The importance of a study is often inappropriately defined by the P value, the authors write, and this problem is highlighted by the use of “trend”—a term for which there is no definition—to refer to statistically non-significant results. To make a case in point, they set out to characterize the degree of overreliance on P values in high-impact oncology literature.
Because there is no definition of a trend toward statistical significance, describing “almost significant” results as a trend introduces substantial subjectivity and the opportunity for biased reporting language that could mislead a reader, the authors argue. They reviewed all original research articles published from November 2016 to October 2017 in the Journal of the National Cancer Institute (JNCI), Journal of Clinical Oncology (JCO), JAMA Oncology, and Lancet Oncology, and found that trend statements were frequently used to describe statistically non-significant results, commonly those with large P values and minimal supporting data. In addition, when P values approached statistical significance, promising clinical significance was often deemphasized in order to highlight the proximity of the P value to .05—demonstrating an overemphasis on P values.
In closing, the authors point out some central tenets of the American Statistical Association’s statement outlining primary P value principles, and point out that trend statements violate these principles. We must deemphasize P values and shift our focus to the clinical relevance of the finding, the power of the study to address a clinically meaningful difference, and the appropriateness of the study design, they conclude, and the oncology research community—in particular, leading oncology research journals—should take the lead.