February 9, 2009
Know Thyself (or At Least Thy Data)
By Beth W. Orenstein
Vol. 10 No. 3 P. 14
Efficient data mining in your RIS and other systems can better inform clinical and business decisions.
Gerard Durney, MBA, FAHRA, has a productivity rule for the Lenox Hill Hospital radiology department: Schedule inpatient CT scans for 7 am daily. The vice president for clinical services for the 652-bed, acute care hospital on Manhattan’s Upper East Side tracks how the department performs using an analysis his associate, Dana Ostrow, has written using Nuance Communication’s RadCube, a Web-based business intelligence tool for advanced clinical data mining, analysis, and classification. Durney can check whether his rule has been followed when he logs on to his computer every morning. He clicks on the icon on his desktop and can see technologist productivity for the morning, as well as patient wait times and report reading turnaround times from the previous day.
“They all know that I look at it,” Durney says. “And if, for example, I don’t see that an inpatient [CT] scan was done at 7 am, they’re going to get an e-mail from me asking why not.”
The staff, knowing that Durney has access to this information and that he checks it regularly, will head him off if, for some reason, no exam was scheduled then. “I will get an e-mail at home or on a Saturday morning explaining what happened even before I see the report,” he says.
Radiology departments are accustomed to working with data—perhaps even more so than other departments within the healthcare enterprise. It could be because imaging modalities based on digital technology have been in radiology departments longer than most. Radiology departments moving to PACS and RIS helped facilitate the emphasis on the data that were acquired. But only recently have many radiology departments started taking advantage of all the data available, searching their database electronically to gather information that helps provide better patient care and make better business decisions.
“If a radiologist wanted to view previous studies with similar diagnoses, for example, he had to take time away from what he was doing to call his information technology department,” says Matt McLenon, CEO of Softek Solutions. “He had to describe what he was trying to accomplish and ask the IT person to write a query that matched what he was looking for. It could be done, but it could take days depending on what else everyone had to do at the time, and the process was such a barrier and so time consuming that the radiologist didn’t bother with it.”
What has changed is that recently a number of medical technology firms have developed intelligence tools that enable physicians, as well as radiology department administrators, to more easily mine data from the masses of information they now keep electronically.
In May 2008, Nuance’s Dictaphone Healthcare division introduced RadCube, which works in conjunction with a radiology operation’s other programs and systems, including its RIS, computerized physician order entry with decision support (RadPort), and speech reporting (PowerScribe and RadWhere) applications. RadCube can produce reports that analyze processes within a radiology department from modality utilization and physician ordering patterns to exam throughput, reporting variation, quality measures, and patient outcomes.
RadCube is powered by LEXIMER, a proprietary natural language processing engine designed in partnership with Massachusetts General Hospital in Boston specifically for the medical imaging domain. With LEXIMER, RadCube uses data-mining algorithms to extract, structure, and classify free-form, narrative dictation found in radiology reports. A structured brief is created for each report and includes report positivity, follow-up recommendations, follow-up time frames, report findings, and more.
zVision, Clario’s radiology practice management software, helps physicians search through all past patient records, including reports and notes. To keep searchable data current, MedQuist has integrated its speech recognition solution, SpeechQ for Radiology, with Clario’s zVision. By combining the two solutions, radiologists can query their information in real time, says Carole Spangler, PhD, MBA, vice president of business development for Clario. “There’s never a time lag. It’s really a nice marriage,” she says.
Also, physicians who are dictating reports can easily go back and forth between the images on their workstation and prior patient information, which can help with making a diagnosis and subsequent treatment decisions, says Chris Spring, MedQuist’s director of speech recognition solutions.
Searching Report Content
zVision was developed at Seattle Children’s Hospital because radiologist Ed Weinberger, MD, wanted to do keyword searches on his report data. The program was designed to be user friendly and intuitive, and those who are using it say that it is, Spangler says.
Weinberger says using zVision to track more than 400,000 cases has resulted in tremendous gains in workflow efficiency for the radiology department. “It can save radiologists up to 90 minutes a day by providing searchable report data at their fingertips,” Spangler says.
Softek’s product, Illuminate, has both advanced searching and alerting capabilities among its other benefits. At the heart of its engine is a proprietary natural language processing algorithm proven on millions of documents, enabling radiologists and administrators to find what they are looking for quickly. Softek introduced version 2.0 of Illuminate at RSNA 2008 in November. Illuminate is designed exclusively for use with iSite PACS.
“Its plug-and-play integration with iSite allows users split-second access to their reports and images during the first day of on-site installation,” McLenon says.
One of Illuminate 2.0’s new capabilities is to allow users to set different alerts. One type notifies the radiologist when a recommended follow-up scan has been performed. The alert also notifies the doctor if the follow-up scan was not performed in a timely manner. Users can send an exam finalized alert to another physician to view critical findings immediately or set a track patient alert to follow a specific patient over a period of time and be notified every time a study is completed for that patient. Also, a search criteria alert can be set to notify physicians when new studies are indexed on iSite that match a specific search criterion, such as a cancer diagnosis that a physician may be interested in tracking.
“Every function of Illuminate is designed to improve patient care,” McLenon says. “Plus, the audit trail and delivery confirmation of messages are a welcome addition for reducing liabilities to radiology groups.”
