Study design synopsis : bias can cast a dark shadow over studies

Abstract

The study of free-living populations is important to generate knowledge related to the epidemiology of disease and other health outcomes. These studies are unable to provide the same level of control as is possible in laboratory studies and thus are susceptible to certain errors. The primary categories of study errors are random and systematic. Random errors cause imprecision and can be quantified using statistical methods including the calculation of confidence intervals. Systematic errors cause bias, which is typically difficult to quantify within the context of an individual study. The three main categories of systematic errors are selection, information, and confounding bias. Selection bias occurs when enrolled animals are not representative of the target population of interest in respect to characteristics important to the primary study objective. Information bias occurs when data collected from enrolled animals deviates from the true value. Information bias is most damaging when errors vary among comparison groups. Both selection and information bias are prevented through the application of good study design procedures. Researchers should select study animals after careful consideration of the primary study objective and desired target population. Investigators can reduce information bias through standardised data collection procedures and the use of blinding. Confounding bias occurs when the measured association between a predictor and an outcome ignores the influential effect of an additional variable. Confounding is common and analysts must implement the appropriate statistical adjustments to reduce the associated bias. All studies will have some errors and biased data with high precision are the most damaging to the validity of study conclusions. Authors can facilitate the critical evaluation of their research by providing text related to the limitations and potential sources of bias within the discussion section of their manuscripts.