Fingers, hands or patients? The concept of independent observations

In hand surgery trials, it is often possible to take several measurements from the same patient, because many disorders here affect bilateral or multiple structures, such as the hand itself, the finger joints or the tendons. Most conventional statistical analyses that take place on the level of hands, digit rays or joints rather than patients violate the assumption that observations should be independent. Furthermore, ignoring the multiplicity of data inflates sample size and thus may lead to spurious significance. This article describes three options to deal with such problems. First, the analysis can simply be restricted to only one measurement per patient. Second, a self-controlled design may be advantageous for conditions that usually have a bilateral pattern. Third, complex statistical modelling (involving generalized estimating equations) can be used to analyse all available measurements with adjustment for data dependency.