Study Designs 1: Observational Studies Flashcards
(36 cards)
What are observational studies?
Studies that observe distributions and determinants of health: they do NOT involve interventions
What are some examples of questions observational studies can ask?
How much disease is there?
What causes a disease?
What puts people at greater or less risk for a disease?
What can observational studies inform?
Health policy Planning + provision
Future research
What are the 2 types of observational studies?
- Descriptive studies that examine DISTRIBUTIONS e.g. how much measles is there in different UK regions?
- Analytical studies that examine DETERMINANTS e.g. is vaccination related to measles incidence?
What designs would be relevant for descriptive studies?
Ecological
Cross-sectional
What designs would be relevant for analytical studies?
Cross-sectional
Case-control
Cohort
What are the general outcomes of observational studies?
- Measure of effect size: How big is effect? How strong is association? - assess CLINICAL significance
- 95% CI: how precise is our measure of effect? - depends on error factor
- P-value: how likely is it that size of effect is due to chance? - assess STATISTICAL significance
What is the rule of thumb of P-values?
If it is less than 0.05, we reject our hypotheses, however if it is more than 0.05 we dont reject the hypothesis but not necessarily accept it (might need more data)
Null hypothesis P-value = 0 (no significance between variables)
In order of increasing reliability and validity, list the types of study.
Expert opinion Case reports Observational: - Ecological - Cross-sectional - Case-control - Cohort RCT SR/MA
What is an ecological study?
Studies where the unit of analysis is NOT the individual but a group e.g. school, area, region or country generally using available administrative data or population-level data
What are the advantages and disadvantages of ecological studies?
Adv:
- Good for hypothesis generation
- Cheap and simple to do
- Low or no ethics involved
Disadv:
- Can only establish association, not causation
- Data may be unreliable
What is the common measure of effect in ecological studies?
Correlation coefficient (R) using methods such as Pearson (ND), Spearman (skewed) + Kendall’s tau (ordered) - possible range -1 to +1 and there is a 95% CI for R value
Null hypothesis value = 0
What can R^2 be used for?
To evaluate the proportion of variation in the outcome (y-axis) explained by variance in the predictor (x-axis)
What are the potential biases of ecological studies?
Ecological fallacy
Confounding variables
Quality of data (timing)
Selective reporting
What is ecological fallacy?
Assuming that the population values are the same as the individual values
What is a cross-sectional study?
Studies where data are collected from a sample at a point in time
Can be repeated (usually with a different sample)
DESCRIPTIVE or ANALYICAL
What are the advantages and disadvantages of cross-sectional studies?
Adv: good for hypothesis generation
Disadv: can only establish association but not causation
What is the common measure of effect of cross-sectional studies?
Standardised difference (absolute difference if data is not ND’d) in means - 95% CI for difference in means + null hypothesis value = 0
Cohen’s D = (Mean group 1 - Mean group 2)/pooled SD
Small ES = 0.2
Med ES = 0.5
Large ES = 0.8
What are the methods for statistical testing in cross-sectional studies?
2 independent groups:
- Unpaired T-test (ND)
- Mann Whitney U test (skewed)
What are the potential biases of cross-sectional studies?
Sample selection bias Response bias Recall bias Responder/social desirability bias Confounding variables Direction of causation
What are case-control studies?
Analytic studies comparing exposure for a group with a condition (cases) and a group w/o the condition (controls) so look backwards - can have matched (1+ per case) or unmatched controls
What are the advantages and disadvantages of case-control studies?
Adv:
- Useful for rare diseases or diseases with long latent (undetectable/undetected) periods
- Fast + cheap
- Loss to follow-up not an issue
- Good to examine multiple exposures/risk factors
Disadv:
- Cannot measure incidence or prevalence
What is the common measure of effect of case-control studies?
OR = odds of being a case if exposed/odds of being a case if not exposed (may use adjusted OR to control for potential confounders)
95% CI + null hypothesis value = 1
How do you calculate an Odds Ratio (OR)?
Odds (case if exposed) = A/B
Odds (case if not exposed) = C/D
OR = A/B / C/D
