Electronic Health Records

Adoption & Implementation News

Does Speech Recognition Aid Clinical Documentation Improvement?

A review of literature on speech recognition shows mixed results for clinical documentation improvement.

By Kyle Murphy, PhD

- Despite the availability of speech recognition software and natural language processing over the past two decades, research shows limited evidence proving these technologies to have a clearly positive impact on clinical documentation improvement.

Speech recognition for clinical documentation improvement

That conclusion comes from a systematic review of literature on the risks and benefits of speech recognition for clinical documentation by Tobias Hodgson and Enrico Coiera published recently in the Journal of the American Medical Informatics Association.

"SR is a widely used input modality for modern computer devices and has a long pedigree in the clinical setting," Hodgson and Coiera write. "Surprisingly, our review revealed that the evidence base documenting the benefits and limitations of SR’s use for clinical documentation is limited, incomplete, and relatively neutral to its benefits. Recent studies, which would benefit from more modern SR technologies, are absent."

The team of researchers pared the number of studies down from 538 to 23, which comprise the focus of their quantitative and qualitative analysis, spanning from the efficiency and accuracy of speech recognition to errors introduced by speech recognition and the cost–benefit of speech recognition.

The authors did note evidence supporting system-level benefits for improving clinical documentation speed in "dramatic reductions" to turnaround time (TAT) for report creation, but they exercise caution in divining the meaning of these findings:

This is mainly due to the virtually instant delivery of reports possible with SR based systems. This improvement hides an editing and document creation time cost that falls directly on the clinician. The effective clinical adoption of technologies often depends on local costs being offset by local benefits, and the relatively low uptake of SR to date might in part be due to an imbalance in cost over benefit for the clinician preparing reports.

Improved accuracy was another win for speech recognition in the systematic literature review. Yet the margins for error in medicine make the less-than-100-percent accuracy of this technology still problematic, according to Hodgson and Coiera.

"In fact many SR software developers now claim accuracy rates of up to 99%," they state. "However, high accuracy rates do not necessarily mean that SR is clinically safe, and several studies have reported a range of errors, some of which are clinically significant and could lead to patient harm."

Numerous errors are listed in the review:

  • Creating documentation for the wrong patient
  • Wrong drug name or dosage
  • Wrong lab values
  • Left/right anatomical discrepancy
  • Medical discrepancy
  • Age or gender mismatch
  • Wrong doctor name
  • Wrong date
  • Made up words and acronyms
  • Irregular spacing
  • Spelling errors, omissions, or duplications

In the review, the researchers also call attention to a lack of data identifying the numerous variables having an effect on speech recognition (e.g., user training, environmental conditions).

"In the absence of such data, there is the need to cautiously generalize the performance reported in these studies to expected real world performance. In other words, good performance under controlled conditions may not be replicated in clinical settings," they maintain.

Hodgson and Coiera list six areas for future research to focus on to evaluate the merits of using speech recognition for clinical documentation improvement:

  • Impact on clinical processes and outcomes
  • Impact on clinicians
  • Impact on patient safety
  • Comparative effectiveness
  • Effectiveness for non-documentation tasks
  • Alternate input platforms

Recently, CHRISTUS Health in Texas shared details about its recent initiative to improve the quality of clinical documentation by implementing health IT from Nuance — both a quality reporting system for radiologists with speech recognition and natural language processing technology for providers in ambulatory and inpatient settings.

According to CMIO Luke Webster, tangible benefits are already emerging in the form of reducing turnaround times and quality reporting such as reducing the duration for issuing an imaging order and receiving an official sign-off from providers. “That has improved substantially,” said Webster.

And the health system is eagerly awaiting more.

“One of the hopes (and obviously the plans that we have in place) is to leverage that foundational technology to improve clinical documentation real-time, such as prompting providers as they document,” added Webster.

Continue to site...