam] is protocolled incorrectly because of either incorrect [computerized physician order] or insufficient information about what the clinician is actually worried about, or me not understanding what needs to be done or lacking the proper integration to make sure the technologist understands the right protocol, you can imagine that I could do a study and miss the abnormality if it's not done correctly," Chang said.
The University of Chicago dealt with this problem by developing an imaging protocolling tool that can extract clinical information from the EMR to ensure that the right protocol is used for the study, according to Chang. This improved both efficiency and quality.
Correct image interpretation
Radiologists are humans, of course, and they are imperfect observers. This means that Bayes' theorem applies -- meaning that how a study is interpreted is a function not just of the test performance but also of prior knowledge of conditions that might be related to the exam.
"I must be exposed to not just the image in the vacuum, but I also need clinical and knowledge context [for the study]," Chang said.
CPOE, however, does not adequately communicate the pertinent clinical context for the imaging study being ordered.
"I can't interpret the study without this information," he said.
Radiologists either have to do without this information, or they waste time by going into the EMR to try to figure out what the referring physician wanted in the first place, he said. But IT can be leveraged to solve this problem as well.
Chang's institution developed a clinical context lookup tool that could access data such as pathology or lab results from the EMR to augment the sparse information included in the computerized physician order. They're now also applying natural language processing tools that can review prior reports and patient history and provide a rapid abstraction of what the radiologist needs to know to interpret the imaging exam.
"Instead of having me spend time inefficiently searching the EMR for that information and resulting in variability with respect to quality, we can use machines -- not to replace me, but to augment me -- to review this information and present it to me in an efficient manner," Chang said.
"Just-in-time" CDS tools could also be used to assist the radiologist in interpretation and to facilitate peer review, he said.
Correct communication, collaboration
One size does not fit all when it comes to communicating radiology results to referring physicians, Chang said. The traditional radiology report is just one of many communication models.
"The goal is to match the appropriate communication method [to a specific clinical context] and not just use a phone call or report," he said. "Electronic-based communications and messaging models can be very helpful."
Optimizing communication requires optimized, context-specific messaging mechanisms for each type of radiologist-to-clinician communication, according to Chang. Examples include a messaging tool for stat consults with emergency room physicians, a patient-follow-up tracking application for follow-up of incidental imaging findings, and a lesion-tracking dashboard.
These show how IT can help radiology go beyond the narrative report and orchestrate optimal communication and collaboration with referring physicians. This improves the efficiency of not just radiologists, but also the quality and efficiency of all stakeholders throughout the enterprise, he said.
Business intelligence and analytics
Business intelligence is a critical enabling tool for realizing measurable improvements in efficiency and quality throughout the enterprise, Chang said.
"[Business intelligence and analytics] is more than just a bunch of static reports and more than just a bunch of scorecards," he said. "It requires a comprehensive strategic perspective and governance model."
IT can help, he said. However, radiology -- and healthcare in general -- is several years behind other industries in terms of implementing business intelligence and analytics.
"We're so behind that we only care about scorecards and dashboards, [which are] usually retrospective scorecards," Chang said.
He noted that there is often confusion about the differences between dashboards and scorecards. Dashboards are a real-time performance monitoring tool with an operational, tactical scope. Scorecards, on the other hand, are performance management tools that focus on metrics for defined strategic goals, he said. Due to their retrospective nature, scorecards aren't great for facilitating performance improvement, however.
Business intelligence and analytics in radiology should go beyond scorecards and dashboards, which are designed for human consumption and depend on people to do the right thing.
"This dependency on knowledge workers can be a liability, especially for complex real-time processes," Chang said.
Predictive analytics --- combined with IT-enabled, data-driven workflow orchestration -- can help keep human error from the workflow process, according to Chang.
Big data/machine intelligence
Although they're subject to lots of hype, big data and machine intelligence do have significant potential in healthcare and radiology, Chang said.
"What we need is data- and contextually driven, human-machine cybernetic collaborative workflow orchestration," he said. "Augment the human. Build evidence-driven workflow engines using IT to make it almost impossible for the human to do the wrong thing."
Machine intelligence can then be used to augment the radiologist by gathering and presenting information or knowledge, anticipating what the radiologist wants when he or she wants it.
