Designing Patient Insights Module for an Incubation Lab
Author
David Kim
Scope
UI/UX Design
Architecture
Concept
The Patient Feedback module was conceptualized to demonstrate how an incubation lab might develop a data-driven insights dashboard utilizing AI. Drawing from experiences doing customer discovery with research labs in various industries, research teams often struggle to manage large amounts of patient data during clinical trials.
Our solution enables researchers to collect, analyze, and act on real-time patient feedback, leveraging AI to automatically gather insights and interact with an extensive database of previous clinical trials and research.
The Incubation Dashboard serves as the cornerstone, with the example project "CardioTrack" demonstrating how tools like Patient Feedback can enhance trial outcomes through real-time decision-making.
First, let’s briefly explore the architecture of our (mostly) React application.
We’ve offloaded resource-intensive AI/ML processing to the backend, while simpler, real-time analysis is handled by IRIS, our AI assistant, directly on the frontend. This structure allows us to scale efficiently to accommodate larger trials, more participants, and evolving features.
A robust set of data sources is interfaced through a GraphQL API, managed by Apollo Client on the frontend, with Redux for state management and React Query (or alternatives like TanStack Query or HTMX) for optimized data fetching. Additionally, we’ll incorporate modern tools like Vercel’s AI SDK for generative UI, enabling dynamic charts and interfaces from IRIS.
The layout of the module is built on a 12-column grid system to ensure the interface remains both flexible and functional. This grid allowed me to organize complex content types—such as interview transcripts, patient insights, and AI-generated summaries—into distinct, easy-to-navigate sections without cluttering the interface.
I chose blues for accent colors, which tend to give a product a more understated, "scientific" feeling, to interact and layer on top of warmer parchment-like neutrals. When paired with a sophisticated serif heading like Kilm Type Foundry's Martijn Plantijn, the application starts to feel like a journal, beckoning the user to interact and engage with the platform as though they would sit with it next to a cup of coffee, cross-legged at the breakfast table.
These are invocations that are absolutely necessary when dealing with the health of patients, where mindfulness and presence are essential to the quality of care & research.
Top Insights
IRIS
Hi Nadia, how can I assist you with the trial insights?
Can you tell me how Linda’s positive outcomes compare with other patients?
Linda has experienced notable improvements, particularly with her cholesterol levels due to the dietary intervention. Compared to other patients, Linda’s outcomes align with those who have adhered well to the diet changes. Patients like James Miller and Sarah Thompson also reported similar improvements in cholesterol and blood pressure.
How effective is this treatment compared to the CardioGuard trials last year?
In comparison to the CardioGuard trials conducted last year, the CardioTrack trial shows similar improvements in cholesterol and blood pressure metrics across participants. However, CardioGuard had a lower overall patient adherence rate, with adherence dropping by nearly 42% due to more severe side effects such as dizziness, fatigue, and nausea. Here’s a visualization of the key data comparing adherence rates and reported side effects between the two trials.
CardioGuard vs CardioTrack Trial Comparison
Adherence Rates
100
75
50
25
0
The blues are applied more liberally within the Insights pane. The most critical insights are displayed on the top. which can be hovered over, or clicked, to provide more information. The elevation of the card expresses the primary focus for the researcher-user: the interaction with IRIS (see tooltip).
We imagine this AI assistant to have deep capabilities; able to extract critical insights from the clinical trial, link insights to hypotheses, and even create a graph using generative UI, comparing effectiveness to last year's trials.
Thank you for taking the time to read this case study.
This project, including the case study, was completed within 14 hours.