Seeing the Big Picture Through Big Data Analysis
Published on July 11, 2013
New business intelligence tools are empowering radiologists and driving clinical documentation improvements
By: Aaron Brauser and Stephen Willis
Originally published in ITN Imaging Technology News
We’ve lived through decades of innovation in imaging technology, from developing radiology information systems/picture archive and communication systems (RIS/PACS) architectures that digitize and store millions of images, to building out workflows and Web services that expand access and drive efficiencies. Industry consolidation and other factors in the last few years have made an indelible impact on radiology by commoditizing these services and forcing radiologists into a volume-based mindset. The push/pull in the years ahead will not make this any easier — aging baby boomers will increase the imaging-prone population, while the growing influence of accountable care organizations (ACOs) and integrated health networks will look to eliminate unnecessary imaging.
Where does this leave imaging reporting and the role of radiologists going forward? ACOs are moving the industry to value-based care, and to stay relevant radiologists need to better understand the patient and provide reporting that is meaningful, measurable and impactful. New technology advances are helping make that a reality by enabling radiologists to better analyze and communicate findings, then drive actions based on these interpretations.
Uncovering Unstructured Data
If the goal is measurable improvements in efficiency, productivity, cost-effectiveness, quality and patient outcomes, today’s radiology reporting has to change. According to a 2012 Journal of American College of Radiology study, one out of every five radiology reports contain incomplete information that, through under-coding and under-billing, results in up to five percent in lost revenue. In addition, 80 percent of documentation currently submitted would not properly assign new ICD-10 codes (to be implemented across the U.S. in 2014).
Part of the issue is the structured nature of imaging reports. Today’s heavy reliance on templates and checkbox entries in reporting make it very difficult to capture the detail, rationale and other variables that create a meaningful and complete story on each patient.
In driving quality improvements and enriching patient care, radiology documentation must provide context, not just structured data. Clinical tools must offer access to prior reports and patient narratives to give radiologists a holistic view for better interpretations. Trying to extract and distill this information manually is too time-consuming and expensive. Leveraging huge datasets of patient information captured in hundreds or thousands of reports, forms and narratives in PACS, electronic medical records (EMRs) and other hospital information systems can provide access to free text and give radiologists visibility into priors, family history and other patient data to deliver quality interpretations.
New cloud-based business intelligence technologies can now allow radiologists to aggregate and analyze both individual records and Big Data record collections. These platforms take a cue from Big Data pioneers like Google and Amazon, which don’t need heavily structured data but instead search billions of pages of free, unstructured text routinely, guided only by algorithms, to answer inquiries. Today, radiology-focused data engines can ingest previously inaccessible information in narrative documentation from systems across an enterprise and make it easily, quickly and cost-effectively available.
Natural Language Understanding
“Natural language understanding offers significant advantages by better understanding the actual meaning that radiologists convey in their reports, and generating data based on that meaning.”
All data mining engines are not created equal. Natural language processing (NLP) technology has been around for some time, offering “text matching” functionality in aggregating data across reports. What it lacks is context, however — searching for a patient with diabetes can result in matches to text that says “No evidence of diabetes.” In addition, this technology is often hard coded to specific problems and cannot handle local variances (“code blue” vs. “code yellow” arrest situations).
In contrast, healthcare IT systems using more advanced natural language understanding (NLU) algorithms provide a deep understanding of context. Using advanced querying language, NLU focuses on syntax, semantics and pragmatics (context contributing to meaning) to improve how structured data, free text and system data are understood and coded. It is based on formal abstraction of clinical information, and its formal expression uses ontologies like SNOMED CT and RxNorm, incorporating both symbolic and statistical NLU technology.
Natural language understanding offers significant advantages by better understanding the actual meaning that radiologists convey in their reports, and generating data based on that meaning.
In the past, radiology reporting has been done in a vacuum, with no feedback provided on the quality or accuracy of the documentation. Real-time alerts during interpretations and report creation can bring significant benefits in productivity and back-end processes such as coding and billing. By correcting or completing missing information at the point of care, time is not wasted later in the care cycle. It can also help address compliance issues, such as when a pre-signature alert notifies the radiologist when a report matches criteria for PQRS reporting, prompting the correction of a critical deficiency in the documentation.
Today, ACOs and value-based care are changing what is required of interpretations and their impact against pre-defined measures. Were VQ scans diagnostic or non-diagnostic? What percentage of reports was BI-RAD 3 inconclusive across mammography scans? Natural language understanding systems play a huge role for health information management (HIM) departments by identifying — in real time — whether reports are meeting quality measures and key performance indicators specific to an organization.
By moving from static, text-oriented documents to living, dynamic documents, healthcare professionals can gain instant feedback on whether reports will be impactful. Figure 1 shows examples of how real-time alerts and the combination of structured/unstructured reporting can benefit radiologists, referring physicians, HIM departments and patients.
Getting the Most out of Radiology Systems
As our industry continues its focus on clinical documentation improvements, radiology providers need to be smarter and more efficient than ever. Choosing the right business intelligence and analytic tools — and leveraging critical data trapped in PACS, EMR and hospital IT systems — can empower the radiologist to deliver the right data to the right people at the right time, resulting in decisions that build healthier populations.
Figure 1:Traditional Real-Time Measures (Structured Data Only)
- RVU productivity physician, group, etc. in real time
- RVU productivity versus moving average for assigned work list
- Work outstanding (completed but not read/finalized)
- Work incoming (arrived but not completed, or in progress)
- Turn around time (TAT) averages for providers, facilities, patient types and others
- Aging of stat studies in order to ensure appropriate TATs
- Ratio of subspecialty exams being interpreted by fellowship-trained radiologists
New Real-Time Data (Structured and Unstructured Data)
- Real-time notifications to address possible PQRS measures before signing report
- Example: 55 year old with wrist fracture would need a DEXA recommendation before report is signed
- Real-time notification of whether required body parts/organs are addressed before signing report
- Example: Abdominal ultrasound needs mention of several organs and systems
- Real-time visual indication of related/pertinent historical reports, surgeries or measures to the current study being interpreted
- Example: For MR spine patient with recent spine surgeries, prior notes with highlighted verbiage are available
- Real time notification of best-practice measures or hospital-needed measures
- Example: All incidental pulmonary nodules include Fleischner criteria in recommended follow-up
Aaron Brauser is the director of solutions management responsible for M*Modal’s Catalyst products and NLU capability. Brauser has more than 20 years of software experience specializing in bringing new product technologies to market. He earned both his undergraduate degree and MBA from the University of Pittsburgh.
Stephen Willis serves as chief information officer of Canopy Partners, which offers technology and management services to physician practices and hospitals. He has more than 12 years experience in IT, with eight of those years in the healthcare IT management across multiple disciplines. He previously served as CIO and lead systems administrator for Greensboro Radiology and IT manager for Physician’s Medical Enterprises.