A new technique has potential to make information transfers easy and more accurate.
Technology that can record and classify the clinical data transferred in a typical nursing handover could improve the accuracy of the information shared and reduce administrative work.
The approach consists of automated processing and human feedback. “First, speech recognition (SR) converts verbal information into written, free-form text,” Hanna Suominen, a senior researcher at National ICT Australia (NICTA), says. “Second, information extraction (IE) is used to analyse the speech-recognised text and fill out a handover form by automatically identifying the relevant text snippets for each slot of the form.” The pre-filled form is then given to a clinician to proof and sign off on.
Dr Leif Hanlen, technology director and principal researcher at NICTA, says, “The spoken text is transformed into clinical information – it’s effectively pre-filling a form.”
Whilst Suominen says any pre-defined form can be used, the NICTA team evaluated the approach with two types: one with six headings and 50 subheadings and another that had five high-level headings.
Hanlen says, “Typically when a nurse handover occurs, it’s verbal, it’s a group dynamic and there’s a number of people speaking and providing information, but the clinical information within that is recorded only at some later point. “There’s obviously an efficiency problem because somebody who could have been giving care is now typing on a computer.”
Maricel Angel, an RN at Calvary Hospital in Canberra who was involved in a trial of the technologies, agrees that missing important patient information is a problem with a typical verbal nursing handover.
“Failures in the communication can lead to some mistakes,” Angel says. “Sometimes it can compromise patient and staff safety. One of the advantages [of the speech-to-text handover] is it reduces medical errors and the time spent on clerical work,” she says.
Suominen adds, “Studies from Finnish and US hospitals have reported turnover time from 1.3 to 5.3 times faster for [speech-to-text handovers] combined with proofing by hand, when compared with manual transcription. This semi-automated approach also brings 60 per cent of the document drafts to sign off in less than an hour, whilst the [number is 30 per cent] for manual transcription.”
Another problem with traditional nursing handovers concerns loss of information, Hanlen says. Oftentimes a handover means summarising, in which something may be forgotten, lost or changed.
“All of that is a loss of information from the original handover, so what’s in the official record, and what’s going on to the next shift and the shift after that is only tangentially related to what was said at the time,” he explains. “We’ve found the loss of information is quite dramatic, even from one shift to the next, but if you go for multiple shift changes and someone who’s in a hospital for a period of time, the information loss … is quite substantial.”
Hanlen says the aim is to take all of the information at the source. The belief is that information loss leads to clinical errors and adverse events, he explains, but adds the more beneficial aspect of a shift to speech-to-text handover is a move towards the way nurses work. “You don’t need to go back and type it all in again having done your rounds, so it’s effectively more closely aligning with the way the nurses do their work as it is.
“What we’re looking for here is not to replace the nursing handover completely with some automated lapel mic. What we’re doing is trying to fit within a nursing practice.”
Suominen says, “Structuring free-form text, through IE or by hand, eases finding and using relevant information, as well as making this content available for computerised decision-making and surveillance in healthcare.”
She says when using SR and IE in combination, nurses can keep the context of information, track changes and perform searches on both the transcriptions and forms.
Angel says one of the benefits about speech-to-text is that you have access to the data in front of you, but this can be a negative if nurses don’t have access to computers or if there are any technological failings. “It’s easy to use, so I don’t think there will be a problem unless there is a computer breakdown,” she says.
Angel would be interested in using speech-to-text if the system were implemented in her hospital, but says the right training must be involved.
Suominen says, “The approach makes the document drafts available and accessible almost instantly, to all people with an authorised access to a particular patient’s documents, as well as to other technologies for computerised decision-making and surveillance.”
In addition, she says, it avoids problems related to writing the handover documents towards the end of the shift. Instead, nurses can proof their documents when all details are still fresh in their minds.
“I believe nurses can fully focus on their patients and technologies and will seamlessly collect information to document drafts,” Suominen says. “Then the nurses and patients can together read and proof the outcomes. This is a major contrast to the current reality, where either no or limited notes are taken or the patients are left alone for substantial amounts of time while the clinicians are typing notes into their computers.
“I think the nurses can be better present and available to their patients without compromising the careful documentation side.”
Whilst the entire system has not yet been trialled, the team has tested methods for recording the voice for nurses and feasibility of various technologies for recording.
Four out of five clinicians interviewed preferred the lapel microphone over a headset and liked the smaller recorder size.
“One of the concerns the nurses raised early on was they didn’t want to have a headset with a boom mic out the front of it,” Hanlen says. “We’ve shown that reasonably simple and reasonably low-cost microphone technology can be used.”
Suominen says, “[Nurses’] preference is multipurpose recorders, such as tablets, that allow them, for example, to proof and sign off the resulting handover form as well as see other electronic clinical records.”
There was a focus on using technology that would not be too expensive. Suominen says money is not a huge factor when it comes to SR quality: “A $15 noise-cancelling lapel microphone in combination with a $200 digital voice recorder gives the best correctness.”
Another aspect that was looked at in trials was the accuracy of transcriptions of clinical text.
“SR achieves an impressive word correctness [90–99 per cent], with only 30 to 60 minutes of training to a given nurse’s speech,” Suominen says. “In other words, correcting SR errors by hand as a part of clinical proofing is not likely to be time-consuming.
“Although these percentages originate from noiseless test laboratories, SR can recognise over 77 per cent of words correctly in settings with background noise, interruptions, and other factors complicating the situation in clinical reality.”
In terms of IE, it was found to perform well at filtering out irrelevant text. Suominen says it is “close to perfect” at extracting information such as the patient’s current room and bed. However, she adds more work is needed to improve its performance with abstract and verbose information such as identifying goals, tasks to be completed and expected outcomes for future care.
Suominen says, “The speech-to-text approach that combines SR with IE … makes the resulting electronic structured data available for efficient re-use.”
She says whilst several clinical SR engines and services have been developed and are commercially available, they are not applicable to nursing handovers, as typical use is in a peaceful office where one person uses their personalised SR engine at a time.
“This is far from the clinical reality of a noisy ward where nurses have their handover discussions and personalised solutions are not applicable,” she says.
She adds that whilst IE has been widely studied for clinical applications, she doesn’t know of research that includes filling out handover forms or surrounding combined use with SR.
“If considering form-filling together with SR, the current commercial approach is menu based; a clinician chooses a relevant heading from a menu and then speaks text that is relevant for that heading,” she says.Do you have an idea for a story?
Email [email protected]