Making sense of the huge amount of data generated during clinical trials
The client is one of the largest pharmaceutical companies in the world. It spends hundreds of millions of dollars in clinical trials to come up with newer drugs. Clinical trials are long, expensive and their timely success is imperative for the long-term growth of a life sciences organization. For every successful molecule that passes through the clinical trials and reaches the market, there are 1000’s of molecules that do not pass through the entire cycle. Their trials are terminated during either drug discovery or development phase.
However, during this entire cycle enormous amount of data is generated at each stage. Our client wanted to leverage technology and cognitive computing to analyze this data, help generate insights regarding rationale for failed clinical studies, and leverage that to monitor studies going forward. This will allow them to predict failures early, take corrective action or terminate the studies and save costs.
Clinical studies and trials include large volumes of data, which comes in both structured and unstructured format, has various semantic and contextual nuances for driving analytics and deriving any meaningful insights. This makes it challenging to read, understand and interpret the data. Moreover, to learn something from that data and apply it on future studies requires a framework that learns from past data and applies it on future projects as well. read more...
A comprehensive cognitive computing framework
Nagarro developed a comprehensive big data and machine learning based cognitive computing framework which leveraged the client data and guided a learning system to extract similar information in future from documents and generate insights.
Scoping the problem
Given the nuanced and ambiguous nature of this engagement, we ran a series of natural language processing and machine learning based functional and technical experiments (each ranging from a few hours to few days) and documented those results to understand opportunities, feasibility and limitations of what could be accomplished.
Our problem solving approach
- We leveraged ‘signal extraction framework’ to mine the right insights determining success or failure of clinical trial, from a large unstructured set of data.
- Using our functional expertise, we annotated the signals and developed a self-learning system which could conduct a ‘sentiment analysis’ and help identify the polarity strength of each signal.
- We employed a disease area specific clinical trial success framework to develop detailed compound profiles and interpret the signal from.
The computing framework
Using the insights coming from the above exercise, we developed a dashboard wherein our client could look at the summary profile of each candidate being experimented in the clinical study or any competitor compounds, and based on that answer a series of question such as:
- Correlations observed on the data set and what can be learned from it for future studies on same compound
- Could the client have terminated the study earlier and at what point
- What are the successful predictors for a project of this nature
- What could the client have learned from the competition during the course of the drug discovery project
Channelizing time and resources to the right initiatives
- The client can now leverage disease and treatment pathway contextual insights to identify early on the compounds most likely to be successful.
- The state-of-the-art solution will potentially save millions of dollars every year across various research efforts.
- The solution enables the client to save critical time and resources which can be channelized to the right research initiatives to help develop treatments.read less...