Abstract 698 Medical Informatics Platform, Saturday, 5/1

Dictation is widely used to improve the completeness and legibility of medical records. Recent changes in federal documentation requirements have increased the demand for medical transcription. Transcription costs are $0.17-$0.25 per line and may amount to 11% of collected fees for outpatient visits. Computerized dictation is a means of decreasing the cost and improving the turn-around of transcribed documents. Little research has been conducted examining the utility of computerized dictation in medical practice.

METHODS: The times spent dictating and editing outpatient notes were measured with a stopwatch. Dictations were randomly assigned to human or computer transcription. Computer transcription was performed by IBM ViaVoice Software on a Dell 266 mHz Pentium computer with an AWE64 sound card. Data collected included (1) date of visit, (2) date of dictation, (3) time spent dictating, (4) time spent editing, (5) date the final note was completed and filed, (6) length of the final note, and (7) type of visit (initial consultation or follow-up visit).

RESULTS: Data are summarized in the table and are expressed as mean ± S.D. During the study period, there were 434 dictations; 203 were completed by a transcriptionist and 240 by computer. The proportion of initial and follow-up visits transcribed by the two systems was comparable. There was no difference between the two systems in the amount of time spent dictating, however time spent editing was significantly greater using the computer than using human transcription. Time to the completion of notes and the length of notes were significantly shorter using the computer than using a human transcriptionist.

Table 1 No caption available
Full size table

CONCLUSIONS: Currently available computerized dictation systems are practical and fast enough to replace human transcription. There are differences in the types of errors made by computer and human transcription. Human transcriptionists make spelling errors whereas computerized transcription makes word substitutions that are due to probabilities of its statistical language model. These different error types likely explain why editing computer generated reports takes longer than editing reports generated by a human transcriptionist.