Confounders, mediators, and colliders

 Confounding is a distortion of the estimated effect of an exposure on an outcome. The distortion is caused by a confounder that impacts both exposure and outcome.

Exposure <- Confounder -> Outcome

An observed exposure effect may then reflect the effect of the confounder rather than exposure. For example, when studying the effect of alcohol consumption on cardiovascular disease risk, prior health status, is a confounder if it affects both alcohol consumption and the risk of a cardiovascular event.

Confounding can be dealt with in two ways. The first is to perform an experiment with randomised assignment of subjects to exposure and a control group. The advantage with this strategy is that it prevents against confounding from all confounders, known as well as unknown. However, the disadvantage is that it cannot be used to investigate effects of harmful exposures as that would be unethical and in many cases practically impossible. Other disadvantages are that trials are expensive and logistically complicated, which make them unsuitable for studying long-term or infrequent events.

The second strategy is to perform an observational study and adjust the effect estimate for confounding effects. This strategy has several advantages as observational studies are simpler to perform, much cheaper than trials, and they can easily be used to study late and infrequent events, for example to estimate revision risks using an implant register. The disadvantage is that only known confounders can be adjusted for because the adjustment procedures require data on the confounders as matching criteria in matched analyses, treatment predictors in studies based on propensity scores, or as covariates in multivariable statistical models.

In practice, all confounders are rarely known, and attempts to identify them in the statistical analysis, for example using stepwise regression, are often attempted, but it is not a particularly good idea. To be able to interpret the outcome, the statistical models require assumptions regarding cause-effect relationships, and these assumptions must be based on clinical knowledge, not on exploratory findings that may differ from sample to sample. Consequently, the findings from observational studies are more uncertain than is usually recognised. Apart from the quantified sampling uncertainty, which is presented using confidence intervals or p-values, the additional model uncertainty is often substantial, typically unknown, and frequently ignored when authors present their conclusions from an investigation.

A mediator is a variable caused by the exposure and on the pathway to outcome.

Exposure -> Mediator -> Outcome

For example, when studying the effect of cigarette smoking on the risk of cardiovascular disease, carrying matches, is a mediator, on the pathway between cigarette smoking and cardiovascular disease. Adjusting for the mediator would be a mistake. It would likely remove a part of the otherwise estimated effect of cigarette smoking.

A collider is affected by common effects of both exposure and outcome.

Exposure -> Collider <- Outcome

For example, when estimating the effect of cigarette smoking on the risk of respiratory diseases, hospital admission may be a collider. Conditioning on a collider can induce endogenous selection bias, creating a noncausal association between exposure and outcome. It should not be adjusted for.


Comments

Post a Comment

Popular posts from this blog

Randomisation and alternation in clinical trials

  In early clinical trials patients were allocated to treatment and control groups by alternation, i.e. patients were assigned to the treatment group and control group in a strict order. The advantage with this method was that it produced groups with equal size. The two groups would also be fairly compared if the order of the patients included in the trial was random. However, the group allocation for a patient was predictable, which could potentially introduce selection bias into the trial's outcome through differential invitations to participate. The only way to avoid this issue was, of course, to make the group assignment concealed, which necessitated a non-systematic group assignment sequence. D'Arcy Hart (1) describes the development with two trials from the 1940s. The first one investigating the effects of patulin (a mycotoxin) on the common cold in 1943-1944 with strict alternation. The second one, having a randomised group assignment, investigating the effects of strept...
Read more

Conditional and marginal models

  Statistical analyses of related data need to take the relations into account to avoid misleading results. Two main types of statistical models are used, conditional and marginal models (1). A conditional model can be fitted within the framework of generalized linear mixed models (GLMM) and a marginal model using generalized estimating equations (GEE). Analyses based on conditional and marginal models give answers to different questions. While a conditional model can be used to estimate the outcome within a subject or cluster after conditioning on covariates, a marginal model can estimate the average outcome for the population accounting group-specific effects. For example, in a longitudinal study comparing patients receiving diet recommendations (treated) with patients receiving exercise recommendations (controls), uncontrolled blood pressure at repeated visits can be modelled using both a conditional and marginal model. Neither of these two models is automatically more correct ...
Read more

Pretesting normality

One common misunderstanding demonstrated in medical publications is the testing of normal distribution as part of a decision process: If Shapiro-Wilk Test or Kolmogorov-Smirnov Test indicates that a variable has a statistically significant departure from normal distribution, group mean differences are tested using the Mann-Whitney Test instead of Student's t-test. This decision rule may sound rock solid, but it may be a serious mistake. First, the practice treats absence of evidence as evidence of absence. A statistically nonsignificant normality test is simply not evidence of a normal distribution. Second, for a two-sample t-test or a linear statistical model, the relevant distributional issue concerns the difference or model residual conditional on covariates, not whether each observed treatment group's raw outcome values are exactly normal. Third, with small sample sizes, where the normality assumption may be important, the normality test has low power to detect non-normalit...
Read more