Frequency and spectrum of errors in final radiology reports generated with automatic speech recognition technology

Area of Science:

• Radiology
• Medical Informatics
• Speech Recognition Technology

Background:

• Automatic speech recognition (ASR) technology is widely used in healthcare, particularly for generating radiology reports.
• ASR systems frequently produce transcription errors, necessitating thorough proofreading and editing to ensure report accuracy.
• Understanding the types and frequency of dictation errors is crucial for improving the quality of finalized radiology reports.

Purpose of the Study:

• To quantify the frequency of significant dictation errors in radiology reports generated using ASR technology.
• To identify the specific types and spectrum of these dictation errors.
• To compare radiologists' perceptions of error rates with actual findings.

Main Methods:

• A review of 265 radiology reports prepared for multidisciplinary thoracic oncology conferences.
• Comparison of each report with its corresponding imaging study to identify significant dictation errors.
• A survey of departmental radiologists to gather their estimates of overall and individual report error rates.

Main Results:

• 78% of reports had no significant errors, while 22% contained errors.
• The most common errors included wrong-word substitution, nonsensical phrases, and missing words.
• A majority of radiologists surveyed underestimated the actual error rates, with most believing their own error rates were below 10%.

Conclusions:

• Over 20% of finalized radiology reports contained potentially confusing dictation errors.
• There is a significant discrepancy between perceived and actual error rates among radiologists.
• Increased awareness of error frequency and types is essential for developing strategies to enhance report accuracy and patient safety.