2. www.medicaltranscriptionservicecompany.com 918-221-7809
Accuracy in the electronic health record (EHR), which experienced medical transcription
companies help physicians ensure, is critical to quality care. Health IT Analytics recently
reported on a new study which found that patient safety could be compromised as a result
of clinical documentation produced with the help of speech recognition (SR) software. The
study, published in JAMA Network Open, found an error rate of more than 7 percent in
speech recognition–generated clinical documents before review by clinicians and medical
transcriptionists. The researchers point out that their findings highlight the significance of
manual editing and auditing of HER documentation.
How SR Software leverages Natural Language Processing for Healthcare
Many US clinicians rely on dictation services supported by SR technology and professional
medical transcription outsourcing companies. SR technology leverages natural language
processing (NLP) to produce clinical documentation. Using computer algorithms, NLP
processes and extracts meaning from the natural language like speech or written text. SR
and medical transcription have been adopted successfully in the dictation and generation of
radiology reports.
Medical NLP systems specifically perform the following tasks:
- Carrying out speech recognition, allowing users to dictate notes that can be instantly
turned into accurate text format
- Mining, locating, and summarizing key concepts or phrases from narrative texts such
as clinical notes or a patient’s account for further analysis
- Mapping unstructured text data to structured fields in the EHR record to improve
clinical data integrity
- Answering unique free-text queries that require the blend of multiple data sources
- Using optical character recognition (OCR) to turn PDF documents or scans of care
summaries and imaging reports into text files for parsing and analysis
According to a study published in the Journal of the American Medical Informatics
Association in 2016, using SR software dramatically reduced overall turnaround time (TAT)
for report creation by clinicians and healthcare organizations. SR-based systems make
reports instantly. However, the study notes that this improvement in TAT “hides an editing
and document creation time cost that falls directly on the clinician”. The new study
published in JAMA Network Open revealed the higher error rate (number of mistakes per
3. www.medicaltranscriptionservicecompany.com 918-221-7809
100 words) in speech recognition–generated clinical documents, which demonstrates the
importance of editing and review by a clinician or professional medical transcriptionist.
Errors due to Speech Recognition – Study Findings
The cross-sectional study published in JAMA Network is based on a stratified random sample
of 217 notes (83 office notes, 75 discharge summaries, and 59 operative notes) dictated by
144 physicians between January 1 and December 31, 2016, at two health care
organizations using SR software. The findings are as follows:
- Up to 96.3% of the 217 notes had at least one error directly after dictation and
before review by human transcriptionists or physicians themselves.
- 15% of these errors were related to clinical information, and 5.7 percent of the
mistakes were clinically significant.
- Nearly two-thirds of documents had at least one clinically significant error, with an
average of 2.7 such errors per note.
- The highest percentage of errors was medication-related, but after transcriptionist
and physician review diagnosis errors were the most common issue
- Discharge summary documents had the highest proportion of errors.
- Surgical notes had the fewest data integrity lapses after NLP production
To detect the error rate, the SR engine–generated document (SR), the medical
transcriptionist–edited document (MT), and the physician’s signed note (SN) were compared
with a criterion standard generated from the original audio recordings and medical record
review.
According to the study published in the Journal of the American Medical Informatics
Association, other studies noted the error types in SR documents as word omission, word
substitution, nonsense phrases, wrong word, punctuation errors, incorrect measurements,
missing or added “no” other added words, verb tense, plural, spelling mistakes, or
incomplete phrases. Errors that change meaning in a way that is not obvious to immediate
inspection are the most serious. Four of the five studies reported such error types in
document, including: wrong patient, dose, lab value, and anatomical side errors.
According an article in Health It Analytics citing Hilary Townsend, MSI, in the Journal of
AHIMA in 2013, clinical notes make heavy use of acronyms and abbreviations, which have
4. www.medicaltranscriptionservicecompany.com 918-221-7809
multiple meanings, and more than half of terms, acronyms, or abbreviations typically used
in clinical notes are “puzzlingly ambiguous”.
For instance, the term ‘discharge’ can refer to either bodily excretion or release from a
hospital. A human transcriptionist can decipher these types of differences by relying on the
context of the surrounding words for clues, but NLP technology has cannot.
Professional Medical Transcription Services reduce EHR NLP Errors
It is clear from these studies that, despite the efforts to enhance SR accuracy and reliability,
unstructured clinical notes and narrative text present a major problem for even the
smartest NLP algorithms. Experts say that even a small proportion of clinically significant
errors can have a negative impact on patient care.
Medical transcriptionists in established medical transcription companies are knowledgeable
about anatomy and physiology, disease processes, signs, and symptoms. Highly trained in
complex medical terminology, they can interpret physician dictation and profession-specific
shorthand and even catch errors and supply edits or corrections. They have adapted to
changing industry needs and deliver high quality EHR-integrated documentation. The new
study found that error rates fell significantly to just 0.4 percent when transcriptionists
performed checks of clinical documentation generated by the NLP system.
Natural language processing tools have great potential to reduce the complexities of EHRs.
Success depends on developing algorithms to improve the extraction and presentation of
clinical decision support data.