During the worst of Hurricane Harvey, the University of Texas MD Anderson Cancer Center was “an island surrounded by flood waters,” according to hospital executives. A team of staff rode out the storm at the Houston hospital with more than 500 inpatients; radiologists who could not get to the cancer center because of wind, rain and flooding assisted electronically, using the integrated electronic health record and picture archiving and communication system.
“Although there were two radiologists on site, most of the interpretation was done remotely by radiologists who still had electricity and Internet connections at their homes,” says Kevin McEnery, MD, director of innovation in imaging informatics for the facility. Images could be examined from the PACS via a secure connection between the system and radiologists’ homes.
That offsite assistance from radiologists would have been impossible only a few years ago, when CT, MRI and other radiological images were stored on film or CDs and transported with patients between providers. While radiologists are still dealing with CDs more often than they’d like, a fully electronic, integrated approach to diagnostic imaging is quickly becoming the norm.
This is just one example of the evolution of imaging, which is occurring rapidly as the radiology profession faces profound new challenges from changing economics, reimbursement schemes and organizational strategies. In sum, radiologists are looking to new information technology approaches to cope with the rising pressure.
Improving the integration of radiology systems and other clinical information systems is one of the ways in which radiology can improve cost savings and improve efficiency, which will become increasingly important as healthcare shifts from a fee-for-service approach to one centered on value-based care.
Improving value means providing “the best level of care to the patient at a reasonable cost and in a reasonable timeframe to obtain the best outcome,” says McEnery. Achieving that vision in radiology means addressing a number of challenges, including confirming that orders are clinically appropriate, accurately interpreting scans, and developing easy-to-read reports for physicians and patients.
Tech-savvy radiologists have started using data analytics, machine learning and artificial intelligence to help them achieve these and other value-based care goals. However, additional building blocks are needed—including structured reports, standardized data lexicons and clinical decision support systems—to fully enable these tools.
Gregory Nicola, MD, vice president of Hackensack (N.J.) Radiology Group, believes that radiology will serve as a trailblazer for other medical professions. “We’ve been a technology-driven profession for a long time. Some things have hit us, including machine learning, before most other medical professions. There are going to be a lot of lessons learned from radiology.”
Using artificial intelligence
Radiology is emerging as a key proving ground for testing the use of artificial intelligence in healthcare. Industry watchers believe that this advanced technology may improve image interpretation and assist radiologists with mundane tasks and readings, perhaps enabling them to focus efforts on images that need closer examination by professionals.
For instance, University of Virginia (UVA) Health System is collaborating with Zebra Medical Vision to test the effectiveness of machine learning algorithms in identifying the presence of five medical conditions or abnormalities on scans: emphysema, coronary artery calcification, fatty liver, spinal fractures and low bone density in the spine.
“These apps run in the background and provide us with alerts if the algorithm identifies an abnormality,” says Arun Krishnaraj, MD, associate professor of radiology and medical imaging at UVA. “As a subspecialty abdominal radiologist, I can’t say that the algorithms have made any findings that I overlooked. But since I work in an academic hospital with a lot of trainees, I think it is useful as another set of eyes to double check that we acknowledge abnormalities that are present.”
Of the five algorithms, the one used for bone density offers information that radiologists do not routinely look for in abdominal scans. This algorithm computationally determines a bone density score. “There are subtle things that the algorithm identifies that are not visually easy to see,” Krishnaraj says. “For instance, it calculates bone mineral density based on a standard deviation.”
While the use of artificial intelligence is only beginning in radiology, Krishnaraj believes it will eventually usher in a major shift in the value proposition that radiologists offer patients and other clinicians.
“Up until now, the best radiologists were those with the keenest eyes, or the ability to consistently identify abnormalities on imaging studies without mistakes. But I think that will be supplanted by machine learning algorithms that will become more consistent and more accurate than our eyes.”
The new value proposition for radiologists will be as diagnosticians who can synthesize all available information about a patient’s case into an easy-to-digest report with recommendations for next steps. “Radiologists are sitting in the middle of an information storm,” Krishnaraj says. “We may be the only clinicians who can see inside the body, noninvasively, while also being able to look in the EHR to review the patient’s clinical history, laboratory values, tumor markers, etc. Our job as radiologists is to synthesize all this information into something actionable.”
New reports and nomenclature
A key step to using artificial intelligence and data analytics to improve value in radiology will be the development of structured reports and lexicons, according to Nicola.
“Machine learning algorithms have a much easier time if the exact same words are used by everyone,” he says. “Some people think that natural language processing can be used to address this problem. But being very consistent in the language you use is a more effective strategy. We’re laying the groundwork first by standardizing and structuring so that, when machine learning is universally applicable, we’re already there to use it.”
Standardizing nomenclature also can result in near-term benefits that can bring efficiency to how radiologists work and support overall care delivery.
