ABSTRACT
The number of clinician burnouts is increasing and has been linked to a high administrative burden. Automatic speech recognition (ASR) and natural language processing (NLP) techniques may address this issue by creating the possibility of automating clinical documentation with a "digital scribe". We reviewed the current status of the digital scribe in development towards clinical practice and present a scope for future research. We performed a literature search of four scientific databases (Medline, Web of Science, ACL, and Arxiv) and requested several companies that offer digital scribes to provide performance data. We included articles that described the use of models on clinical conversational data, either automatically or manually transcribed, to automate clinical documentation. Of 20 included articles, three described ASR models for clinical conversations. The other 17 articles presented models for entity extraction, classification, or summarization of clinical conversations. Two studies examined the system's clinical validity and usability, while the other 18 studies only assessed their model's technical validity on the specific NLP task. One company provided performance data. The most promising models use context-sensitive word embeddings in combination with attention-based neural networks. However, the studies on digital scribes only focus on technical validity, while companies offering digital scribes do not publish information on any of the research phases. Future research should focus on more extensive reporting, iteratively studying technical validity and clinical validity and usability, and investigating the clinical utility of digital scribes.
ABSTRACT
BACKGROUND: Patient experience surveys often include free-text responses. Analysis of these responses is time-consuming and often underutilized. This study examined whether Natural Language Processing (NLP) techniques could provide a data-driven, hospital-independent solution to indicate points for quality improvement. METHODS: This retrospective study used routinely collected patient experience data from two hospitals. A data-driven NLP approach was used. Free-text responses were categorized into topics, subtopics (i.e. n-grams) and labelled with a sentiment score. The indicator 'impact', combining sentiment and frequency, was calculated to reveal topics to improve, monitor or celebrate. The topic modelling architecture was tested on data from a second hospital to examine whether the architecture is transferable to another hospital. RESULTS: A total of 38,664 survey responses from the first hospital resulted in 127 topics and 294 n-grams. The indicator 'impact' revealed n-grams to celebrate (15.3%), improve (8.8%), and monitor (16.7%). For hospital 2, a similar percentage of free-text responses could be labelled with a topic and n-grams. Between-hospitals, most topics (69.7%) were similar, but 32.2% of topics for hospital 1 and 29.0% of topics for hospital 2 were unique. CONCLUSIONS: In both hospitals, NLP techniques could be used to categorize patient experience free-text responses into topics, sentiment labels and to define priorities for improvement. The model's architecture was shown to be hospital-specific as it was able to discover new topics for the second hospital. These methods should be considered for future patient experience analyses to make better use of this valuable source of information.
