Structured reporting rewards early adopters

March 6, 2003

From traditional transcription services to speech recognition technology to structured reporting, the way to streamlined reports seems straightforward. But appearances can be deceptive, and many objections may be raised against structured reporting.

From traditional transcription services to speech recognition technology to structured reporting, the way to streamlined reports seems straightforward. But appearances can be deceptive, and many objections may be raised against structured reporting. Radiologists must understand why the technology is worth the effort.

Structured reporting offers three primary advantages over traditional reporting methods: It streamlines the multistep dictate-transcribe-edit-approve process, saving both time and money. It standardizes imaging descriptions so that reports are reproducible and comparable across an institution, regardless of which radiologist provided the interpretation. And, finally, if integrated with the Internet and HIS/RIS systems, it can ensure before the radiologist moves on to the next case that every report contains reimbursable billing information, provides clinically relevant answers to the question posed on the requisition, and is immediately available on the Internet.

These improvements are critical in today's healthcare environment. Even as the number of practicing radiologists declines, the number of imaging studies continues to rise. It is becoming increasingly rare for a patient with headache; chest, abdominal, or pelvic pain; or trauma to escape the emergency department without the blessing of the radiologist and a multislice reconstruction. To keep up with the increasing number of cases per radiologist per day that has resulted, doctors must use improved methods of reviewing cases and providing final interpretations.

One strategy to maximize reimbursement while payers continue to tighten the belt is to decrease average length of stay without compromising patient care. While only so many corners can be cut before the nose goes to spite the face, radiology reporting has suffered from delays that are entirely avoidable given today's technology. This is why a structured reporting system that decreases report turnaround time will inevitably be adopted. Radiologists who refuse to do so risk losing even more turf battles to clinical specialists hungry for a piece of the imaging pie.

Structured reporting had been available since the early 1980s at Harvard Beth Israel Deaconess Medical Center, when it was Beth Israel Hospital, in the form of a Unix-based interface. At that time, radiologists had the option of traditional Dictaphone-type dictation or a new keyboard-driven structured reporting system. The report was created by selecting words in nested menus that corresponded to individual keyboard letters. A physician reporting a chest x-ray, for example, might press L to describe the lungs, then C for clear, to form the sentence "The lungs are clear." It was an early form of computer-assisted stenography.

The major drawback to this system was that the keyboard menus had a fairly steep learning curve because the letter choices were not always intuitive. If you wanted to describe a mass lesion in the brain, for instance, would pressing M give you a menu of masses? Or would it instead allow you to describe the appearance of the medulla, mastoid air cells, or the middle meningeal artery? Those doctors who took the time to learn the system found that it could speed up their reporting time for normal or near-normal reports. The problem came with complicated cases that were better suited to dictation?and once the radiologist had the microphone in hand, it seemed easier to finish the remaining reports that way as well, complicated or not.

Hindrances to this early structured reporting system multiplied when new modalities arrived, necessitating the addition of new nomenclature to the structured reporting glossary. More abstruse keyboard shortcuts (R for hypeRintense and O for hypOintense) resulted, with a new learning curve for each radiologist. Eventually, the system atrophied from disuse.

"People couldn't figure out the code, or they complained that it precluded the nuances of their dictations," said Dr. Jonathan Kleefield, director of neuroradiology at the Beth Israel Deaconess Medical Center. "Anything slower than dictation did not have a very good shelf-life."


You can teach most minds to accept a structured reporting system for relatively simple descriptions, but when a study contains multiple abnormalities, there are nearly an infinite number of reasonable ways to describe it. The independent mind, a hallmark of many physicians, balks at containing itself to someone else's idea of a perfect report. Trying to describe an intricate imaging finding while performing mental acrobatics to fit descriptive words into a fairly rigid format may lead to little or no savings in efficiency.

The challenge comes in designing a structured reporting system that harnesses the power of computer processors to make the reporting interface as intuitive and topographically simple as it is internally complex. A mouse-driven point-and-click interface such as eDictation offers a method of report generation similar to the keyboard interface. This system is more intuitive than keyboard input, since anatomy and imaging findings are visually linked on the screen in a tab-folder and radio-button interface.

Implementation of these mouse-driven systems can be just as cumbersome as the keyboard input method, however. Click the lung tab, click the volume tab, click the hyperinflated tab. Click the mass lesion tab, click the attenuation tab, click the soft-tissue tab, click the size tab, type in the dimensions, click the margins tab, click the spiculated tab, click the lobe tab, click the left upper tab, click the extension tab, click the pleural surface tab. Or pick up the Dictaphone and say, "The lungs are hyperinflated. There is a 4 x 5-cm spiculated soft-tissue mass in the left upper lobe which extends to the pleural surface." Done.

As any linguist will tell you, speech is remarkably efficient. It is the transcription and finalization process that retards the process. On paper, the trend toward speech recognition dictation software seems intuitive. Massachusetts General Hospital uses a self-learning voice recognition program that requires an initial 15-minute training session with each user, with periodic fine-tuning whenever its recognition rate falls short. Unfortunately, this technology does not always provide smooth sailing. People initially refused to use the new technology, until the traditional dictation stations were physically removed from work areas.

