The 144-year-old pharmaceutical team built a platform that uses AI to process data about adverse events.
By Clint Boulton
Assembling information about adverse reactions to drugs vexes pharmaceutical companies, which are required to produce reports on any incidents to federal regulators. The far-flung data arrives in different languages and formats, making it hard to manage.
Eli Lilly has cracked the code on this task, known in industry parlance as pharmacovigilance, using artificial intelligence (AI) software that automatically routes and processes information about adverse events.
“Our primary business driver is to increase productivity and reduce the cost of processing adverse events, while maintaining a high standard of quality and compliance,” says Tim Coleman, vice president and information officer of Eli Lilly’s medicines development unit.
Pharmacovigilance is a labyrinthine process. Safety departments within global pharmaceutical purveyors verify and review tens of thousands of cases before transferring them to a reporting database. The cases are then sent to governing agencies within a certain timeframe.
Converting each case from the format in which it is reported into one that can be easily reviewed by regulators is a complex, laborious process. Moreover, the goalposts frequently move in accordance with ever-evolving regulations. Assessing the impact and efficacy of these efforts, from informatics to reporting, is also challenging.
Convening a cross-functional team
With the number of adverse events growing between 10 percent to 15 percent each year, improving the intake process for cases is critical for Eli Lilly, which must reduce operational expenditures so that the company can shift funding to developing medicines and improving patient outcomes.
“Historically this [case management] process has been very manual, labor-intensive and costly,” says Coleman, who led the initiative.
In 2017, the company assembled a cross-functional team composed of case managers from its global patient safety organization, procurement, quality and IT. Subject matter experts and other business staff communicated to IT the pressing need to improve the cycle time of reporting adverse events.
The team built MosaicPV, an intelligent intake platform that automates the business rules used to triage and manage these reports. The software ingests both structured and unstructured case data about adverse events from global call centers, email systems and patient connection platforms into a data lake.
There natural language processing (NLP) software “reads” the adverse event information, as machine learning (ML) algorithms and rule-based models interpret the information and make predictions. At a high level, MosaicPV is comparable to how robotic process automation technologies help insurance and financial services companies process claims and other critical documents.
The platform has exceeded initial business case estimates based on improvement in cycle times in processing cases, as well as a reduction in time case processors spend examining case files, while adhering to data quality and compliance standards, says Coleman. “They are able to focus on higher value-add activities,” Coleman says, of staff who previously processed claims. The tool also garnered a IDG 2020 FutureEdge 50 award for use of emerging technology to transform business operations.
Think big and be agile
Coleman credits adopting an agile development methodology, something Eli Lilly has found success with in various other areas of it business, with the team’s ability to roll out MosaicPV. Using waterfall, the team wouldn’t have delivered improved business process and cost savings, according to Eli Lilly.
Following two releases in 2018, Coleman’s team rolled out a third version in August 2019, adding several new features, new user personas and new data sources. It also refined the machine learning and NLP models and migrated the system to Amazon Web Services, affording it the ability to rapidly scale while reducing Eli Lilly’s need to manage infrastructure. Release 4, scheduled to go live this year, will include new data analytics functionality and the ability to handle cases in other languages.
Rapid software development under the aegis of an agile, product-centric ethos was the company’s greatest hurdle. Employees undertook a significant amount of reskilling to learn how to work with and refine AI software, Coleman says. The team also had to learn to be “product owners” as they worked side-by-side for months, participating in agile scrums to build software for a highly regulated area. It required business and IT teams to break down the siloes in which they are accustomed to working, Coleman says, adding that PwC came onboard to assist and train the team.
“We challenged the team to leave their titles and respective roles at the door,” Coleman says, adding that the team adapted well and learned from the experience through three releases.
Keys to agile success
Coleman offers tips for fostering successful agile product teams.
Talent makes the difference. It’s hard to succeed without good people in the team, so Coleman made sure he included the right expertise to solve the problem, starting with staff who value teamwork and are aligned for a purpose to achieve a desired outcome.
Understand strengths and weakness of team. Knowing what you have and what you need regarding expertise requires examining both internal and external resources to mitigate risks of failure.
Don’t be afraid to think big. “Thinking out of the box is the key to success for projects and programs,” Coleman says.