OpenRoom
Utilizing the power of AI to automate and streamline the hospital discharge form generation.
UX Design - AI Health Web Application
Project Overview
Hospitals lose time and money to discharge delays. Hospital discharges take up to 2.2 hours per patient and on average discharge summaries consume 135 hours a week for the medical workforce.
Delays in completing hospital discharge summaries have also been linked to a greater risk of patients returning to hospital within 30 days — as each day’s delay in finalising a summary was associated with a 1.6% increased chance of readmission.
We proposed OpenRoom, an AI system that extracts medical notes, generates accurate draft summaries and updates bed managers in real time.
Gabrielle Versace
Pragrun Rashid
Pete Argent
Emannuelle G
Team
Role
UX Designer / Researcher
Discover
Our team began the design process by contextualising the issue through academic sources which gave a good overview of the problems that medical practitioners face, followed by interviewing potential users / people familiar with the issue, these people were:
Senior Doctor
Junior Intern Doctor
Radiologist
Research Goals
Understand how discharge summaries are currently written (tools, workflows, pain points).
Validate if OCR + NLP automation could reduce effort and backlog.
Test whether clinicians trust/edit AI-generated discharge drafts.
Identify what would make an AI-assisted discharge tool adoptable in a high-pressure hospital setting.
Findings from academic research
Writing a discharge summary took a median of 14.5 minutes, with almost six minutes of that spent on “thinking time” or searching for information rather than typing.
Some doctors are not familiar with using the new electronic discharge summary system vs interim paper based system.
Lack of computer terminals.
Reviewing the quality of discharge summaries by printing anonymised versions of discharge summaries and having a group-based assessment of those summaries by all grades of doctors. With individuals free to give suggestions for improvements. Learning points from discussion of the anonymised summaries were fed back to team members.
‘I didn’t recognise myself in my own discharge summary. Consultants and registrars have to take some responsibility – it can’t just be dumped on a poor junior doctor.' — Dr Carl Mahfouz, GP and University of Wollongong academic
Canberra women’s health GP Dr Gillian Riley said the issue is that discharge summaries are not considered important.
Findings from user research
Searching for the relevant information in the clinical notes takes most of the time.
Lack of planning in the preparation of the discharge summary— afterthought rather than a planned process.
The discharge summary is often given to junior doctors who have not treated the patient. Which causes extra time because the [junior] doctor needs to review and learn the entire patient history to summarise it which is timely.
The senior doctor then takes over the discharge summary but has no ownership over it. "It's chaotic."
A draft which needs to be edited by the clinician, would only work well if the person using it is the treating doctor (which is what should be happening). Doctors feels that the treating doctor should be [handling discharge summary].
Define
I ran a collaborative workshop with the team consisting of our UI designer and AI developers to bring the team to a collective understanding of the problem space and brainstorm solutions that were viable for the AI developers of our team to implement and understand the 'why'.
We utilised a pre-existing discharge summary template and referred to a real-life example of a discharge summary from a group member who had been in hospital to understand how it had been used on the patient side.
Ideate —> Test
Based on our collaborative workshop, the team collectively proposed the solution of an intelligent document processing system that automatically generates complete discharge forms by analysing existing medical records.
We prototyped our solution in AI assistant Bolt and asked a small sample of people to test the usability and give feedback.
Try it out below ↓
I then took screenshots of the AI prototype and collaborated with the UI designer to translate the feedback from user testing into UX recommendations and UI changes.
As users voiced concerns over the accuracy of AI in extracting information from documents, I recommended an 'Inspect Mode' within the 'Edit Mode' which would allow users to inspect which documents the system had extracted information from.
Information that had been extracted from documents would be highlighted and users would be able to view the source document/s in the side panel and where the information was specified.
I quickly wireframed this recommendation in Figma to visually represent to the developers, along with a brief explanation on the shared Miro board.
Implementation
Impact
Although this system has not yet been implemented in actual medical settings, success metrics of the system would include:
Decrease in time spent writing discharge summaries.
→ Under the current median of 14.5 minutes.
Decreased rates of patients readmitted within 30-day window post-discharge.
Decrease in cost per patient (delayed discharge cost).
>90% clinician satisfaction with clarity and usability of summaries (via surveys or metrics of Edit function used).
Increased efficiency in bed turnover: Average time from discharge decision to bed availability reduced from 6 hours to <5 hours
Key Takeaways
AI was a major asset in rapid prototyping and testing of our solution.
Collaborative workshopping was a major asset in bringing developers on-board with UX recommendations.