AI Identifies Previously Unrecognized Harmful Drug-Drug Interactions
核心洞察
Researchers have developed an AI-driven method to detect potentially dangerous drug interactions missed in clinical trials, enhancing post-market drug safety monitoring.
The AI analyzed over 160,000 scientific articles and verified findings against 3.4 million patient records, identifying 111 drug-drug interactions linked to severe adverse reactions.
Confirmed interactions include methadone with ciprofloxacin causing respiratory depression (搜索) and ibrutinib with amiodarone increasing atrial fibrillation (搜索) risk, among others.
A new artificial intelligence (AI) approach developed by researchers at Vanderbilt University Medical Center (VUMC) is helping to identify previously unrecognized, harmful drug-drug interactions. By analyzing millions of real-world patient records and sifting through a vast amount of medical literature, the AI system can detect dangerous combinations that may have been missed during clinical trials.
The study, published in Clinical Pharmacology & Therapeutics, confirmed nine severe adverse drug interactions not previously logged in DrugBank, a widely used database for medication risk assessment. This includes the combination of methadone and ciprofloxacin potentially causing dangerous respiratory depression (搜索), and the co-administration of ibrutinib and amiodarone increasing the risk of atrial fibrillation (搜索).
Identifying Missed Interactions
According to You Chen, PhD, associate professor of Biomedical Informatics at VUMC and senior author of the study, many serious drug reactions are not apparent during pre-market testing due to the limited patient population and exclusion of vulnerable groups in clinical trials. "Our approach allows us to detect these problems by analyzing millions of real-world patient records," Chen stated.
The current system for after-market drug safety monitoring relies on voluntary reporting of suspected adverse reactions. The AI-driven approach offers a more proactive method to identify potential issues.
AI Methodology
The research team utilized natural language processing (NLP) to analyze 160,321 scientific articles published between 1962 and 2023. This process identified 111 potential drug-drug interactions linked to severe adverse drug reactions. These findings were then validated using medical records from over 3.4 million patients from VUMC and the National Institutes of Health’s All of Us research program.
Eugene Jeong, a doctoral student in the Department of Biomedical Informatics and lead author of the study, emphasized the importance of the validation step. "When we find the same drug interaction problems occurring in large patient databases, it gives us confidence that these are real effects that doctors need to know about," Jeong explained.
Specific Drug Combinations and Risks
The study focused on drugs processed by five liver enzymes, representing approximately 80% of all pharmaceuticals. The researchers found that when two such drugs are taken together, one drug may slow the breakdown of the other, increasing the risk of side effects. Confirmed risks included hallucinations (搜索) from combining tramadol and fluconazole, and kidney damage (搜索) from combining clarithromycin and voriconazole.
The researchers reported that patients taking these drug combinations concurrently had over a 90% greater likelihood of experiencing these dangerous side effects compared to those taking a single drug alone.
Clinical Implications
Chen noted that these findings could significantly improve patient safety, especially for older patients or those with complex conditions requiring multiple medications. "Better understanding of drug interactions helps them make safer choices," Chen concluded. This AI-driven approach represents a significant advancement in post-market drug safety surveillance, offering a more comprehensive and data-driven method for identifying and mitigating potential risks associated with drug-drug interactions.