Could data mining medical literature uncover side effects before they cause serious harm to patients? A new study maintains the effort could effectively complement existing methods, such as combing through the FDA’s Adverse Events Reporting database, because prior research suggests that up to 98 percent of searches are irrelevant to side effects and may skew results toward false positive links.

To prove the point, two researchers from the Rank think tank developed an algorithm to sift through the PubMed literature and searched for mentions of least one of 38 drugs and 55 side effects. From there, they determined the relevance of the articles and forecast expected rates of adverse events. They would up analyzing 9,133 articles published between 1949 and September 2009, plus others that incorporated side effects from classes of drugs as opposed to individual compounds.

“We aimed to prototype a process for collecting and analyzing relevant literature while minimizing false positive and false negative drug-AE associations,” the researchers wrote in the Journal of the American Medical Informatics Association (here is the abstract). “…we evaluated whether the literature-mining process might have detected drug-AE associations that were the subjects of FDA warnings. We used the entire literature in some analyses and simulated prospective analyses by restricting the data to literature available prior to the warnings.”

One example cited was Merck’s Vioxx; the researchers wrote they found evidence of heart disease caused by painkiller, which was withdrawn in 2004, due to links to heart atttacks and strokes, as early as 2001 by analyzing literature that was published by 2002. At the end of the day, they believe their “literature-based method” discovered real associations between drugs and adverse events with greater than 70 percent sensitivity and 40 percent positive predictive value. They also maintain they were able to detect “numerous associations” prior to FDA warnings, “suggesting that literature mining did not simply provide a lagging indicator” of widely known links between drugs and side effects.

“Mining FDA’s AERS data has been fruitful but there have clearly been (as expected with any method) false positives and false negatives,” Rand’s Siddhartha Dalal writes us. “We hypothesized that analyses using a novel source, in which the data had been filtered by the editorial and peer review process, would provide useful information that would complement results found when analyzing AERS. We think that our results supported this hypothesis, and hope to collaborate with researchers using AERS in the future.”

Hat tip to InformationWeek