10.2: Observational Studies ✅ Flashcards
Cross-sectional studies: characteristics
1) Descriptive
2) Analytic
-Random sample taken from the source population
-All assessments happen at a specific time point
-Exposure and outcome are assessed simultaneously
-Mostly deal with the present
-Many countries have annual cross-sectional studies to assess health of the population (called surveys)
Descriptive
Assessment of only 1 variable at a time
Analytic
Assessment of association between 2 variables
Descriptive measures for Cross-sectional studies
Proportions (categorical)
Prevalence (binary categorical)
Mean/median (numeric)
Measures of association for Cross-sectional
Odds ratio (binary categorical)
Mean difference (categorical exposure vs numeric outcome)
Regression/correlation coefficient (numeric exposure vs numeric outcome)
Pros of Cross-sectional studies
Easy
Cheap
Gives the opportunity to readily assess the prevalence of disease in a population
Cons of cross-sectional studies
Observational design often leads to information bias
-as mostly self reporting
Prone to confounding
-as can’t assess all potential confounders
Impossible to assess risk of a disease (incidence)
Impossible to establish temporality in an association
-as exposure and outcome are assessed simultaneously
Not appropriate for inferring causality between an exposure and an outcome (high likelihood of reverse causation)
Case group
A group of patients with a specific disease
Control group
Random sample of individuals not suffering from the specific disease
-has to be recruited from the source population
Case-control studies: characteristics
Case group usually matched for key characteristics to eliminate confounding
Overmatching should be avoided and restricted to 2-3 factors
-rest should be adjusted at the analysis stage.
Final step involves assessment of a series of exposures in cases and controls which occurred in the past
Retrospective
Retrospective
Start with the disease and then look back in time for identification of exposures
Measure of association for Case-control
Odds ratio (categorical)
Mean difference (numeric)
*note: outcome is ALWAYS binary
Descriptive measures for Case-control
Never calculated
->only analytical
Pros of Case-control
Can investigate determinants of rare diseases
Easy to perform
Cheap
Possible to assess several exposures and potential confounders for a single outcome
Possible to investigate exposures that happened in the long past
Cons of case-control
Observational design, prone to information bias (measurements are based on self-reports)
Prone to recall bias
(assessments of exposures are based on self reports of events that happened many years back)
Very prone to confounding
If the control group is not a representative sample, can lead to selection bias
Impossible to assess the risk of disease
Difficult to prove the causality due to confounding and bias
Cohort studies characteristics
Random sample is selected from the source population
Before study, all participants suffering from the disease of interest are excluded
At the study baseline, all exposures of interest and potential confounders are assessed
Following this, participants continue to live normally and researchers record new cases of disease during follow-up
Ascertainment of disease cases occur from clinical examination, hospital or national registries or self reports in specific time intervals
In a follow up, reassessment of exposures to see if they have changed
Main aim is to investigate whether the risk of different diseases differs based on different exposures
Prospective due to time-lapse between exposure and outcome
Descriptive measures of Cohort studies
Incidence (binary)
Mean/median (numeric)
*note: descriptive measures are rarely calculated in cohort studies
Measures of association for Cohort studies
Risk ratio (binary categorical)
Rate ratio (binary categorical when person-years can be calculated)
Mean difference (categorical vs numeric exposure)
*note: Odds ratio could be calculated when the binary outcome is very rare
Pros of Cohort studies
Can assess risk of disease
–> calculate relative risk
Can assess temporal associations
Can assess several exposures and outcomes so results are more valid
Can investigate risk factors for chronic diseases
Study samples are usually large, increases internal validity
Cons of cohort studies
Observational design, so is prone to informational bias
(as measurements based on self reports)
Selection drop out can lead to risk of selection bias
Observational design = prone to confounding
Hard to assess risk of rare disease outcomes
(as a large sample would be required)
Difficult to prove causality due to possible bias and confounding, but easier than compared to other observational study designs
