L3B - Confounding and Bias
These terms are used to describe factors (that…
L3B - Confounding and Bias
These terms are used to describe factors (that are not a specific target of study) that may influence, or are responsible for, a study’s outcome
Last week we learned….
- That different clinical questions are answered by specific study
- The results obtained by these different study designs have a different
value in the hierarchy of evidence
-Their relative values are a reflection of the internal validity of these studies, or… their risk of bias
- It refers to the reliability or accuracy of the study results
- An internally valid study is one which measured what it set out to
- Is an absolutely essential feature of clinical research
Lacking Internal Validity
If a study LACKS internal validity, then something ELSE might be influencing the outcome => CONFOUNDING FACTOR
Are the results generalisable to different populations outside of the study population?
Trialists vs Epidemiologists
- Trialist =>
Selection Bias, Performance Bias, Detection bias, Attrition Bias, Reporting Bias
- Epidemiologists =>
- Selection bias
- Information bias
Any factor, other than the factor of interest (independent variable), that influences the outcome of a study/experiment
E.g. The objective of an experiment might be "Does A cause B". A confounder is anything that could cause a change in B, but is not A.
A uncontrollable factor
Mitigating the Effect of Confounders
- For RCTs; Randomisation should eliminate the influence of confounders
- For Observational studies, techniques include:
• Statistical techniques (multivariate analysis)
Because the confounding factors are evenly distributed between the groups, and thereby their effect is mitigated
- Observational Studies cannot Randomise
Restriction (Exclusion or specification)
The simplest way of avoiding a confounder
- participants are screened and excluded from the study if known to possess the confounder.
- May slow recruitment
- Increases Internal validity but may result in poorer External Validity
Involves “matching” the confounder in the case group with a participant in the control group who also has the confounder (age and sex)E.g. if smoking is a confounder, when enrolling a smoker in the case group, the control group must also enroll a smoker
- If there are several confounders, this can become onerous
- By doing this, by definition the study can no longer examine the effect of a matched variable.
- This strategy is applied after study completion
- Essentially uses the same approach as "restriction" => stratifies consituents into two groups during analysis stage
=> Enables comparisons to be drawn and illuminates the impact of the onfounder
- Exposed to confunding Factor
- Unexposed to confounding Factor
Any systematic error involvd during collecting and/or interpreting data
- May result in either: Over or Under estimation of study effect
- Undermines internal validity
- Some biases are inherent to a specific study design, whilst others may be omnipresent
Systematic differences between baseline characteristics of the groups that are compared.
Systematic differences between groups in the care that is provided, or in exposure to factors other than the interventions of interest.
Sometimes it's impossible to avoid without encountering ethicl dilemmas - e.g. effects of major surgery
Systematic differences between groups in withdrawals from a study.
- Reduction inthe number of participants as the study progresses
- Loss of particpiants alters group characteristics, effecting interpretaton of outcome
- Attrition may be random, but the bias lies in the fact that there must be something different between those who stay and those who leave
Example; Interventional study of diet and Depression
- Attrition of <5% is unlikely to be significant
Attrition of >20% is likely to be sigificant
Systematic differences between reported and unreported findings.
- ~50% of studies are not published
Studies with positive outcomes tend to be preferentially published to those that are negative, and published faster
Originatea from highly competitive research environment, career promotion concerns, and competition over funding
Using the Funnel Plot to Identify Biases
- A scatter plot plotting the result (often standard error of the effect) and precision (odds ratios) of multiple studies
- Assymmetry is indicative of publication bias
- Multiple biases interact with one another to eventuate in the assymmetry - not just Publication bias