Coder+
Redefining Medical Coding with Intelligent Automation
Coder+ is a web application that combines medical coding with advanced machine learning algorithms to deliver the most efficient coding solution in the market.
As the lead designer, I played a pivotal role in spearheading the discovery, strategy, and design efforts for this innovative solution. Once the project evolved, my role as the sole designer expanded to include a junior designer. Together, we prioritized a user-centered design approach, conducting extensive research, refining concepts, and incorporating user feedback to ensure a seamless and intuitive experience for medical coding teams.
Problem
Balancing Automation and Human Expertise in Medical Coding for Clinical Trials
Medical coding processes in clinical trials face inefficiencies and high error rates due to reliance on manual input. These challenges hinder research progress, compromise data accuracy, and create resource allocation difficulties. While the introduction of machine learning and automation holds promise, the lack of trust in software accuracy poses a significant barrier.
To overcome this problem, I needed to design an automated coding solution that strikes a balance between leveraging the advantages of automation and preserving the expertise and control of human coders. The solution should instill confidence in the accuracy and reliability of the automated coding algorithm, ensuring that human experts can be the ultimate drivers of the coding process when necessary.
Discovery
In collaboration with the product team, I curated a diverse group of end users from various organizations, representing different sizes and operational demographics. These users were carefully selected to ensure a comprehensive representation of the coding landscape and their active engagement throughout our design process.
To gain deep insights into their workflows and pain points, I employed a range of user research methods, including:
​
User Interviews
In-depth interviews with Coding specialists and managers to uncover their unique operational challenges, expectations, and preferences within the coding process.
​
Shadowing End Users
Observing users during coding activities helped us understand how they do their work and where the pain points are.
Competitive & Analog Analysis
Studying existing coding solutions and related industries informed our approach.
Data Analysis
Analyzing user coding decisions showed a strong alignment between manual decisions and algorithm predictions.
Key Insights
Task-Specific Experience
Many coding specialists specialize in either Medical Events or Concomitant Medication coding tasks. Due to the distinct structures of the medication and events dictionaries, it made sense to separate the coding experience by events and medications as well. This insight allowed us to tailor the user experience and optimize efficiency based on task specialization.
Prediction-Based Bulk Coding
Over 96% of the time, manual coding aligned with algorithm predictions. This insight led us to prioritize bulk coding features to enhance user trust and promote automation.
Simplifying Decision Path
Coding specialists using MedDRA Dictionaries primarily rely on the SOC (System Organ Class) and the Coding level (LLT or PT). We simplified the user interface by displaying essential information only, reducing cognitive load.
Design Goals
Armed with new insight and in collaboration with the Product team, design goals were defined to help guide the strategy for success with Coder+
Concept Development
During the concept development phase, we focused on aligning our goals of efficiency, trust in predictions, and task specialization with the insights gained from user research. Here are the key concepts we developed that would guide our product design solutions:
Streamlining Coding Tasks with Predictions
-
Our solution assigns predictions with confidence levels to every task, even if autocoding features are disabled. This promotes trust in the algorithm and enables users to see potential coding decisions.
-
Strong alignment between manual coding decisions and algorithmic predictions led us to prioritize applying predictions to multiple coding tasks at once.
Tailoring User Experience for Task Specialization
-
To accommodate task specialization, we divided coding tasks into Events and Medications tabs. Each tab provides a tailored experience, offering task-specific tools and streamlined workflows that meet the unique needs of coding specialists.
Simplifying Dictionary Search for Manual Overrides
-
When users need to manually override predictions, we simplified the dictionary search results by displaying shortened paths, reducing noise and enhancing visibility.
User Testing & Feedback
To validate and refine our design solutions, we conducted user testing sessions with coding specialists and managers who had participated in our initial interviews.
During the sessions, each user was provided with a guided setup and observed as they interacted with a prototype. Throughout this process, we documented and cataloged our findings, and identified recurring patterns.
Simplify Confidence Percentage Scores
Users found the confidence percentage scores confusing, particularly in determining the appropriate automated threshold to set. The granularity of the scores was leading to concerns about making incorrect decisions based on slight differences (e.g., setting it at 80% versus 90%).​
Enhanced Result Comparison
Users expressed the need to quickly compare the details of one dictionary term result with another, particularly when dealing with similar terms. It was crucial for coding specialists to access specific details to determine the correct coding decision, which the expandable rows made difficult.
.jpg)
Streamlining Manual Override
Coding specialists, having likely already applied predictions to confident tasks, expressed the need for the next task to load automatically after selecting a code in the dictionary, enabling them to complete tasks quickly without interruption or excessive back-and-forth navigation
Iteration & Refinement
Based on findings during testing we refined our design to further improve the experience, a few of these improvements were:
1. Confidence Levels
We realized the need for a new framework that presents the confidence of each prediction in a more user-friendly way, enabling coding specialists to understand and differentiate between confidence levels effectively. Our solution was to abandon the percentage score, and use a more intuitive confidence level framework of “High, Medium, and Low”, as well as a more intuitive definition of these categories:
High = likely to be more accurate than a human coding specialist
Medium = likely to be as accurate as a human coding specialist
Low = likely to be less accurate than a human coding specialist
Slide-in Details Panel
We also identified the importance of providing an efficient and intuitive way for users to compare the details of coding options, converting the layout and interactions using a slide in panel design, as it allows users to quickly compare details of each result facilitating accurate decision-making.
Keep momentum going
By automatically loading the next task from the task list into manual dictionary search, we enable users to complete tasks quickly without interruption or excessive navigation.
Design Challenges
The design process for Coder+ presented unique challenges, including accessing a diverse user group. We leveraged a survey tool within our platform, inviting users to provide feedback and participate in interviews and testing sessions. This approach connected us with committed users and provided valuable insights for informed design decisions.​
Another challenge was gaining stakeholder buy-in due to required backend investment. We involved the product and development leads as observers in interviews and testing sessions and shared condensed video footage called 'Discovery Snacks' to showcase user perspectives. This effectively communicated the potential impact of Coder+ and aligned stakeholders.
By navigating these challenges with creativity and strategic collaboration, we ensured Coder+ became a user-centered solution that transforms the clinical trials landscape.
Outcome & Impact
Transforming Medical Coding in Clinical Trials
Since the MVP release of Coder+ in summer 2022, the impact has been remarkable. Currently, close to 20 organizations have embraced Coder+ as their go-to solution for medical coding in clinical trials. What's even more promising is that over 40 additional organizations have committed to migrating their trials in Spring 2024, once Coder+ reaches full feature parity with the previous coding solution. Notably, many of these organizations were not using our old coding solution, signaling a significant expansion of our user base.
​
As of June 2023, Coder+ has achieved another impressive milestone, with over one third of the trials using Coder+ utilizing the full automation features. Coding managers now rely on reviewing targeted pieces of data, rather than every coding decision, streamlining the coding process and enhancing efficiency. This transition to automation highlights the trust and confidence users have in Coder+'s algorithmic predictions and its ability to deliver accurate coding results.
Conclusion
In conclusion, the Coder+ project has been a journey of insights, innovation, and success. By prioritizing users' needs and involving stakeholders throughout the process, we developed a seamless coding experience that balances the efficiency of automation and the power of human expertise. With growing adoption and commitments from organizations, Coder+ is transforming clinical trial coding.
The knowledge gained through this design process, combined with the successful implementation of Coder+, has solidified our understanding of the importance of user-centered design, the strengths of automation, and the potential for transformative solutions to impact any market.