The radiologist’s primary end product is the interpretative report for each imaging study. “People may not even see us in many cases,” says Cree Gaskin, MD, professor and vice chair, operations and informatics at UVA. “We read a study and produce a report. That’s why many improvements in radiology need to revolve around ensuring that our reports are timely, accurate and relevant, and that they convey information efficiently for whoever is consuming that report downstream.”
To this end, Hackensack Radiology Group is working with ordering physicians to standardize the structure of their reports. “We talked to the referring physicians and asked, ‘What do you need to know when you treat this patient?’” says Mohit Naik, MD, director of quality.
The radiology reports are divided by specialty or body part. For instance, an abdominal CT report has separate sections for liver, pancreas and other organs. This enables a pancreatic cancer surgeon to go straight to the pancreas section rather than having to search through the entire report for relevant information.
Hackensack Radiology Group is also working to incorporate standardized nomenclatures to reduce variation in how radiologists share key findings. To explain the concept, Naik points to the Breast Imaging Reporting and Data System (BI-RAD) tool. Developed by the American College of Radiology, BI-RAD divides mammography findings into six standardized categories along with recommended follow-up steps. These range from “negative” (or no significant abnormality found on the scan) to “highly suggestive of a malignancy” (biopsy strongly recommended).
“These standardized nomenclatures help clinicians understand the radiologist’s level of suspicion about a malignancy or pathology,” Naik says. “This, in turn, helps guide the appropriate management for that patient.”
Hackensack Radiology Group has implemented other reporting tools that have been developed by medical societies, such as PI-RADs for prostate cancer screening. However, there are not yet nationally accepted reporting and data systems for many conditions. In these situations, Hackensack Radiology Group works with referring physicians to standardize their own nomenclatures.
For instance, its radiologists worked with gastroenterologists and surgeons to develop a lexicon for pancreatic cancer staging so that physicians can easily find out whether a tumor can be surgically resected, or removed, depending on the location of the tumor. “Our surgeons have been very happy with the structured template,” Naik says. “Most radiologists already knew what needed to be reported, but they weren’t necessarily reporting in a coherent, structured way.”
One challenge in redesigning radiology reports is to avoid increasing documentation time. Hackensack Radiology Group worked with IT staff to automatically populate portions of the reports by linking reports to associated CPT codes.
When radiologists click on a report template for a certain imaging scenario, the template knows that the scenario is associated with a specific CPT code and automatically fills in information that must be documented for that test, such as how the study was performed. Previously, radiologists would have to dictate all that information for each image. Now, for the most part, they only have to dictate their findings.
“Physicians often feel that these initiatives position them as data entry personnel,” Nicola says. “But you can actually save physicians an enormous amount of time if you approach these initiatives correctly.”
Improving access and integration
Before radiologists can accurately interpret a scan and issue a report that effectively guides clinical decision making, they need access to all relevant images and patient information. Toward this end, better integration is key.
For example, to accurately interpret a scan and issue a report, radiologists first need access to the imaging in question, as well as to other relevant scans and patient information. A group of Level I and II trauma centers in North Carolina, including Duke University Medical Center, agreed to purchase the same platform for image sharing to ensure rapid exchange of scans. While North Carolina has a statewide health information exchange, it does not currently allow sharing of images.
Almost all hospitals in the state had signed onto the commercial platform as well, and the number of images being stored on CDs has gone down dramatically. “It spread virally throughout the state,” says Christopher Roth, MD, director of imaging informatics strategy and vice chairman of radiology, Duke Health. “Care is faster and more reliable because of this electronic sharing.”
One goal was to ensure that trauma patients across the state could get urgent images rapidly reviewed by expert radiologists at Duke Health and other tertiary care hospitals. Some research indicates that these secondary reviews often turn up key findings that are missed on the initial review by local physicians.
The cloud-based platform enables participating organizations to connect their scanners or PACs to a central repository to post and download images. “Because of the way we architected the workflow, it’s very straightforward for physicians to put in an order for an outside interpretation,” says Roth. “We have set up the dataflow and people workflow to support very smooth and effective patient care.”
In addition to being able to quickly obtain needed images, radiologists often require easy access to key patient information in the EHR to help them interpret scans as precisely as possible. Often, radiologists only know the reason for the exam, such as “rule out fracture.” But additional information is often needed, such as the associated diagnosis, surgical history, laboratory results, prior radiology reports and images, and pathology results.
To address that issue, UVH Health System developed a radiologist dashboard in its EHR that displays key patient information that radiologists typically like to refer to when reading a scan. In addition, the organization’s EHR and PACS were integrated. Via the dashboard, radiologists can easily view relevant patient information and images they need at the same time they are reviewing new scans.
“Radiologists sometimes complain that the EHR wasn’t made for them, but that’s not really true,” says Gaskin. “It’s just a matter of telling the EHR what information you want on a dashboard display.”
