Журнал лекарственного метаболизма и токсикологии
Открытый доступ

Абстрактный

Reproducing Tests: Recommendations for Designs, Statistical Analysis, and Execution of Toxicological Investigations

Giacomini Zur

The challenge of reproducing tests has received a lot of attention. Problems with replication arise for a range of causes ranging from experimental design to laboratory mistakes to inadequate statistical analysis. Here, we go through a number of recommendations for the design, statistical analysis, and execution of toxicological investigations. In general, hypothesisdriven trials with sufficient sample sizes, randomization, and blind data collecting methods can increase replication. Both publicly and privately within the scientific community, science is going through a kind of crisis of faith. Some high-profile cases of fraud have garnered public attention, including the debunked Stimulus-Triggered Acquisition of Pluripotency (STAP) method for STEM cells. The practicing scientist sees retractions of articles much too frequently because of shady data and methods. Although it is regrettable, grant review panels, reviewers, editors, and observant readers at least seem to be able to spot fraud. More harmful are the claims that have surfaced in recent years about publications across several fields having poor replication records, with no proof of fraud. According to a recent article in Science, less than half of psychological studies could be repeated. Although it seems reasonable that the "soft sciences" should not replicate well, the biological sciences also perform badly when replication pressure is applied to them. According to reports from the pharmaceutical industry, failure rates for attempts to reproduce published studies to progress medication development are considerably over 50%.