Observational study biases. Many environmental risk factors that are the subject of analysis such as smoking, head injuries, electrical shock and chemical exposure are investigated because they are known to cause other diseases. It is therefore unethical to investigate the effect of these kinds of exposures on an outcome using a randomised controlled trial to experimentally demonstrate causality. In these cases, observational research is undertaken where the proportions of people who have encountered an exposure with and without developing the disease are compared. While this is ethically necessary, there are several methodological pitfalls that make determining causality more difficult. Observational studies have the potential to suffer from unmeasured confounder bias due to the unrandomized nature of the design. Unmeasured factors that are associated with the factors under investigation in the study may be driving the association, rather than the measured factor (Greenland and Neutra, 1980). Multivariate analysis, where outcomes can be adjusted for the presence of confounding is common practice to mitigate some of the effects of confounders (Xxxxxx, Xxxxxxx, & Xxxxx, 2012). Adjusting for all possible confounders is almost impossible, and confounders could be important at stages of life much earlier than when the disease develops (Xxxxxx et al., 2005). If an exposure and an outcome both affect how likely someone is to be sampled for a study this can lead to spurious causal associations driven by collider bias (Xxxx et al., 2010). For example, moderate exercise may slow disease course in people with ALS (McCrate and Xxxxxx, 2008). Additionally, people who are enrolled in clinical studies tend to have a slower disease course to have enough time to be included in the study. Therefore, sample selection bias could produce a correlation between physical activity and ALS where there isn’t one in the ALS population more generally – so population level data are preferred. A recent, large population-based study using data from the Netherlands, Ireland and Italy found a small but significant increased risk of ALS with different levels of physical activity, although the project was population based, the percentage of responders, and their clinical phenotype compared to non-responders, was not reported (Xxxxxx et al., 2018). Retrospective observational studies that reply on questionnaire data may suffer from self-reporting biases (Xxxxxxxxxx, 2016). One such bias is recall bias, which causes peo...