Other forces are pushing healthcare organizations, and radiology departments in particular, to reduce the number of unnecessary imaging studies they do. That will become particularly important in value-based care approaches, which leave providers at risk for patient care costs.
Industry studies suggest that 20 percent to 50 percent of imaging tests are unnecessary, exposing patients to needless radiation and contributing to high costs. Recognizing this, Hackensack Radiology Group is participating in the American College of Radiology’s R-SCAN Initiative, which aims to improve imaging appropriateness through the use of criteria.
The radiology practice is collaborating with emergency department physicians at an affiliated hospital to reduce ordering of imaging tests for certain patient conditions, such as CT scans for minor head trauma. Various specialty societies have identified 12 questionable imaging scenarios for the Choosing Wisely campaign, which is encouraging dialogue around the necessity of tests and treatments.
A key step in R-SCAN is identifying patient cases where imaging was inappropriately ordered so that specific education or feedback can be provided to physicians. “This turned out to be labor intensive because we have to do manual chart extraction,” Nicola says. “Sorting by ICD-10 code is not good enough. We had to have a clinician read the clinical note to find out what the indication for the study was.”
To automate the process, Hackensack Radiology Group is asking emergency department physicians to use standard terms in their reports, such as “minor head trauma, no risk factors,” Nicola says. “By using a lexicon, we will be able to have the computer mine those reports and determine if the imaging ordered was appropriate.”
In addition, radiologists are working with hospital IT staff to build a homegrown CDS tool in the EHR to query physicians when they order imaging deemed inappropriate by Choosing Wisely, Nicola explains. “For example, before they can place an order for a head CT, the EHR lists questions: ‘Did the patient have minor head trauma?’ ‘Is he or she on anticoagulation therapy?’ Then the CDS-EHR tool would warn the ordering provider that, ‘The Choosing Wisely campaign says this patient does not need head CT. Do you still want a head CT?’”
Using CDS to guide orders is becoming a priority for radiologists looking to comply with a Medicare requirement that ties payment for advanced imaging (such as CT, MRI and PET scans) to the use of Medicare-recognized CDS tools that incorporate appropriate use criteria. The requirement, which this summer was delayed until January 2019, only applies to outpatient imaging studies.
The UVA Health System is all set to comply with this Medicare rule. The Charlottesville-based system uses a CDS system that incorporates appropriate use criteria from the American College of Radiology and other medical societies.
The UVA Health System is working with its CDS vendor to proactively address preauthorization requests from commercial insurers, beginning with a pilot project with Aetna. When providers order a head CT, for example, they are prompted to choose the reason for the study, such as “severe headache.” Then, the CDS system issues an appropriateness score from 1 to 9, with the numeric scores ranging in meaning from “low utility” to “indicated.” The system also lists alternative imaging for the same clinical scenario with appropriateness scores, relative cost and relative radiation dose.
“The providers know in an instant if their orders are appropriate or not and can see scores for other exams that could be ordered instead,” says Gaskin.
For the Aetna pilot, imaging orders for Aetna members that receive good appropriate scores can be performed right away, eliminating the delay and often-tedious manual steps involved in seeking a preauthorization.
Better use of information technology also can help radiology departments meet rising demand for improved efficiency throughout an organization.
For example, when patients arrive at MD Anderson for initial consultations or post-treatment follow-up visits, they often require new imaging studies performed before they meet with their physicians. To improve value for patients, MD Anderson has put electronic processes in place to help coordinate the imaging schedule—both the scan appointment and the radiologist’s interpretation—to coincide with the patient’s doctor appointments.
“We want to minimize the amount of downtime for the patient, meaning that patients get the appropriate imaging performed in the appropriate sequence, see their physician, and get their care plan arranged in a short period of time rather than having to stay in Houston for several days,” McEnery says.
Prior to the visit, computer-based algorithms screen each patient’s schedule to identify imaging that isn’t optimally scheduled or inappropriate. The algorithms also check for redundant imaging, such as scans being repeated too soon or orders for similar imaging tests by different physicians. When issues are identified, the algorithm alerts scheduling staff, who work to resolve the issues. “It allows our scheduling coordinators to focus on those patients where the schedule needs to be optimized,” says McEnery.
Another key goal is to ensure the radiologist’s report is completed before scheduled physician appointments so the physician and patient can discuss next steps. Radiologists access a work list via the EHR that automatically prioritizes which scans should be read first, based on the time of the patient’s appointment.
Radiologists at UVA Health System have improved turnaround times—between 23 and 70 percent for particularly time-sensitive scans—thanks to a home-grown priority reading system.
Register or login for access to this item and much more
All Health Data Management content is archived after seven days.
Community members receive:
- All recent and archived articles
- Conference offers and updates
- A full menu of enewsletter options
- Web seminars, white papers, ebooks
Already have an account? Log In
Don't have an account? Register for Free Unlimited Access