Statistical Guidelines for Contributors to Medical Journals

Most papers published in medical journals contain analyses that have been carried out without any help from a statistician. Although nearly all medical researchers have some acquaintance with basic statistics, there is no easy way for them to acquire insight into important statistical concepts and principles. There is also little help available about how to design, analyse, and write up a whole project. Partly for these reasons much that is published in medical journals is statistically poor or even wrong.1 A high level of statistical errors has been noted in several reviews of journal articles and has caused much concern. Few journals offer even rudimentary statistical advice to contributors. It has been suggested1 2 that comprehensive statistical guidelines could help by making medical researchers more aware of important statistical principles, and by indicating what information ought to be supplied in a paper. We present below an attempt to do this. Deciding what to include in the guidelines, how much detail to give, and how to deal with topics where there is no consensus has been problematic. These guidelines should thus be seen as one view of what is important, rather than as a definitive docu? ment. We have not set out to provide a set of rules but rather to give general information and advice about important aspects of statistical design, analysis, and presentation. Those specific recommendations that we have made are mostly strong advice against certain practices. Some familiarity with statistical methods and ideas is assumed, since some knowledge of statistics is necessary before carrying out statistical analyses. For those with only a limited acquaintance with statistics, the guidelines should show that the subject is very much wider than mere significance testing and illustrate how important correct interpretation is. The lack of precise recommendations indicates that good statistical analysis re? quires common sense and judgment, as well as a repertoire of formal techniques, so that there is an art in statistics as well as in medicine. We hope that the guidelines present an uncontro versial view of the most frequently used and accepted statistical procedures. We have deliberately limited the scope of the guidelines to cover the more common statistical procedures. Readers may find that a relevant section presents information or advice that is unfamiliar or is not understood. In such circumstances, although almost all of the topics covered may be found in the more comprehensive medical statistics text? books,3 4 we strongly recommend that they should seek the advice of a statistician. The absence from the guidelines of specific references is intentional : it is better to get expert personal advice if further insight is needed. Moreover, because mistakes in design cannot later be rectified, professional advice should first be obtained when planning a research project rather than when analysing the data.