Alternative Explanations For An Association Flashcards
What could you look at to help assess the role of chance?
- Confidence interval around measure of effect
- P-value
- Sample size (needs to be large enough)
What is reverse causation? What is difficult to establish?
Reverse causation is the concept by which instead of the exposure causing the outcome, the outcome actually caused the exposure
Timing of onset of disease in relation to exposure is often difficult to establish
When is reverse causation likely to be a problem?
More likely to be an issue in studies that take a snapshot of the population:
- case reports/series
- cross-sectional
- case-control
In these study types it is harder to establish if the exposure really did happen before the disease
How is reverse causation overcome?
By ensuring the exposure measurements relate to the period before the disease onset, allowing for the fact that there may be a preclinical stage of the disease and that the person may experience symptoms prior to a diagnosis being made
What is confounding and when does it occur?
Confounding is when you see an association between exposure and disease/outcome that is not a true association
This occurs when the association is distorted by some other exposure which is independently related to both the outcome and exposure of interest
Give an example of confounding
In a study of occupation and lung cancer, miners were seen to be more likely to get lung cancer than others
Rather than coal dust causing lung cancer, the observed association may actually be due to the fact that miners smoke more than the general population and that smoking is the true cause of the increased risk
In this example, coal dust is the exposure, lung cancer is the outcome and smoking is the possible confounder
When can a variable NOT be a confounder?
When it is on the causal pathway
E.g. Blood pressure would not be considered a confounder in the relationship between exercise and heart attack
If confounder still are known, what should we do?
Collect data on them so we can then include them in our analysis
What are the 3 main ways in which confounders distort an association?
Confounders can work in any direction:
- They can overestimate an association whereby the relation is deceased when the confounder is accounted for
- They can mask a true relation between exposure and disease (initially there is no apparent relation but after allowing for the confounder an association appears)
- They can change the direction of an association completely (an apparent positive association may become negative when the confounder is allowed for, and vice versa)
How can we deal with confounding?
- Study design (randomisation/restriction/matching)
2. Analysis (stratification/multivariate)
What is randomisation and how can it combat confounding?
Individuals are assigned to the exposed and unexposed groups in an entirely random manner
This means that all confounding issues (known/Unknown) are distributed evenly across both conditions I.e. It breaks the relationship between confounder and exposure
However, this won’t work for all study designs, but is good for randomised controlled trials (e.g. Placebo vs. Drug)
What is restriction and what is an issue with this way of combatting confounding?
Here, the study group is restricted to only one level of the confounding variable (e.g. Just focus on non-smoking miners)
However, this limits the number of participants in the study and also limits how generalisable the findings are to the general population
What is matching in the context of confounding?
This is where a control is matched to each case based on potential confounders (e.g. Same age and gender)
Special statistics are required to analyse the data
Is confounding more commonly controlled for by study design or analysis?
Analysis
What is stratification and what does it rely on?
Stratification is when you carry out analysis and compute a measure of effect separately for each level of the potential confounder
It relies on the relevant information on the confounding factors being collected accurately
What is the difference between restriction and stratification?
Restriction takes place at the study design stage, whereas stratification allows all the data to be collect and then the analysis is restricted/stratified
What do multivariate analysis techniques effectively do in the context of confounding?
They adjust the estimated measure of effect, e.g. OR, to control for the confounders
Name 6 common confounders
- Age
- Gender
- Socioeconomic status
- Diet
- Ethnicity
- Smoking habit
What are the 5 possible explanations for an association?
- Truth
- Chance (sampling error)
- Reverse causation
- Confounding
- Bias
What is bias?
Systematic error in an estimate arising from problems in the study design/execution
It results in a measure of effect which can be either above or below the true value
What is the key difference between bias and confounding?
You can correct for confounding in the analysis - you cannot correct for bias
What are the two main types of bias?
Selection bias
Information bias
What is selection bias?
Arises from defects in study design relating to how people are chosen to be, or end up in the study
Bias can be introduced if the criteria for inclusion relate to the exposure of interest
Particular issue in case-control studies
What are three types of selection bias?
- Non-response bias (particular issue with questionnaire studies)
- Loss to follow-up bias (problem in cohort studies)
- Healthy worker effect (issue with occupational exposures - workers generally healthier than general population)
What is information bias?
Occurs when info is collected incorrectly or inaccurately I.e. When study subjects are misclassified according to their disease status, their exposure status or both
Give some examples of reasons why patients may be misclassified?
- Error in instrument
- Subjects give incorrect info
- researcher records info incorrectly
What are three common types of information bias?
- Recall bias
- Reporting bias
- Observer (or interviewer) bias
When can recall bias be a problem?
In case-control studies when cases may recall their exposure more completely than controls
When does reporting bias occur?
When individuals with a disease are more likely to report exposure
What is observer bias?
When the interviewer/researcher knows who the diseased and non-diseased people are and collects exposure info differently for them (or vice versa)
Dealing with bias requires getting the study design right - how would you achieve this?
- Select an appropriate study population
- Maximise response rates
- Minimise loss to follow up
- Use objective rather than self-reported measures of exposure/outcome
- Follow standardised procedures in data collection
- Blind interviewer
