When is a null finding in register-based epidemiology convincing?

Methods have been suggested to evaluate the influence of unmeasured confounding [11] but will not be further discussed in this commentary because unmeasured confounding will in most cases bias the association reported away from the null, for example, confounding by indication where medications appear to cause outcomes they are meant to prevent or healthy user bias where the healthiest continue treatment [12] and will rarely explain null findings in register-based studies. In observational studies, the situation is not similar even though the STROBE statement (STrengthening the Reporting of OBservational studies in Epidemiology) recommends sample size calculation when planning a new study [13]. Information on expected number of exposed persons and number of events is often available, but information on relevant effect sizes and effects of adjustment of confounding variables on the risk estimates is often difficult to include in a formal sample size calculation. According to epidemiological methodology, nondifferential misclassification will generally underestimate the association between exposure and outcome at least for relative risk measures which are the most commonly used effect measures. The CRAN package episensr was used to estimate the observed relative risk under different values of and rubella vaccination and autism, N Engl J Med, Vol. 347, 2002, 1477-1482 4 M. Melbye, J. Wohlfahrt, J.H. Olsen, M. Frisch, T. Westergaard, K. Helweg-Larsen, Induced abortion and the risk of breast cancer, N Engl J Med, Vol. 336, 1997, 81-85 5 G. Edgren, M. Reilly, H. Hjalgrim, T.N. Tran, K. Rostgaard, J. Adami, Donation frequency, iron loss, and risk...