On 2017 Jul 13, Thomas Littlejohns commented:

Thank you for your comment as well as the link to your informative article on selection and collider bias.

The overall aim of the current paper was to investigate the representativeness of UK Biobank participants on a number of characteristics and health outcomes. The cohort has been described as healthier than the general population (due to the low response rate and voluntary recruitment process), however this is the first time the evidence has been provided to confirm this. Whilst UK Biobank was not set up to be representative, providing the empirical evidence for this, as opposed to informing the research community anecdotally, will hopefully help clarify what questions the resource is well-designed to address. Primarily, exposure-outcome associations and not deriving prevalence and incidence rates that apply to the general population.

However, we are in complete agreement with Marcus Munafo and colleagues that researchers also need to be aware of the potential for biases to be introduced due to the ‘healthy volunteer’ nature of the cohort and included the following on pg.10 – “As with all observational studies, it is incumbent on researchers to acknowledge potential sources of bias on a case-by-case basis that might affect the generalisability of exposure-disease associations, such as residual confounding, reverse causation and self-selection bias”.

Essentially, the take home message is that UK Biobank is well designed for providing generalisable associations between exposures and outcomes. But as with all observational studies, researchers should take care when interpreting their findings and acknowledge the range of biases that could drive any associations, including selection bias if applicable.

On 2017 Jul 05, Marcus Munafò commented:

This is an important study which makes explicit the very low response rate within UK Biobank (~5%) and presents some of the factors associated with participation, then discusses their potential impact. However, in our opinion the view that this will make little or no difference to association analyses is overly optimistic. We have recently examined the potential impact of selection bias on results obtained from studies with low response rates [1]. We argue that, because selection can induce collider bias (which occurs when two variables independently influence a third variable, and that variable is conditioned upon), selection can lead to biased estimates of associations in some circumstances. The extent to which this will occur will depend on the particular association being explored, and the selection mechanisms operating. We suggest that researchers consider the potential for selection bias affecting their analyses, and carry out sensitivity analyses to assess robustness of their conclusions to selection bias.

Marcus R Munafò, Deborah A Lawlor, Kate Tilling

1. Munafò MR, Tilling K, Taylor AE, Evans DM, Davey Smith G. Collider Scope: How selection bias can induce spurious associations. bioRxiv, doi: https://doi.org/10.1101/079707

On 2017 Jul 05, Marcus Munafò commented:

This is an important study which makes explicit the very low response rate within UK Biobank (~5%) and presents some of the factors associated with participation, then discusses their potential impact. However, in our opinion the view that this will make little or no difference to association analyses is overly optimistic. We have recently examined the potential impact of selection bias on results obtained from studies with low response rates [1]. We argue that, because selection can induce collider bias (which occurs when two variables independently influence a third variable, and that variable is conditioned upon), selection can lead to biased estimates of associations in some circumstances. The extent to which this will occur will depend on the particular association being explored, and the selection mechanisms operating. We suggest that researchers consider the potential for selection bias affecting their analyses, and carry out sensitivity analyses to assess robustness of their conclusions to selection bias.

Marcus R Munafò, Deborah A Lawlor, Kate Tilling

1. Munafò MR, Tilling K, Taylor AE, Evans DM, Davey Smith G. Collider Scope: How selection bias can induce spurious associations. bioRxiv, doi: https://doi.org/10.1101/079707