Avoiding Bias in Observational Studies

T he randomized clinical trial is a well established and frequently used study design and is generally accepted as the gold standard (1). Nevertheless, many questions can only be answered with epidemiological observational studies, such as the influence of cigarette consumption on the development of lung cancer (2), the effects of sport, nutrition and overweight on cardiovascular diseases (3), or the effects of UV exposure on skin diseases (4). Whereas in experimental, randomized clinical trials, randomization is intended to lead to a comparable distribution of both known and unknown factors in the two groups to be compared, this is rarely possible in observational studies (see part 3 of this series). This can lead to systematic bias and thus to erroneous results. This article will describe how possible sources of error linked to the study design can be recognized in studies in which randomization is not possible for fundamental ethical reasons and how this can be considered in the planning and evaluation. The following sources of bias will be discussed: ❃ Selection mechanisms in recruitment of study participants (selection bias) ❃ Selective recall or inconsistent data collection (information bias), measurement errors ❃ Confounding, and ❃ Simpson's paradox and other errors. Once one is aware of the causes of biased results, these can either be excluded or reduced by intelligent study planning. In addition, these aspects must be properly considered during the analysis. Understanding these problems can help the critical reader to interpret study results. As this article is intended to give an introductory overview of this theme, results from a selective literature search will be presented.