Speech recognition leads to medical transcription errors
October 3, 2011
Speech recognition software is definitely an issue of concern for medical transcriptionists. If perfected speech recognition programs could put them all out of jobs. However complex medical jargon is not easily recognizable and reliable speech recognition software could prove too costly and too problematic for widespread implementation.
Recent research suggests that speech recognition software leads to a high level of errors when applied to medical reports. The study by Toronto’s Princess Margaret Hospital indicated an error rate as much as eight times higher using speech recognition software. 615 radiology reports were analyzed for the study, half automated speech recognition reports and half transcribed by medical transcriptionists, which demonstrated a 23% error rate as opposed to 4% in reports prepared by medical transcriptionists.
To this medical transcriptionist the high rate of failure in automated reports is predictable. Medical terminology is complicated. Multisyllable words are commonplace along with eponymous diseases and procedures. Medical terms are seldom spelled phonetically and sometimes sound nothing like they are spelled. Also regional dialects affect the pronunciation of even the most commonly used terms. For instance angina, which is the medical term for chest pain, is enunciated differently depending on the physician’s birthplace.
Accuracy is by far the most important consideration in medical transcription. Even the smallest error can have a dramatic effect on the patient’s care and ultimate outcome. 23% is well beyond the acceptable error rate for medical transcription and may point to the demise of speech recognition in medical transcription.