Identifying Bias in Academic Journal Articles

Bias is ever-present. While we are generally aware of the basic forms that bias takes, often bias arses unconsciously, which makes it more difficult to detect. All studies have bias -- these biases begin with our own enculturation. The lenses through which we examine every facet of the world have been shaped by our experiences, and thus there is the bias of our own culture, of how we studied science, and of the preferred methodologies that underpins every decision made.

Thus, no study can be said to be free from bias, it is more important to recognize where there is bias, and how such bias influences the study design, our interpretation of the findings and ultimately our vision of the objective truth. Peppard et al. (2013) studied the increased prevalence of sleep-disordered breathing in adults. In this study, the authors examined time series data from an ongoing community-based study in Wisconsin to determine if there is an increase in sleep-disordered breathing in adults, and to seek conclusions about causality.

The authors looked at current rates, which were modelled, versus historical rates, all using data from the survey. Sleep-disordered breathing includes factors like sleep apnea and hypopnea, known to have negative health consequences. The authors felt that things like hypertension are related to these disorders, based on evidence that hypertension and related weight issues are contributing factors to such sleep disorders.

The authors were working on the premise that an increase in obesity would also lead to an increase in sleep disordered breathing in adults over the time period of the study. There are a few potential sources of bias in this study. The researchers are from Wisconsin and the study was based on Wisconsin. Clearly, the availability of this data set going back to the 80s was appealing -- its availability basically facilitated the study.

But the population of the study, which began with over 1500 people, is not necessarily reflective of the overall population. Ethnically, it is not going to align with the U.S., and might not even with Wisconsin. The people in the study were chosen at random, but ultimately the demographics of the sample are not likely the demographics of the population of, say, the United. So there is selection bias -- the availability of data from this particular sample is what led this sample to be used in this study.

They did try to extrapolate this group of people to the entire country, which is not reasonable. Another source of bias is that the patients are getting older each year. When the time series study began, the participants were between 30-70 years of age. They were tested for a baseline level, and then re-tested each 4 years thereafter. So the same people were in the study at every interval.

The problem with this is that health outcomes tend to decline as we age, so the rates of obesity and any related health issues would be expected to increase. This is going to skew the results -- if a random group of, say, 40-year-olds was used each time, this would better track the changes in the population because it would eliminate the impact of aging from the findings. ' The authors also used BMI as a proxy for weight, not really fully discounting the different flaws that BMI.