Awareness of confounding factors (you could also call them confusing factors) is a crucial element in planning medical research projects as well as conducting statistical analysis and interpreting results. Confounding relates to factors that are associated in some way to both the predictor and the outcome variables and may account for all or some of the observed association. It is important to be aware of confounding so that we are not led to false conclusions of causality. In this article, we discuss what confounding is and how to identify potential confounders.

Confounding Definition

Confounding is the distortion of a measure of the effect of an exposure on an outcome due to the association of the exposure with other factors that influence the occurrence of the outcome. Confounding occurs when all or part of the apparent association between the exposure and the outcome is in fact accounted for by other variables that affect the outcome and are not themselves affected by the exposure [1].

Exposures, Outcomes and Associations

It is also important to define exposures, outcomes and associations from the outset.

• An exposure is a factor that may cause the outcome that we are studying.
• An outcome is a factor that may be a consequence of the exposure that we are studying.
• An association refers to exposures and outcomes that are statistically linked so that the probability of the occurrence of one factor varies with the probability of occurrence of the other factor.

It is EXTREMELY important not to interpret associations as meaning the exposure causes the outcome.

The Golden Rule in Epidemiology

The Golden Rule in Epidemiology is: “Association is not necessarily causation”

Associations may be present due to bias, confounding or chance – or true causality. You must do your best to exclude or minimise the effects of bias, confounding and chance and ensure your association meets the criteria of Hill’s considerations for causation before making any conclusions.

Recognition of confounding is important because failure to control for confounding factors can lead to the false belief that a cause-effect relationship exists when in reality it does not.

Identifying Confounding Factors

The criteria for identifying confounding factors are summarised by:

1. It must be a cause of the outcome of interest
2. It must be associated with the exposure of interest in the population under study
3. It must not be on the causal pathway from the exposure of interest to the outcome of interest, or be a consequence of the outcome

Directed Acyclic Graphs (DAGs)

Directed Acyclic Graphs (DAGs) are perhaps one of the best but least known ways to identify confounding factors from the design phase of your research study. They are easy to understand and they make great additions to your research proposal or final report. They demonstrate that you have considered confounding factors and they help you decide on the optimal variables to include in your data collection and analysis.

The example above would be an example of a simple DAG but when you are completing your DAG you will need to include more factors.

You can create a DAG manually using Powerpoint or you could try a free online tool such as Dagitty.net.

Steps to creating a DAG:

1. Identify factors that are already known to be associated with the outcome of interest
2. Draw relevant paths that connect to the exposure of interest and the outcome of interest.
3. Some confounding factors may also be linked or caused by each other so connect those too.

Now you can use this DAG to indentify the optimal set of variables you need to account for in your analysis. The optimal set may be the smallest, easiest or cheapest set to collect.

The advantage of the DAG is that it will get you to think about confounding and identify the variables that may be on the hypothesized causal pathway (in which case you would not consider them confounders). You also do not need to effectively control for the same variable twice, so it is important to consider confounders that may be on each others causal pathways.

Methods to overcome confounding

During the study design phase:

• Randomisation – groups are similar for both known and unknown confounding variables
• Matching – involves the selection of a comparison group that is forced to resemble the index group with respect to the distribution of one or more potential confounders
• Restriction – eliminates variation in the confounded. Confines study to those without the confounder

During the analysis phase: (however the variables must be collected during the study and you cannot adjust for unknown confounders)

• Stratify – produce groups so that the confounder does not vary. Works if there are only few confounding variables
• Multivariate Analysis – useful for when there is a large number of confounders

References:

1. Sim F, McKee M. Issues in public health. McGraw-Hill Education (UK); 2011 Sep 1.