ContraIndicator: A Natural Language Processing-Based Approach to Potential Drug-Drug Interaction Detection in...

The intricate nature of pediatric patients' physiology and the frequent administration of multiple medications in pediatric intensive care units expose hospitalized patients to potential drug-drug interactions. Therefore, this study aimed to examine and characterize clinically relevant drug-drug interactions in pediatric patients admitted to intensive care units, leveraging artificial intelligence to enhance their identification and management.

Prescription data from pediatric intensive care unit admissions were analyzed using ContraIndicator, a natural language processing-based framework developed for the identification of potential drug-drug interactions. Medication information extracted from prescription records was evaluated against the DDInter database, and its clinical relevance was determined using the severity ratings of the database (major, moderate, and minor). Using univariate and multivariate logistic regression, the associations between severity and age, as well as between severity and the number of prescribed medications, were investigated.

The examination of 8010 prescriptions for 899 hospitalized patients identified 3884 potential drug-drug interactions. The most frequent interacting drug pairs were midazolam + fluconazole, fluconazole + fentanyl, fentanyl + dexamethasone, and vancomycin + piperacillin. Among the 49.6% (446) of pediatric patients who had at least one interaction (major, moderate, or minor), 38.8% (173) had a major potential drug-drug interaction. In this investigation, a higher number of prescribed drugs was significantly linked with a higher incidence of potential drug-drug interactions (odds ratio: 12.79; 95% confidence interval: 8.26-20.6; p < 0.001).

The study suggests that potential drug-drug interactions are frequent in pediatric intensive care units. The majority of these interactions were moderate, followed by major severity. There is a statistically significant association between the quantity of drugs provided and the prevalence of potential drug-drug interactions. The findings of this study could aid in planning and executing future studies, as well as in monitoring and preventing potential drug-drug interactions in patients receiving intensive care. This study introduces ContraIndicator, a graphical user interface that presents empirical findings and promotes awareness regarding potential drug-drug interactions.