AJR: Speech recognition linked with errors in half of breast imaging reports

Reid Conant, MD, of Tri-City Medical Center, using front-end speech recognition software.

Breast imaging reports generated using speech recognition were eight times as likely as reports dictated with conventional transcription to contain major errors, according to a study published in the October issue of American Journal of Roentgenology. The researchers estimated that nearly 25 percent of the reports in the study contained a major error that could have affected understanding of the report or altered patient care.

Automated speech recognition (ASR) is associated with multiple advantages: reduced report production time and cost savings compared with conventional dictation. Previous studies, however, have suggested that ASR-generated reports are plagued by a high error rate.

Sarah Basma, of the research and development division at Mount Sinai Hospital, Women’s College Hospital in Toronto, and colleagues sought to determine the frequency and spectrum of dictation errors in breast imaging reports generated with ASR compared with conventional transcription.

The research team reviewed 615 breast imaging reports discussed at multidisciplinary tumor boards from January 2009 to April 2010; 308 were generated with ASR and 307 with conventional transcription. Modalities in the reports included mammography, breast ultrasound, breast MRI, interventional procedures and x-ray. Most studies were diagnostic exams and of higher complexity than screening studies, the authors wrote.

The researchers categorized errors as major if they affected understanding of the report or patient care. Among the 308 reports generated with ASR, 52 percent contained at least one error. The corresponding rate among conventional reports was 22 percent.

Basma et al reported that major errors were more common in MRI and combined mammography and ultrasound reports. However, academic rank and native language were not found to be independent predictors of error.

The authors provided suggestions for reducing the error rate. They noted, “Reviewing reports between six and 24 hours after dictation may be helpful in detecting errors that can be missed when reports are verified immediately, as with ASR.” They also suggested alerting radiologists to the high prevalence of errors in ASR reports to curb the tendency to superficially edit reports.

In addition, because background noise may contribute to errors, it may be advisable to take steps to reduce noise.

According to Basma and colleagues, errors may create problems beyond compromising patient care. “Even though most errors detected in our study did not alter patient care because all cases had been discussed in multidisciplinary team meetings, the presence of major errors tends to make reports confusing and difficult to read.” 

They noted that clinicians often judge radiologists based on the radiology report. Reports riddled with grammatical errors, they wrote, can indicate a lack of professionalism and carelessness.

After acknowledging that the incidence of errors may have been artificially inflated in the study due to the complexity of the imaging studies and lack of standardized templates, Basma and colleagues concluded, “Careful editing of reports generated with ASR is crucial to minimizing error rates in breast imaging reports.”