Week 1 Flashcards
What are the three questions to find out confounders?
Q1: is the variable a risk factor/protective factor for the outcome?
Q2: is the variable associated with the exposure?
Q3: is the variable not an intermediate in the causal pathway?
What is selection bias in a cohort study?
When there is a difference in the loss to follow up or enrollment procedures
What is the target population in a cohort?
The population you want the study to be generalizable to
What is the source population in a cohort?
People who can join in the study
What is the study population in a cohort?
People who start the study with you (reply to the invitation)
What is the population for analysis in a cohort study?
People with all data available
Which populations differ from each other when external validity is not good?
The source population differs from the target population.
Which type of selection bias is likely with cross-sectional and case-control studies?
Selective enrollment issues
Which type of selection bias is likely with prospective cohorts and RCTs?
Selective loss to follow up?
Which populations differ from each other when selection bias occurs?
study population from the source population or population for analysis
What are the disadvantages of a cross-sectional study?
reverse causality, assessment of prevalence, selection bias (enrollment)
What are the disadvantages of a cohort study?
Expensive, time-consuming, confounding, external validity (effect modification), long enough follow up time?, selection bias (loss to follow up), information error
What are the advantages of a cohort study?
Cause and effect is clear, rare exposures can be studied, possible to investigate exposure over time, possible to study many different outcomes
What is the study outcome of a cross-sectional study?
Prevalence ratio
What is the outcome of prospective studies?
IRR/IPR = RR
What is the outcome of case-control studies?
OR (estimate of IRR)
With what study designs do you match cases with controls?
(nested) case control
What are the weaknesses of a case-control?
Selection bias (enrollment), information error (recall bias), confounding (reduced by restriction/matching/stratification), effect modification, reverse causation, hard to define exposure afterwards
What are the advantages and disadvantages of looking at prevalence vs incidence in a case-control?
Prevalence: more cases but they could have changed their exposure
Incidence: no changed exposure but you need to interview people very soon after their diagnosis
What are the advantages and disadvantages of looking at hospital-based vs population-based controls in a case-control?
Hospital based controls: potential information errors are the same as in the cases, but selection could be related to the exposure
Population based controls: they are more representative for the source population, but it is hard to reach the population
What are the advantages of a case-control?
You can study rare diseases and many risk factors at once
What are the weaknesses of a RCT?
Ethical, short term, fixed exposure, intermediary endpoints, high risk populations, effect modification, selection bias
Is the case cohort a prospective or retrospective design? And what type of bias is therefore limited?
Prospective, recall and selection bias
What people are the comparison in a case cohort?
A random sample of the whole cohort: sub-cohort. They are all free of disease at baseline
What are the advantages of a case-cohort?
Efficiency, reduces selection bias and recall bias, can be used for multiple diseases and sub-cohort can be used to calculate person-time risk
What are the disadvantages of a case-cohort
Need large cohort, reduces power, loss to follow up, no matching, potential sample degradation
Is the nested case control a prospective or retrospective design? And what type of bias is therefore limited?
Prospective, recall and selection bias
Describe the selection of cases and controls in a nested case control study
Cases are all new cases of the disease and each control is sampled from the cohort with incidence density sampling
What are advantages of the nested case-control?
Efficiency, reduction of selection and recall bias
What are disadvantages of the nested case-control?
Reduced power, case-control pairs can only be used for 1 outcome, person time is not estimate
When does selection bias occur?
When those selected to be in the study differ from those not selected
Is selection bias a threat for internal or external validity?
Internal
What are two types of selection bias?
Enrollment and follow up
What are two types of information error?
Measurement errors (continuous variables) Classification errors (discrete variables)
What will happen with a distribution when there are random errors?
The distribution will become wider -> more people in the tips.
What is sensitivity and specificity regarding information error?
Sensitivity: percentage of the exposed that are correctly classified
specificity: percentage of the unexposed that are correctly classified
How can you check for confounding when looking at graphs?
Draw an imaginary line through the data
If the crude OR is 3, and the stratified ORs are 1,5 and 1,5 is it confounding or effect modification?
Confounding
If the crude OR is 3, and the stratified ORs are 4 and 1,5 is it confounding or effect modification?
Effect modification
How can you control confounding?
Through the design (restriction and matching)
Through data analysis (stratification, multivariable modelling)
Will the power increase or decrease when the effect size will get larger?
Increase
Will the power increase or decrease when the variance will get smaller?
increase
Will the power increase or decrease when the sample size will get smaller?
Decrease
Will the power increase or decrease when the alpha will get bigger?
Increase
What is counterfactual thinking?
What if the situation would be exactly the same but with 1 thing different
How can you imitate counterfactual thinking?
With well controlled RCTs
What is additivity, synergism and antagonism in the causal model of rothman?
No interaction, and interaction
If the background risk is low, are you more or less likely to pick up a high RR?
More likely
What do you do when the heterogeneity is high with a meta-analysis?
Try to stratify or subgroup analysis
What is a sensitivity analysis?
Exclude studies with a high risk for bias
How can you check for publication bias?
With funnel plots, if it is symmetric, there is no publication bias.
What is the difference between fixed and random effect models?
Fixed: you assume that there is 1 true association
Random: assume populations differ from each other
