We would like to finalize our work with some take home messages. “What it is” is surely important but “What it ought to be”, is perhaps more important in Understanding Evidence Based Rheumatology, as our title reads.

We first have to realize that many of our diseases are still constructs and as such they need continuous clinical and scientific scrutiny. In this context the rather slow materialization of “The Human Phenome Project” should gain speed as we consider it at least as important as the long time finished The Human Genome project. On hindsight, we reflect that it would have been much more scientifically fruitful had the ubiquitous genetic association studies in every conceivable construct – disease we read during the last several decades had paid more attention to the phenotypes of the constructs they were studying. This wish of more emphasis on the phenotype applies both to the biomedical and the bio-psychosocial models of approach to better understand our diseases.

In the Preface we have jokingly said that the authors to the statistics chapter (see chapter “A review of statistical approaches for the analysis of data in rheumatology”) were instructed to avoid the integral sign in their manuscript, to be better understood. On the other hand perhaps this is not what it ought to be. As we have tried to underline, a sound grasp of methodology, which in turn requires at least a familiarity with basic arithmetic and probability theory is a must for being a better clinician and/or a better investigator. A good clinician and/or investigator should surely also be able to recognize, appreciate and avoid many of the cerebral forms of ethical transgressions (see chapter “Ethical issues in study design and reporting”) and this is only possible with a sound insight into scientific methodology. The investigators should learn to invite the help of a statistician right from the start of their work, seeking advice not only on best methodology to prove themselves right but more importantly how to falsify their hypotheses and on many occasions even come up with new hypotheses. As a corollary, statisticians should no longer be content with their current role of the Wizard of Oz and should be right at the bed side or bench of any investigation before the investigator goes astray, both inadvertently and, on occasion, intentionally. In brief we should make research methodology and statistical thinking integral parts of teaching, practice and research. The more and more widely available software packages for statistical calculations are both marvellous and to be cursed. They, very often, do not help us to understand what we do. It is to be noted that the p value tells us only how surprising the obtained result is if the null hypothesis were true. Taking a randomized controlled trial as an example, assume we have reason to believe a new drug causes a cure 9 times more than the old drug. We design a controlled study and find out a p value of 0.001 related to the difference in outcome between the two arms of our trial. This p value is nothing more than the probability of observing a difference in the disease outcome between the two arms of the study at least as big as what was actually observed, if all what we observed was due to chance. It does not however tell us that our drug will work 9 times better. The investigator actually is surely more interested in the probability of the stated hypothesis being true (9:1 odds in our example). Confidence intervals and perhaps even more the Bayesian methodology are of more helpful here (see chapter “A review of statistical approaches for the analysis of data in rheumatology”). Specifically we should make every effort to be more knowledgeable about what the Bayesian approach with its conditional probabilities is, how it can be very helpful to us both in decision making and setting up and interpreting drug studies.