JAMIA: Data mining could improve drug safety
Medical data mining proves helpful for predicting adverse drug events, according to a study in the May edition of Journal of the American Medical Informatics Association.
Prescription drugs can be associated with adverse effects that are unrecognized despite evidence in the medical literature, as shown by rofecoxib's (Vioxx, Merck) late recall in 2004, wrote researchers Kanaka D. Shetty, MD, and Siddhartha Dalal, PhD, MBA, from Rand, in Santa Monica, Calif.
For the study, the authors retrieved citations published in MEDLINE, the National Library of Medicine’s (NLM) bibliographic database, between 1949 and September 2009 if they mentioned one of 38 drugs and one of 55 adverse events. A statistical document classifier using Medical Subject Heading—which comprises NLM’s controlled vocabulary for indexing articles index terms—was constructed to remove irrelevant articles, according to the researchers.
In testing, the statistical document classifier identified relevant articles with 81 percent sensitivity and 87 percent positive predictive value (PPV), according to the study abstract. Using data filtered by the statistical document classifier, base-case models showed 64.9 percent sensitivity and 42.4 percent PPV for detecting FDA warnings.
Base-case models discovered 54 percent of all detected FDA warnings using literature published before warnings, the researchers wrote. "For example, the rofecoxib–heart disease association was evident using literature published before 2002. Analyses incorporating literature mentioning adverse events common to the drug class of interest yielded 71.4 percent sensitivity and 40.7 percent PPV," Dalal and Shetty reported.
Results from large-scale literature retrieval and analysis compared favorably with and could complement current drug safety methods, the report concluded.
Prescription drugs can be associated with adverse effects that are unrecognized despite evidence in the medical literature, as shown by rofecoxib's (Vioxx, Merck) late recall in 2004, wrote researchers Kanaka D. Shetty, MD, and Siddhartha Dalal, PhD, MBA, from Rand, in Santa Monica, Calif.
For the study, the authors retrieved citations published in MEDLINE, the National Library of Medicine’s (NLM) bibliographic database, between 1949 and September 2009 if they mentioned one of 38 drugs and one of 55 adverse events. A statistical document classifier using Medical Subject Heading—which comprises NLM’s controlled vocabulary for indexing articles index terms—was constructed to remove irrelevant articles, according to the researchers.
In testing, the statistical document classifier identified relevant articles with 81 percent sensitivity and 87 percent positive predictive value (PPV), according to the study abstract. Using data filtered by the statistical document classifier, base-case models showed 64.9 percent sensitivity and 42.4 percent PPV for detecting FDA warnings.
Base-case models discovered 54 percent of all detected FDA warnings using literature published before warnings, the researchers wrote. "For example, the rofecoxib–heart disease association was evident using literature published before 2002. Analyses incorporating literature mentioning adverse events common to the drug class of interest yielded 71.4 percent sensitivity and 40.7 percent PPV," Dalal and Shetty reported.
Results from large-scale literature retrieval and analysis compared favorably with and could complement current drug safety methods, the report concluded.