Cirrhosis-Rx

How an AI-enabled clinical support system can impact the management of patients with liver disease.

UI/UX | RESEARCH | FACILITATION | PROTOTYPING | TESTING

Purpose

To help clinicians better manage the treatment of patients with liver disease (cirrhosis).

One of many solutions to address the purpose

We designed a clinical decision support system (CDS) that standardizes treatment recommendations for patients with liver disease.

Our impact

We allowed for conversations about ways to standardize practices and understand nuances across different practitioners who have different workflows.

The Messy

DESIGN PROCESS

Understanding the purpose (discussions)

Our client had a vision that stemmed from a challenge in clinical practice for liver disease care management. We had conversations about the initial NIH grant, the client’s early version mockups, and stakeholders’ input to ground us on the purpose throughout the design process.

CLEARING THE FOG

Laying out assumptions and what we need to learn (literature review)

We laid out what we know and what we need to find out about existing technologies out there, current state of practitioners in their clinical care, sentiments around AI in decision making, and other barriers in the ecosystem.

Literature review - secondary research

Finding out the answers to questions (focus groups)

Almost everyone loves a co-design activity. For clinicians, it was no different. It helped them use a different part of their brain and they had fun! We co-created the CDS, recognizing their expertise in the field, while I facilitate the discussions around the exercise.

Client’s initial sketch of his vision.

First round of focus groups with providers (4 sessions) to lay out current state and co-create a potential solution.

SETTING THE RIGHT DIRECTION

Using insights to steer us in the right direction

Through a series of focus groups and meetings with internal stakeholders, we uncovered these key insights:

Cirrhosis management is not standardized and different across different types of providers. However, generalists tend to be less confident in their clinical decision making.
We can leverage existing technology built by UCSF team with a few changes in the algorithm.
Most practitioners wouldn’t trust AI unless there is a clear explanation of how AI generated the recommendations. They would still use their clinical judgment.

These insights allowed us to have greater confidence to experiment with designing a system that could mostly benefit general medicine practitioners to be more confident with their clinical decision making.

Collected feedback from focus group.

First design iteration in black and white after focus groups.

Second design iteration after internal discussion with clients.

Focus groups round two

We conducted another round of focus groups to provide updates on the design, get more feedback, and most importantly, understand how they are currently using the EHR for cirrhosis care, and understand where and how they would want to use the new tool in the context of their current workflow.

SPEEDING AHEAD

Second round of focus groups with the same providers in first focus groups (3  sessions)

Survey testing

Due to the clinicians’ time constraints, we decided to send out a survey after two iterations of the mockups to gather feedback around the interaction. This got us closer to our MVP design.

Survey to get more feedback on the prototype and the interaction.

Iterating on our designs

We iterated four times and received feedback from the users via focus group sessions. We focused on:

What information would be most valuable in making treatment decisions
How this information should be presented, and
What other visual aids are important in the decision-making process

Iterating on the mockups four times to get to final design

Building a Minimum Viable Product with cross-functional teams

Our final mockups informed our MVP build, which facilitated more conversations around data integration, recommendations based on calculated algorithms (not AI driven), and visual design of the dashboard.

Minimum Viable Product

LOOKING BACK TO LOOK AHEAD

AI adoption will take time

Incorporating AI in clinical decision making is full of unknowns and can cause resistance; but this is a good conversation-starter for creating a more standardized practice to manage care and the resources that are needed to do that.

But, there’s hope...

This process helped uncover some of the underlying issues around the standardization of care even without AI. Imagine if this, combined with trained data sets and a good change management plan, can increase confidence in clinical decision making.

Minimum Viable Product.