How Dragon Copilot is helping clinicians spend more time with their patients

Maintaining medical records for tens of millions of people in the UK has become a colossal administrative challenge for the NHS. A recent trial of Microsoft’s AI-driven healthcare tool Dragon Copilot has improved efficiency and enhanced the doctor-patient relationship.

One of the essential responsibilities for clinicians (healthcare professionals, such as a doctor or nurse, who provide direct patient care) is accurate record-taking. This adds a substantial administrative burden to each consultation and takes up a considerable amount of time.

The NHS has been undergoing a period of fundamental change, following the rollout of the electronic patient records system, a digital tool that brings together all of a patient’s medical information into a single, unified platform.

A small group of clinicians at Manchester University NHS Foundation Trust were invited to be the first in the UK to try Dragon Copilot. Part of a private trial, the aim was to assess the impact the technology could have on efficiency.

The Dragon Copilot operates by transcribing notes based on the conversation, together with commentary provided by the clinician during an examination. Previously, note-taking was undertaken by the clinicians themselves, either during or following a consultation.

By connecting a microphone to the system and processing the incoming audio through Microsoft Dragon, the system was able to generate patient consultation summaries based on the recording.

“We've been live for a couple of weeks, so it's early days, but having it running in the workflow and populating into our clinic notes is making a fairly substantial difference,” says Peter-Marc Fortune, paediatric intensive care physician and chief medical information officer with Manchester University NHS Foundation Trust.

During the development process, recordings will remain available for a short period of time so that any issues can be fully explored. Once the production version is deployed, the intention is not to store the audio recordings.

Lost in translation

In general, audio transcriptions are rarely completely accurate. In the trial, this was especially the case for complex and technical terms, which the system was unfamiliar with. It would then tend to use the closest-sounding word in its dictionary. As such, a human review of the generated summary and treatment plan still needs to be undertaken in order to ensure accuracy.

“Whilst it's remarkably good, we have seen some extraordinary things occasionally on it – more so when we first went live, than we're seeing now,” says Fortune.

“Every now and again, it picks up something unexpected, usually around the technical language, where it misinterprets a complex medical word and uses one that sounds fairly similar but is completely different. The necessity to ensure that you read it is something that we need to embed with our clinicians.”

Early in the trial, notes needed to be copied across into patient records. However, an embedded version of Dragon Copilot was released, which was able to automatically populate the patient’s notes once the consultation was complete.

“The greatest challenge was NHS network infrastructure, and particularly Wi-Fi. One of the things that we're in heavy discussions about is the product being optimised,” says Fortune. “We're quite keen to explore a good old-fashioned USB microphone approach, where we've not got to involve Wi-Fi infrastructure.”

As Dragon Copilot was developed in America, much of its structure and formatting was designed with the American healthcare system in mind. One example of this was the date formatting system, which had to be changed from America’s month-date-year format to the UK’s date-month-year. There are also differences in how patient summaries are structured, which need to be updated to suit the NHS’s template.

Greater engagement

Although patient feedback was limited, one of the key improvements that was found in the Dragon Copilot trial was that patient engagement was much improved.

Instead of the doctor having to type their observations as the patient is talking to them, the doctor was able to rely on Dragon Copilot to record the consultation and thereby focus more of their attention on the patient. It was particularly noted that the older patients preferred this approach, as it felt more like the ‘traditional’ doctor’s consultation patients experienced in the past.

The initial trial was akin to a toe-in-the-water, but future trials of the Dragon Copilot will focus on the administration aspect and on obtaining more detailed feedback from patients.

The next stage of the trial will be expanding the functionality of Dragon Copilot so that it incorporates more smart links within the Epic electronic health record system to pull specific patient information directly into documentation. For example, it could automatically generate orders for prescriptions and examinations (such as X-rays), thereby improving efficiency and reducing the administrative burden on clinicians even further.