Artificial intelligence is rapidly transforming the way clinicians document patient visits. Combining speech-recognition technology and AI, digital medical scribes hold promise for streamlining workflows and freeing up valuable clinician time.

However, unlocking the potential of this technology requires more than just the tech itself. The challenge lies in creating a seamless human-machine interface, one that integrates effortlessly into busy clinical workflow, where time for adapting to new technology is extremely limited.

Currently, there are few guidelines and independent case studies for optimal implementation. To ensure AI-powered scribes become powerful allies in the clinician’s digital toolbox, we need evidence-based data on what works and what does not work in adopting scribes, including infrastructure, culture, workflow models, assumptions, and training approaches.

The recent one-year trial of a commercial digital scribe at the South Metropolitan Health Service of Western Australia (SMHS) provides some valuable lessons. The software we trialled is a so-called “human-led” digital scribe, that is, a desktop-based real-time dictation system for clinical documentation, which includes text templates and personalised voice commands.

We trialled the software to dictate clinical notes at the cursor in our Digital Medical Record (we excluded clinical letters from our trial, as these are transcribed by an external transcription service). Despite the impressive accuracy of the system, ultimately our trial was unsuccessful, with adoption by clinicians well below initial expectations.

Albeit unsuccessful, our trial has served as a goldmine of insights into embedding digital scribes and digital technology into clinical workflows. Throughout the trial we collected data on adoption, user experience and impact on clinicians’ wellbeing and workflows through software analytics, surveys and individual interviews with clinicians.

Cool tech v reality

The key takeaway here is that simply throwing cool tech at clinicians does not guarantee a win. The challenge lies in integrating technology in complex IT and clinical workflows, where multiple tasks already compete for clinicians’ attention.

For starters, the trial challenged a key assumption that we made at the start of our journey. We took it for granted that a voice-to-text scribe would be a plug-and-play tool, something that any clinician would be ready to use with little or no training. Instead, what we found is that time required for training was the highest barrier to adoption.

Although a user training session was expected to take less than two hours, clinical staff simply did not have the time to invest in learning a new piece of technology for what was perceived as a marginal benefit.

Lack of time capacity is also linked to another issue: standardising how clinicians write their notes. Digital scribes can realise their full benefit when dictation can be converted into standardised templates. Templates compress documentation time by removing the need to dictate repetitive content (note headings, standardised advice etc.) at each single interaction.

During our trial, we realised that, in most clinical specialties, we lacked standardised templates that clinicians could readily access. Off-the-shelf templates were not a suitable solution, as these needed to be adapted to the conventions of each clinical area (e.g. acronyms, checklist preferences etc.).

Designing new templates from scratch to be used by the digital scribe was deemed too time-demanding, especially as this would have required extensive consultation with time-stretched senior clinicians.

Another key finding is that we should not assume a one-size-fits-all approach. Different clinicians have different workflows and pain points. For some, typing might not be such a big hurdle, especially if existing solutions are already working well.

For example, doctors who already have access to human transcribers for their clinical letters and can delegate clinical notes to junior staff found limited value in being able to dictate their notes in real time. On the other hand, allied-health professionals were the most engaged users in our trial, as the voice-to-text software was the sole alternative to typing that was available to them to take their notes.

Dictation in open-plan offices

The trial also highlighted the challenges of voice dictation in open-plan offices. Multiple users reported discomfort with dictating in open-plan offices, not only because of concerns about patient privacy, but also because of the perceived risk of causing nuisance to colleagues. Dictation in the wards was also generally perceived as distracting and as a possible hindrance to communication between senior staff, junior doctors and the patients.

Complex IT infrastructure in the hospital also hindered adoption. With a mixed of local desktops and virtual-desktop environments, installation of the software proved more difficult than expected, especially to enable seamless integration with hardware (microphones).

And then there’s the tech itself. The program is undoubtably impressive and reflects state-of-the-art in software for clinical dictation. Yes, it is not perfect. Minor glitches and minor design faults can turn into major turn-offs.

As our users were highly time constrained, small technical imperfections often led them to abandon the product altogether rather than reaching out to the project team or tech support for help. Further training might have helped some users to navigate the technical limitations of the software, but this clashed with the lack of time for follow-up training session.

Finally, some clinicians felt the solution did not fully meet their expectations. They envisioned a fully mobile-supported and AI-powered system that could capture the entire consultation and automatically transcribe it into a note.

That is, they asked for ambient clinical intelligence (ACI). This challenged our assumption of an incremental roadmap to digital scribes, starting with human-led voice-to-text software and ending with ACI implementation.

Our trial outcomes suggest that a phased approach might not always work for certain technologies. In some cases, waiting for the ideal solution might be better than asking clinicians to adapt to a stepping stone.

The human element

SMHS’s experience is a valuable reminder: successful healthcare innovation hinges on understanding the human element. It’s not just about the tech – it’s about fitting it seamlessly into the complexities of real-world clinical practice.

By prioritising user needs, thoughtful planning, and continuous improvement, we can ensure that digital scribes and AI becomes a powerful tool to empower clinicians and ultimately, improve patient care.

The learnings extend beyond this specific trial. In our opinion, here is what our experience means for the broader healthcare industry.

• User-centred design is king. Involving clinicians from the get-go, understanding their workflows, and tailoring AI solutions to address their specific needs is paramount.
• Focus on solving real problems. Ensure AI and other health documentation technology addresses genuine pain points and demonstrably improves healthcare delivery.
• Embrace flexibility. A one-size-fits-all approach rarely works. Be prepared to adapt AI solutions to different specialties and workflows.
• Don’t underestimate the learning curve: Be realistic about the time investment required for clinicians to learn new systems. Balance the learning curve with the expected benefits.
• Invest in user training. Make training accessible, quick and relevant, focusing on how the technology benefits clinicians and ultimately, patient care.
• Prioritise user experience. Ensure the technology is user-friendly and integrates seamlessly into existing workflows. Even a slightly clunky interface can be a major barrier to adoption in health care.
• Consider the bigger picture. Think about how the technology will function within the broader hospital environment and what infrastructural and/or cultural change this may require. For example, how will voice dictation work in open-plan offices?
• Be mindful of user expectations. Sometimes, a phased approach might not be ideal. In some cases, waiting for a more comprehensive solution might be better than asking clinicians to investing time in solutions that only deliver partial outcomes.

With AI expected to disrupt the healthcare sector in the coming years, it is essential that we do not forget the humans that will use this technology. Healthcare is people caring for people, and every piece of technology that we want to embed into it must have people’s needs at its core.

Francesco De Toni is the innovation senior project officer at the South Metropolitan Health Service in Perth.

Are you a CHIA member? Reading this Pulse+IT article entitles you to CPD points. Click here to record your participation.