First-year residents naturally have the most difficulty with the system, while they adjust to its idiosyncrasies. Editing reports on the fly requires at least rudimentary typing skills (two-finger typists beware), and sometimes the radiologist's accent or phonation proves problematic for the near-99% recognition benchmark claimed. By the time residents have reached their third or final year, they have shaped the system to their liking and voice no major complaints about its implementation.

Although speech recognition software may shorten the time to approval, the resulting report may appear no different. With no template to follow, the radiologist is just as free to omit pertinent negatives or neglect to address the posed clinical question as he or she was with the traditional dictation method. Is the radiologist a laconic Yankee with a pile of cases left to read at 4:45 p.m. on Friday and a sadistic commute out of the city, or a loquacious Southerner with an icy carafe of sweet tea at hand and all the time in the world? The resulting report may be quite different: "Chest: Normal" versus "Two chest radiographs are submitted to me for interpretation, one in the posteroanterior projection and the other in a somewhat obliqued lateral projection, which I suspect is due to the patient's advanced age and on account of the kyphotic thoracic spine, rather than on any inherent ineptitude of the technologist. In any case, the lungs are expanded to the eighth and ninth rib interspace, which affords a clear view of the parenchyma? [report truncated due to space limitations]."

While there is nothing inherently inappropriate about either reporting style, each has significant drawbacks for the ordering clinician. Did the first radiologist compare the "normal" chest x-ray to the exam done yesterday, the one done two years ago, or neither? Could the Gone with the Wind dictation have gotten to the point more quickly and still afforded the same wealth of information?

Structured reporting can address and solve these issues. At its best, structured reporting allows the radiologist to naturally integrate the interpretive process with the reporting process. As the radiologist sees an abnormality on the screen, he or she can direct the program to save that image to the report and then proceed to describe it with preselected options limited by an artificial intelligence interface to reduce onscreen clutter.

When a radiologist logs on to the system, a customized lexicon of radiology descriptors is loaded as well, so that a chest radiologist who loathes the term "consolidation" never has to see it as an option in the drop-down menu. Working in the background, the computer analyzes the frontal chest radiograph, dividing the two dark lungs from the white mediastinum. When the radiologist touches an opacity at the apex of the left lung, the system is already primed with the phrase, "Projecting over the apex of the left upper lobe on the frontal view, there is . . ." More robust artificial intelligence road maps could be used in an oncology CT to assist with the measurement of a metastatic liver lesion, determine which anatomical segment is involved, and mine the report of the previous exam to insert the prior dimensions, which are required to track interval change. The best and only market-ready example of a system that uses several aspects of this approach, as far as we know, is the REX program from PointDx. Judging by the amount of interest generated at the company's 2002 RSNA debut in December, it seems that many radiology professionals were similarly intrigued by its novel approach. Although the product is new and not yet widely adopted, the interface looks very promising.


At the Beth Israel Deaconess Medical Center, the GE PACS provides a sufficiently stable platform that more often facilitates than hinders rapid interpretation. Despite early adoption of this expensive technology, radiologists at the BIDMC still use the dictation-transcription form of reporting. This process has inherent delays, sometimes mounting to three full days before final interpretations are available. This makes one wonder whether having a PACS by itself truly improves patient care if clinically pertinent final interpretations are sometimes available no sooner than they were in the age of hard copy.

In an effort to hasten the physician-dependent portions of this process, the department provided productivity bonuses to those radiologists who could shave hours or days off the approval process and instituted an internal watchdog system that monitors the efficiency of individual sections at monthly meetings. If radiologists must be paid extra to speed up an inherently inefficient system, then clearly the time has come to seek a more efficient method of reporting.

The film versus PACS debate of just three years ago offers an example of how new technology renders tried-and-true methods obsolete. Back in the days of hard-copy alternator readings, many radiologists took for granted the delays posed: hanging films, crawling around for films dropped inside the alternator, searching through an overstuffed film jacket for the best but often mislabeled comparison study, calling the file room to order previous volumes stored in offsite facilities, throwing their hands up in exasperation when the only good comparison study was locked inside a vacationing surgeon's office, and tediously scrolling to the beginning of the alternator when an interested clinician stops by for a curbside interpretation of a study reviewed hours ago, then scrolling back to board 37.

Does this bring back memories for PACS users? It raises the hackles on our necks even now. PACS has come a long way in improving the interpretation process, depending on the efficiency of your chosen PACS solution. In the future, when the reporting process has been streamlined with the help of computer-assisted structured reporting, the days of dictation-transcription-editing-approval will seem just as tedious and outdated.

Dr. Saurborn is a senior radiology resident at the Beth Israel Deaconess Medical Center? Harvard Medical School and founder of, an Internet-based solution for remote radiological interpretation. Dr. Patidar is a radiologist at Massachusetts General Hospital and an instructor at Harvard Medical School.