Feedback Loops for better Machine Learning
Assignments
Highlighted filters
Healthcare AI/ Pharmacovigilance/ Smart Data ENtry
Streamlining Pharmacovigilance Data Entry with AI and Machine Learning
Design and development of a user-centered solution that automates the complex manual data entry process in pharmacovigilance, reducing the time and effort required for resources to file adverse events into the pharmaceutical database.
Overview
Pharmacovigilance is a globally recognized and regulated process that tracks the adverse effects of any drug. The manual data entry process is time-consuming and requires expertise and experience due to the complexity of diseases and events. This project aims to leverage AI and machine learning to automate the data entry process, ensuring accuracy, efficiency, and consistency.
My Role
Research, UX Design, Prototyping, Visual Design
The Team
2 Designers, 1 Design Lead
10+ Engineers
Timeline
8 Months
PROCESS TRANSFORMATION
NEW DESIGNS
Project Goals
Automate Data Entry
Utilize AI and machine learning to automate the extraction, classification, and entry of adverse event data from source documents.
Improve Data Quality
Ensure data accuracy and consistency by leveraging design systems and real-time validation.
Increase Efficiency
Reduce manual effort and time spent on data entry, allowing resources to focus on high-value tasks.
DESIGN GOAL
How Might We Streamline Pharmacovigilance Data Entry with AI and Machine Learning?
Existing Process
NEW PROCESS
Existing Process
Recieve
Users receive document from external portal
Upload
Uploads document in
PV First
Digitize
System digitizes into structured document and autofills the fields
Review
Users review cases using assisted controls
Submit
Submit Document
Manual
Automated
Upload Document
System Performs Data Entry
Add Feedback for wrong highlights
Accordions to
reduce scrolling
Understanding the Users & Process Flows
We conducted qualitative user research to gain a deeper understanding of user roles, backgrounds, responsibilities, and behaviors.
This involved conducting in-depth interviews with 10+ users individually to gather insights into their experiences and perspectives. Additionally, we shadowed users to observe their interactions and behaviors in real-life scenarios. By mapping out the process flows and understanding the "as-is" journey, we identified problem areas and pain points that informed our design decisions
HIGHLIGHTS & PAIN POINTS
If quality of the documents are not good, it makes my job stressful in reading the data leading to inaccurate cases
H
Hansa Gupta
Case Processor, 28 Years
Going through each and every document individually is a time taking process
Sri vidhya
Case Processor, 29 Years
Our case processing happens on multiple softwares. Procuremnet happens in one, processing happens in one and submission on another
A
Akhilesh
Case Processor, 28 Years
Brainstorming & IDEATION
Visual Strategy & DESIGN LIBRARY
TESTing and ITERATION
DESIGN PROCESS
Snippets from the User Interview
Identifying Strategic Scope
To simplify the data entry process and enhance its efficiency, we brainstormed with the Tech Team and discussed how we can leverage AI and machine learning to automate and streamline the process. Here are some potential strategies:
Automated Data Extraction:
Utilize AI-powered data extraction tools to automatically extract relevant information from source documents, reducing manual effort and increasing accuracy.
Smart Data Classification
classify and categorize data, ensuring that it is accurately and consistently labeled, and reducing the need for manual intervention.
Document Standardisation
To make it easier for users to understand and navigate the forms, reducing confusion and errors.
Predictive Form Completion
Leverage machine learning models to predict and complete missing data, reducing the need for manual data entry and improving data accuracy.
Before the finalhandoff, we tested certain critical functionalities of the product with the actual users in a moderated environment. The goal of the test was to check the efficiency and effectiveness of the product.
A sample of 10 people were identified and the prototype was tested with them.Several interesting observations came out of the test which were then prioritised with the business as per their complexity.
Although the accuracy of the system was not that great due to lack of data, but the overall effciency of the product improved considerably. We created a detailed report and submitted it to the product team
Highlights System
Extracted Data
Ability to filter system
highlighted data
Filters extracted & required fields
Switch Between Original Document & Structured Document
Case Processing Screen
Multi Layered Data
organisation for optimum decluttering
Shows density of data in the
document
Highlights key information as
required by users