AstraZeneca: Regulatory Strategy & Usability Design for a Leading Pathology Platform

Prioritized Usability Issues with In-Depth Analysis & Design Fixes

Show/Hide Deliverable(s)

I provided a detailed heuristic analysis to highlight discovered issues, prioritizing them by type, severity and frequency, & the resources needed to solve them (i.e. forming initial framework for a future Use-Related Risk Analysis [URRA], and design review process [21 CFR 820.30 (e)]).

For each identified usability issue, I provided a detailed explanation of how it affected users, the costs associated, any regulatory implications (i.e. lack of RUO labelling due to no such requirement for development under EU MDR), and simple solutions that solved for that difficulty.

What This Achieved:

• Provided Precise, Actionable Design Fixes - With a prioritized set of issues with detailed fixes in hand, the project team could establish the timelines and resources needed to steamline the most essential usability concerns.
• Scoped Plan for Iterative Improvements - By prioritizing usability issues according to urgency, the project team was given a long-term roadmap that could further improve the research software as future resources became available.

Identified Workflow Improvement Opportunities & Regulatory Pathways for Potential Commercialization

Show/Hide Deliverable(s)

I surfaced AI pattern recognition improvements to accelerate the core diagnostic workflow while preserving SME oversight, along with missed user needs along the workflow prioritized for the team's discretion (seeding any future Use-Related Risk Analysis [URRA] & potential CDS/SaMD HF Guidance).

With this in hand, I outlined the applicable Device-CDS & SaMD pathway risk categorization and associated design control requirements, and provided an initial cost/effort framework for any future project effort.

What This Achieved:

• System-Based Workflow Improvements - Apart from interface-level improvements, systems enhancements were identifed to steamline key aspects of the users' workflow. I prioritized the opportunity for greatest value first, with additional opportunities available for later consideration.
• Potential Commercialization Pathways - Having examined systems capability in detail, I presented commercialization options to product ownership with all associated benefits, timelines, & regulatory obligations.

Enhanced Insights with Expanded HIPAA-Compliant Datasets

Show/Hide Deliverable(s)

Using the “Safe Harbor” method to anonymize sensitive Protected Health Information (PHI), that information could now be included for deeper examinations, and additional insights shared between pathologists. This also formed the initial outline for any needed Predetermined Change Control Plan (PCCP) as part of a future FDA marketing submission.

What This Achieved:

• A Broader Data for More Meaningful Insights - This provided pathologists with a broader array of data to leverage in their research efforts, potentially revealing new insights.
• A Conscientious Regulatory Approach - Proactively offered a compliant way of including anonymized Protected Health Information (PHI), encouraging stakeholders to broaden AI training data.

Increasing AI Clinical Accuracy Through Systems-Level Design

Show/Hide Deliverable(s)

For both research and potential clinical use (FDA AI/ML Guidance), the model's accuracy needed improvement without compromising underlying training. Introducing a system-level "escape hatch" to allow the system to flag ambiguous edge cases increased measured accuracy from 94% to 99%+, promoting trust in the system while accelerating pathologist judgment for ambiguous cases.

What This Achieved:

• Made Problematic Results Immediately Visible - Problematic results caused by misidentification (i.e. pathway overlap, wrong origin, market confusion, etc.) were recategorized for rapid examination by pathologists.
• Demonstrated Systems-Level Thinking - This change went beyond interface usability concerns, and provided a key systems change to get the best results possible from the QCS AI model.
• Demonstrated Design Investment in Product Success - Far beyond the orginal design scope, offering this additional value showed a deeper commitment & ability for the Global UX Design Team to enable product success.

Deliver a Clearer Design System with "Human-in-the-Loop" Patterns

Show/Hide Deliverable(s)

Using Tailwind CSS as an accessible framework (ADA/WCAG 2.2), I established a unified design system for all system elements, supporting pathologists across complex workflows and setting the groundwork for a potential commercial release (i.e. as design outputs in a Design History File [DHF]).

With AI central to most workflows and a key component of a potential CDS/SaMD release, I redesigned the interface so pathologists could identify, review, and act on system-generated results. This meant building user trust by applying IEC 62366-1 principles to define clear design inputs/outputs, creating the interaction pattern in alignment with FDA AI/ML Guidance.

What This Achieved:

• A Clean, Comprehensible User-Iterface - Integrating all of the usability issues previously identifed, a new design system was created to address these challenges with a clean, simple aesthetic that was evocative of cutting-edge medical technology.
• Provided "Human-in-the-Loop" Design Focus - Incorporating design best practices from both FDA AI/ML Guidance and Google best practices, users were always aware of what was AI-generated, how the system arrived at these findings, and what could be done with them.