News & Events

Tippie-Led UI Team Designs Award-Winning Model to Spot Unknown Drug Side Effects

A research project at The University of Iowa could help health care professionals protect lives by spotting previously unknown harmful side effects of prescription drugs.

Nick Street, an associate professor of management sciences in the Tippie College of Business, said the project team designed a prediction model that suggests harmful drug effects and helps clinicians design further tests to determine if the effect is real.

“Our model doesn’t prove the side effects because that can only come from a clinical trial, but we can find correlations that help set up those trials,” Street said.

Street was part of the development team with Lian Duan and Mohammad Khoshneshin, Tippie doctoral students, and Si-Chi Chin, a doctoral student in information science. The team’s model recently won first prize in a challenge in the Observational Medical Outcomes Partnership (OMOP) Cup, a competition in which participants develop a groundbreaking approach to ensure the safety of prescription drugs.

The contest was a simulation in which the team was given a database of drugs and patient conditions that resembled information taken from medical records, insurance claims, and other health care documents. The team members then had to find as many potential negative side affects as they could before the April 1 competition deadline. (Although the competition lasted for six months, the UI team had only two months because they didn’t learn about it until January.) The contest also gave credit for finding the true correlations sooner, which involved spotting the side effects more efficiently by reviewing fewer records.

The team won $5,000 for winning the challenge by finding the potentially dangerous drug/condition pairs from 22 million such pairs in the database. The UI researchers found the interactions using what is known as data mining to arrange and analyze the information in the database to find patterns.

The team designed and developed its own software, which combined a statistical model with two specialized correlation techniques.

“We used these tools to see if the relationships were real and if our conclusions would hold up to future data, to determine if there really was a drug interaction, or if it was just background noise in the data,” Street said.

One particular challenge involved conditions that took effect on the same day that the patient started taking a drug, so the team had to determine whether the condition was a reaction to the drug or to the ailment being treated.

Street said the team’s design could be the basis for developing a tool that would look for real-life drug side effects in the future. Now that the health care profession has decades worth of data in electronic form about patients and their interactions with prescription drugs, projects like the UI’s could mine those massive databases and save lives by spotting previously unseen harmful, and sometimes fatal, side effects.

Return to top of page