Towards a Responsible Use of Statistics in Biomedical and Epidemiological Research

Abstract

Although the application of numerical methods to the study of typical research problems in the medical and health sciences dates back to the 1830s with the contributions of the French physician Pierre A. Louis, the formalization of mathematical statistics as an analytical tool is historically associated with the contributions of Karl Pearson, Ronald Fisher, and Austin Bradford Hill, who laid the foundations for what we know today as evidence-based medicine, beginning in the mid-20th century. Since then, the presence of some analytical tool from inferential statistics has been considered an indispensable and irreplaceable mark of quality in research. Without denying the instrumental value of statistical resources, the purpose of this essay is to argue that their irresponsible and superficial application has contributed to fostering an attitude of intellectual laziness among medical researchers and to the establishment of rigid norms that are contrary to the rigor and quality of scientific thought. The topic has universal interest and relevance, but the arguments in the article refer mainly to the author's experience in national scenarios, and refer in particular to two analytical resources whose irresponsible use is particularly frequent: statistical significance tests and predictive models.

Keywords: observational studies; randomized controlled clinical trials; sample size; confounding variables; linear models; prediction vs. explanation; hypothesis validation.