Lecture 3. Causality and Error Flashcards
In a causal diagram, what do solid arrows represent?
A causal relationship between two variables
In a causal diagram, what do dashed lines represent?
Not a causal relationship, but a relationship driven by an unmeasured risk factor
What can causal relationships be?
Necessary and/or sufficient
What is an examples of a necessary and sufficient causal relationship?
e.g disease if and only if S. typhi pathogen is present
What is an examples of a necessary but not sufficient causal relationship?
e.g dietary factors enable S. typhi to adhere to the intestinal wall
What is an examples of a sufficient but not necessary causal relationship?
e.g. Shigellosis may also cause disease or influence severity (might have the S. typhi pathogen and not have the disease)
What is Hill’s Criteria?
The seven tests for causality
What are the seven tests for causality?
Strength of association
Dose-response relationship
Correct temporal relationship
Independent or recognised confounders
Consistency with other knowledge
Biologically plausible
Reversible
What is strength of association?
High Relative Risk or Odds Ratio – these are measures of association and estimate the increase or decrease in exposure/disease in a cohort or case control study.
What are examples of confident and non confident association strengths?
More confident: smoking increased the risk lung cancer by a factor of 22
Less confident: Oral contraceptives increased the risk of breast cancer by 1.2
What is dose-response relationship?
The disease increases as the exposure increases
What is an example of dose-response relationship?
The longer duration of smoking and higher daily number of cigarettes, the greater the reduction in life expectancy
What is a correct temporal relationship?
The exposure occurs before the disease
What study type are the best and worst at informing correct temporal relationships?
Easier to establish the sequence of events in a cohort study (prospective) than in case-control study due to potential imprecision in records
Cross-sectional studies do not inform correct temporal relationship
What does independent of recognised confounders mean?
Once known ‘causes’ are accounted for there is still a significant association
What is an example of independence of recognised factors?
Asbestos increases the risk of lung cancer, independent of smoking cigarettes
What does consistency with other knowledge mean?
If a study is the third or fourth to report the same findings then one can be more convinced that the relationship is causal
What are examples of consistencies with other knowledge?
There are many studies suggesting a link between smoking and cancer
Early studies on the risk of breast cancer following oral contraception use varied in findings
What does biologically plausible mean?
Coherence with the current body of biological literature
i.e. The association makes sense with other sources of data (makes sense mechanistically within the disease)
What does reversible mean within Hill’s criteria?
Remove the risk factor and there is a change in the outcome
What is an example of reversibility within Hill’s criteria?
If you stop smoking, life expectancy increases again
When might a study have insufficient power?
Failure to detect an association between two variables that is truly present in population
What are measurement errors and what do they result in?
Random, measurement error results in the dependent and/or independent variables being misclassified, but not systematically. Result is reduced likelihood of detecting a significant effect.
What do random errors result in?
Measurement error
Results in a reduced likelihood of detecting an effect
What does bias result in?
Results in misclassification, but occurs in a systematic way
Bias potentially modifies the association
What are the four types of bias?
Selection Bias
Recall Bias
Funding Bias
Bias due to confounding
What is selection bias?
Individuals are not representative of the population from which they are drawn because they were not randomly selected to be in a study
What study types are at risk of selection bias?
All of them
When does selection bias occur?
The stage of recruiting or retaining participants
What are the three types of population?
Target population (generalise to)
Source population (who do you invite)
Study population (who do we sample)
What is an example of selection bias in a case-control study?
1929 Study of 7,500 autopsy records
Hypothesis: Tuberculosis protected against cancer
Cases & Controls = 816 cancer cases (autopsies with malignant tumours) and 816 non-cancer cases (autopsises without cancer tumours) at John Hopkins Hospital
Lower rate of TB in cancer patients: 6.6% of patients with cancer had TB; 16.3% without cancer had TB.
Concluded TB was protective against cancer
What are the three common and important types of selection bias (ranked in descending order of how reliable they are)?
1) Case control studies - Selection of cases or controls on the basis of an exposure which is associated with the disease in question
2) Randomised Clinical Trails - Non-random allocation of patients to treatment groups in a clinical trial -Younger people, or sick people are offered the “new” drug
3) Cohort Studies - Loss to follow up (when one group is less likely to come back)
What is information/recall bias?
Patients with a disease recall their histories differently from patients without the disease
What are examples of information/recall bias?
A diagnosis of liver cancer might influence the recall of the amount of alcohol consumed in the past
Women generally report fewer sexual partners than men
What is funding bias?
It is very important when doing work that the results are impartial of the sponsors
When reading papers check who are the sponsors
What is bias due to confounders?
Confounding occurs when a variable is associated with a dependent and independent variable
A problem for observational studies
Well-recognised potential confounders are age, sex, social class, race
Most confounders are not measured because they are unknown (If we measured and accounted for it, then it isn’t a confounder)
What is an appropriate study design?
Type of study
Quantify power and significance and estimate required sample size
Measure the confounder and include in the analysis
What can ignoring confounders lead to?
Identifying associations that are false or failing to identify true associations
