The timely identification of previously unknown toxicities of cardiovascular drugs is an important, unsolved problem. In the United States, 20% of the 548 drugs introduced into the market between 1975 and 1999 were either withdrawn or acquired a new black box warning during the 25-year period following initial approval by the Food and Drug Administration. Adverse drug events (ADEs) are an important cause of morbidity and mortality in patients, yet 95% of ADEs are unreported, leading to delays in the detection of previously unknown ADEs and underestimation of the risk to known ADEs. It is known that Electronic Health Record (EHR) notes and lab results contain ADE information and biomedical natural language processing (NLP) provides automated tools that facilitate chart review and thus improve patient surveillance and post-marketing pharmacovigilance. Optimal use of anticoagulants requires accurate and timely detection of ADEs from EHRs. The objectives for this proposal are to develop intelligent NLP approaches to extract disease, medication, and structured ADE information from EHRs, and then evaluate extracted ADEs for detecting known ADE types as well as clinically unrecognized or novel ADEs whose pattern or effect have not been previously identified.
This project proposes innovative intelligent biomedical natural language process approaches to detect ADEs in patients' electronic health records (EHRs), an important step towards ADE surveillance and pharmacovigilance. The focus of this proposal will bon the national priority area of anticoagulants ADE detection, although the algorithms and systems we develop can be extended to other drug classes and diseases. It is anticipated that the data resources, algorithms, and the Anticoagulant Pharmacovigilance Toolkit (APVTK) developed will significantly enhance ADE detection, patient surveillance and post marketing pharmacovigilance.
Showing the most recent 10 out of 13 publications