Scripps Health is a nonprofit, community-based healthcare delivery network in San Diego with five hospitals and multiple imaging centers. La Jolla Radiology (LJR) contracts to provide services to three of the five hospitals. LJR has 24 radiologists on staff who produce 450,000 reports per year on average. To handle this tremendous volume, we embraced speech recognition in our radiology practice early on, implementing it at one of Scripps’ facilities in 2002.

Prior to adopting speech recognition, we relied heavily on digital dictation and a staff of in-house medical transcriptionists/editors (MT) to complete our reports. If one of the MTs took sick or went on vacation, making up the difference and completing reports could become backlogged for days. In one extreme instance, we lost two transcriptionists and then had one go on vacation, and in one week’s time we fell so far behind that it took us more than a month to catch up. This was a major impetus for us to look at a speech recognition solution.

An Imperfect Solution

Once we started using speech recognition, its value quickly became clear, even to our initially skeptical financial officers. We were able to show an ROI of $200,000 to $250,000 for the one hospital over a 3-year period by reducing the use of medical transcription by 50 percent. Typical report turnaround time (TAT) from exam order to report signed shortened to 24 hours, down from an average of 36 to 72 hours prior to employing speech recognition; however, we had low adoption of self-editing by the radiologists, due to the difficulty of switching between the system’s editing and transcription options.

Also, the limited space the speech recognition system devoted to fields or comments from the hospital information system (HIS) and the radiology information system (RIS) hindered performance. The radiologists had to interact with the different fields via mouse clicks to access patient information, including history and diagnoses. Moreover, when we switched to PACS in 2005, the speech recognition system was not integrated with it; therefore, exam orders continued to require paper requisitions with dictation initiated through the use of a bar-code reader.

In this workflow, radiologists are given paper versions of patients’ physician orders for radiology procedures that include basic patient demographics and order information. Each radiologist then must enter the patient’s image accession number (or numbers for multiple images) by either scanning a bar code, manually typing it, or dictating the numbers for the transcriptionist to enter. This can slow down radiologists. Occasionally, bar codes are incorrect, which can result in radiologists dictating reports for the wrong patients or exams. Radiologists would not necessarily know that they had dictated an erroneous report or case unless they recognized it when the report was returned for their review and signature, at which point identifying the error can be very difficult. This resulted in approximately 5 percent of our studies getting lost, requiring re-dictation.

A New Solution

Knowing that technology continues to advance and hoping to expand the use of speech recognition at Scripps, we went to the 2004 Radiological Society of North America conference to explore various solutions for a potential upgrade. We became interested in SpeechQ for Radiology from MedQuist because of three advantages that we observed at a demonstration.

The first thing we noted was a markedly improved speech engine. Secondly, it operated in a Microsoft Word environment, enabling us to create a “paperless requisition” that would include on a single screen all the pertinent patient demographic information, and HIS and RIS info (including history and reason for examination). In addition, in the proper HIS environment, we could associate patient images with the reports — a valuable improvement. Finally, it eliminated the barrier to self-editing, enabling radiologists to selectively self-correct certain reports, while sending others to a medical transcriptionist/editor.

The Chairs of the three LJR-contracted Scripps hospitals ultimately chose this solution partly due to the vendor’s willingness to provide an integrated, customized solution, and also because of the system’s extensive microphone functionality options that reduced mouse clicks and voice commands. (For example, the radiologists could select editing or transcription with just one click of a button on the hand-held microphone.) The new system also would fully integrate with our Stentor PACS, and because the vendor offered transcription services, the transition would be smoother at our largest facility, which employed outsourced transcription services.

Radiologist Training

In January 2006, we went live with the new speech recognition system for radiology. Our on-site transcription leads attended a 1-week training course and our radiologists completed 20 minutes of voice training prior to implementation. After only about 45 minutes of post-implementation training, the radiologists could dictate reports and create their own templates and auto-texts. Two hours more training enabled the radiologists to operate independently, and because we had our lead transcriptionists now trained as trainers, they circulated around the department daily and assisted the radiologists in their use of the system.

We went from a non-integrated environment to one where speech recognition is fully integrated with our RIS/PACS. The radiologists can now view each exam with all of the information included, and sign off on complete reports. Full integration also means less chance of lost reports; the RIS/PACS sends the patient data to the speech recognition system as the radiologists open each exam, enabling them to visually verify that the patient’s name and order information are identical in both.

The accuracy and speed of the speech recognition system, as well as the flexibility to either self-edit or send to transcription, became major factors in achieving an initial adoption rate whereby 70 to 80 percent of all reports were self-edited. In addition, knowing that the reports were going to be released immediately to the electronic environment strongly incentivized the radiologists to self-edit. Even where the transcriptionist/editor received the report, we asked each radiologist to go ahead and self-correct their preliminary impression.

Results

The immediate availability of image-integrated reports that enabled the radiologists to go back and review as they were dictating, resulted in more detailed impressions and improved referral satisfaction.

Scripps’ referring physicians also were pleased with the rapid availability of the finished reports online, particularly for ED patients — they no longer had to chase down the reports. Phone calls requesting “wet reads” (dating back to the days of waiting for X-ray films to dry) of inpatient and outpatient exams also were markedly diminished; these reports are now available immediately for those with electronic connectivity to the hospital’s HIS. For those without connectivity, the system faxes a report within 15 minutes to the physician’s office.

With our previous speech recognition system we reduced our average final report TAT to 24 hours, and with the new system, we reduced it even further to six hours. The preliminary report TAT also was reduced to less than one hour for most studies, and we reduced the total number of in-house transcriptionists by two-thirds. Nearly all of our radiologists continue to self-edit between 70 to 80 percent of their own reports, a true testament to the successful adoption of speech recognition technology by both Scripps Health and La Jolla Radiology.

Kris Van Lom, M.D., is medical director of quality for radiology at Scripps Mercy Hospital and chairman of the department of radiology at Scripps Memorial Hospital Encinitas. Contact him at [email protected] .

February 2009