The new search tools are also helping radiologists with their research. Ostrow says RadCube makes it easier for her to accommodate physicians who may be looking for specific findings for a study or a presentation. “Before we had RadCube, if someone were looking for MRIs where a specific form of cancer was detected, it was a tedious process, but I could go through hundreds of exams and pull 10 that came close for them. With RadCube, I can do it in seconds and construct a short list for them to review,” she explains.
McLenon says any radiologist putting together a presentation would save hours of work by locating the cases he or she needs to reference with Illuminate. “Researchers also can compare findings of previously unlinked studies to support a hypothesis, confirm a trend, or possibly identify new curative measures,” he says.
This new generation of data-mining intelligence tools helps radiology management make better decisions on both the clinical and business sides of healthcare. Durney says at Lenox Hill, which does more than 170,000 imaging studies per year, he has used RadCube to improve report turnaround time, especially at certain times of the day.
Durney said one departmental goal is to provide a preliminary read on every emergency CT report within one hour of the images being available. The department chairman asked to see whether the department was meeting this goal. So Ostrow set up a query that would show turnaround times for all emergency CTs. The information RadCube provided “made it very easy to plot it out and see when there were delays and what the causes were,” she says. Ostrow notes that causes were not necessarily what Durney or the chairman had expected—one of the most valuable benefits of mining your own data.
“We thought they were going to be related to angiographies, which require more postprocessing,” Durney says. It turns out the delays were related more to shift changes—something they may not have determined without tracking the data. They discussed the findings, which helped both the radiologists and technologists become more sensitive to the issue and make the changes to address it, Durney says.
The data-mining tools also are proving to be cost-effective. For example, Softek’s Illuminate contains a peer review function that eliminates the need for a separate system to generate the reports that are necessary for ACR accreditation.
“This integrated review capability enables radiologists to perform and document peer-reviews as they are reading studies in the course of their daily work,” McLenon says.
Durney says he has never found radiologists eager to do peer reviews, often because they required radiologists to go offline to complete them. With RadWhere, Nuance’s speech recognition reporting application, Durney says the radiologists are always comparing a current study to one read by a different radiologist one year ago. “It’s done randomly, and the data is easy to grasp. You no longer have to pull cases and beg and plead with the radiologists to stay later or come in on a weekend to do peer review. In RadWhere, peer review is incorporated into the radiologists’ workflow and done in a very even-handed way,” he explains. With RadCube, the data are then easily analyzed and prepared for presentation.
It is not uncommon for radiologists to use different terms for the same conditions, procedures, or results in their reporting. Often, the terms don’t match the CPT codes required for reimbursement. “The intelligence tools make it easy for billing to search the reports for terms and determine the appropriate codes for billing and be sure none are overlooked,” Ostrow says. “With the ease of analyzing data and mining terms out of reports, Lenox Hill Hospital is now using the data to flag patients to go back and review charts and see if their account was coded correctly with all of the appropriate DRGs [diagnosis-related groups] from the admission.”
Driven by Data
Durney says physicians often don’t like to be told what to do and resent when their requests for imaging studies are denied by insurers. But, he says, they are trained to respond to data. “By mining data,” he says, “you’re able to look at groups and individual physicians and modalities and compare the reasons for the test orders with the results. If you find one physician who has ordered 100 CTs for a certain indication while his peers ordered 10 on average and both had only three positive findings, you could say to the physician ordering the higher number, ‘Can we talk about the appropriateness of this exam?’”
The intelligence tools are able to index historical data quickly. Softek’s Illuminate can backload and index 500,000 studies in less than one day, McLenon says. Once the initial indexing is complete, new studies are added automatically. Illuminate indexes the clinical text, as well as the associated discrete fields of each study, resulting in point-and-click access to the images once a user performs a search.
Two Years’ Data
Lenox Hill opted to index more than one year’s worth of studies with RadCube. “We have over two years of historical data, but it wouldn’t have been economical to get any more in there,” Ostrow says.
Jonathon Dreyer, radiology marketing manager at Nuance, says he has found that each site determines who has access to RadCube. “Some facilities will only grant access to those users in charge of analyzing data for the entire department, while others will empower more employees with this information to provide better decision making at all levels,” he says.
While the intelligence tools are plug and play, their vendors provide on-site installation and training, as well as 24/7 technical support.
Ostrow says she found, as with anything new, “You have to learn the language.” But with her background in radiology IT, she found RadCube very intuitive to use. “Once I got a handle on the language, it got a lot easier,” she says.
Spangler says radiologists like the fact that with zVision, they can do all their data mining on one workstation and don’t have to remember multiple logins. “Radiologists also can customize its features to fit their needs. If there is something they want to monitor, they can put it on their dashboard, and it will update every five minutes if they want,” she says.
Dreyer expects that applications for intelligence tools such as RadCube will only grow. “As imaging departments are met with growing competition, the need to not only understand the operational aspect of business but also to track patient outcomes and quality initiatives will, without a doubt, continue to proliferate,” he says.
— Beth W. Orenstein is a freelance medical writer and a regular contributor to Radiology Today. She writes from her home in Northampton, Pa.