Unreliable: Bias, Fraud, and the Reproducibility Crisis in Biomedical Research

Most biomedical research findings can’t be reliably replicated, according to this startling debut report. Szabo, a pharmacology professor at the University of Fribourg, Switzerland, contends that the problem stems in part from such honest mistakes as mislabeling cultures or variations between labs in the composition of reagents. (Even lab-grade water treatment machines can’t fully erase differences in the levels of microorganisms and synthetic estrogens commonly found in tap water.) Another factor is the questionable massaging of data, Szabo posits, discussing how the “borderline acceptable” practice of excluding statistical outliers can make the “difference between an effect that can be declared statistically significant versus statistically nonsignificant.” Explaining how outright fraud has driven other irreproducible results, Szabo describes how in the 1970s, a Memorial Sloan-Kettering dermatologist claimed to have figured out how to transplant skin from genetically unrelated organisms in an experiment on black and white mice. He was fired after the institution discovered he simply colored the white mice’s skin with a black pen. Szabo’s conclusion, based on numerous anonymous surveys, that “one in five people working in biomedical science have engaged or are engaging regularly in fraud” is jaw-dropping, and his sensible solutions include devoting more funding to validating published findings and imposing criminal penalties for fraudsters. It's a troubling wake-up call for scientific researchers. (Mar.)