AI Can Help Identify Potential Adverse Effects from Drug-Drug Interactions

A newly-developed machine learning system may help act as an alert system for potential adverse effects from drug-drug interactions.

Patients who take multiple medications are at an increased risk of experiencing negative effects from drug combinations, particularly when they are prescribed drugs and then take additional OTC medications on their own. However, new machine learning system may help identify potential negative adverse effects from drug-drug interactions, according to a study published in IEEE Journal of Biomedical and Health Informatics.

For the study, the authors designed an algorithm that analyzes data on drug-drug interactions cited in reports from the FDA and other organizations. Using this information, the algorithm was designed to act as an alert system that could warn patients when a drug combination may have potentially dangerous adverse effects.

“Let’s say I’m taking a popular over-the-counter pain reliever and then I’m put on blood pressure medicine, and these medications have an interaction with each other that, in turn, affects my liver,” study author Soundar Kumara, Allen E. Pearce and Allen M. Pearce professor of industrial engineering, Penn State, said in a statement. “Essentially, what we have done, in this study, is to collect all of the data on all the diseases related to the liver and see what drugs interact with each other to affect the liver.”

With artificial intelligence advancements in medicine, a machine learning program that could catch potential drug-drug interactions would be a viable method to improve patient safety. However, the authors noted that “the implementation of DDI alerting system remains a challenge as users are expiring alert overload, which causes alert fatigue.” Alert fatigue occurs due to the high number of possible adverse interactions, which may inadvertently cause providers and patients to ignore alerts.

To overcome this barrier, the authors identified only drug-drug interactions for the algorithm that would be considered high priority, such as those that are life-threatening as well as interactions that may lead to disability or hospitalization, and for which intervention is required.

Additionally, the alert system consisted of an autoencoder model, which is a type of artificial neural network. Because this model is designed to address how the human brain processes information, it is suited for semi-supervised algorithms that can use both labeled and unlabeled data.

The list included approximately 2891 drugs, or approximately 110,495 drug combinations, which showed a total of more than 1.7 million reports on serious health outcomes from drug-drug interactions.

However, identifying the high-priority interactions is just the first step, according to Kumara.

“The reactions are not independent of these chemicals interacting with each other ­— that’s the second level,” Kumara said. “The third level of this is the chemical-to-chemical interactions with the genomics data of the individual patient.”

Kumara noted that further development of the algorithm could lead to more personalized interaction alerts.


Liu N, Chen CB, Kumara S. Semi-supervised learning algorithm for identifying high-priority drug-drug interactions. IEEE Journal of Biomedical and Health Informatics. 2019. Doi: 10.1109/JBHI.2019.2932740

AI could offer warnings about serious side effects of drug-drug interactions [news release]. Penn State University’s website. Accessed October 15, 2019