To start with, a brief comment about why statistics is important in science. As Richard Feynman so eloquently states it in this video, science relies on empirical support:
If it disagrees with experiment, it is wrong. In that simple statement is the key to science. It doesn’t make a difference how beautiful your guess is. It doesn’t make a difference how smart you are who made the guess, or what his name is. If it disagrees with experiment, it is wrong.
Furthermore, empirical support is usually uncertain:
It is scientific only to say what is more likely and less likely.
Statistics is in medical research the main science for developing and confirming hypotheses, for quantifying the benefits and risks of exposure to specific treatments or hazardous agents with quantification of the unavoidable estimation uncertainty, and for finding answers to many other questions.
Uncertainty is fundamental for medical research as this typically is performed on limited groups of subjects, for example 32 healthy volunteers included in a randomized trial, a cohort of 587 asbestos workers followed 18 years, or some 83761 patients having been treated with a specific drug and registered in a patient register.
However, how large the studied sample is, the unavoidable question is still whether the observed outcomes are relevant for others than those studied. Are the results generalizable to other, future patients? Or do the results only reflect sampling variation, or bias?” This is a classic “sample-to-population generalization” issue that is in the centre of statistical inference. The uncertainty is usually presented in terms of probabilities (p-values) for a specific hypotheses, or confidence intervals describing the estimation uncertainty of studied effects.
Given the importance of statistical science, one could expect medical researchers to be experienced and good users of statistical methods, but the unfortunate fact (1) is that wrong techniques often are used, wilfully or in ignorance, and that many published results are misleading. Gore et al. (2) showed, for example, 1976 that of 62 articles published in the British Medical Journal, 32 (52%) had statistical errors and 18 (29%) were seriously flawed. Similar errors appear today, but the problem is now greater, because advanced analyses with complicated calculations can easily be made, without the slightest understanding, using modern statistical software. Totally erroneous results, that may appear convincing, have good chances of being be published.
1. Altman DG. The scandal of poor medical research. Br Med J 1994;308:283.
2. Gore SM, Jones IG, Rytter EC. Misuse of statistical methods: critical assessment of articles in BMJ from January to March 1976. Br Med J. 1977;1:85-87.