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.

Study design synopsis: Observational study design in equine research

“…Depending on the study design, this will alter the approach taken to case recruitment but all require thorough deliberation to reduce bias. For further discussion on bias, refer to Fosgate 18 …”

“…Common pitfalls to consider include obtaining a truly representative sample of the intended target population, if this is not achieved, it can contribute to selection bias and invalid estimates 18 . The methods of sample selection, which should be a random sample, should be clearly defined and the resulting study sample available for analysis will determine how valid the results are.…”

“…Common pitfalls to consider include obtaining a truly representative sample of the intended target population, if this is not achieved, it can contribute to selection bias and invalid estimates. 18 The methods of sample selection, which should be a random sample, should be clearly defined and the resulting study sample available for analysis will determine how valid the results are. The final sample included in the study may be affected by the response rates (eg respondents of a survey) or the sampling frames used or the number of animals that complete data are available for (Figure 2).…”

“…Hospital or practice-based case-control studies are a common approach, but care must be taken when selecting hospital controls because animals presented to hospitals have other conditions which may be related to various other factors, for example, age, breed and use, leading to Berkson's bias. 4,18 Case-control matching on the basis of one or more variables is also possible in circumstances where a confounder is associated with disease and exposure. However, this should be carefully…”

“…For further discussion on bias, refer to Fosgate. 18 Observational epidemiological studies will continue to have a substantial contribution to clinical practice and to equine research and evidence base. To maximise the value of these studies, it is of critical importance that the study design is accurate, valid and appropriately reported in the literature.…”