Abstract

Good practice in research involves considering diverse sources of biases when designing a study for later validation of results. If they are recognized beforehand, it is possible to minimize or avoid them. Selection biases may originate at the time of enrolling the subjects of study; making it necessary to clearly state the selection criteria of the exposed and non-exposed individuals. If people get lost from the original sample, bias may be introduced by the consequences of reducing the sample. Biases of information could originate in loss of evidence at the moment of recording the data. The definition of follow-up protocols may also help to keep registers of all variables, so information will not be missed from the individuals under study or from the observers who conduct the follow-up. It is necessary to apply the same protocols and instruments for measuring and evaluating the health outcomes in exposed and non-exposed individuals in order to avoid biases of miss-classification. Confusion biases can be avoided at the time of designing the study, with the inclusion of confounding variables from the onset. Matching by age and gender is strongly recommended and, finally, using adjustment techniques at the time of the data analysis.