Sorry, you need to enable JavaScript to visit this website.
Clinical Data Management

Can you train a model to identify patterns in clinical data to reconcile data faster and speed up study conduct while maintaining, and in some cases, improving data quality?

Completed on Sep 13, 2019
Doctors having a conversation
Overview

Our Goal

Use machine learning technologies to learn from historical data related to queries and their associated patterns, along with the clinical data from those studies. 

The Challenge

Clinical data managers must merge data across several reports by reviewing the study’s audit trail manually to determine if a data validation error (queries) must be raised or not. Historically, the industry has been limited to manual data review processes to validate data from clinical trials, which are very time-consuming and laborious. With a large training dataset available from several historical studies as reference, the ideal machine learning solution should be able to automate the identification of data discrepancies with accuracy to optimize operational efficiency and reduce cycle time for database locks.

Results

Multiple selected companies participated in this challenge and each applied innovative methods ranging from rules-based techniques, to clustering algorithms, to deep learning with neural networks. The winning solution leveraged a deep learning model that enabled human-in-the-loop feedback mechanism, prediction interpretability as well as empowering data managers to focus on more difficult data reconciliation challenges, while the machine assists in predicting repeatable cross panel data reconciliation patterns.

About the Breakthrough Change Accelerator 

Pfizer’s Breakthrough Change Accelerator is a competitive incubation sandbox created to power rapid and robust machine learning in a secure environment. We invite innovators to collaborate on challenges using large volumes of relevant datasets to create novel technology solutions that speed up development of medicines to patients.