Epidemiology - Observational Study Designs
What is an ecological study?
A study that examines rates of disease in relation to a factor, both described at a population level - May describe groups by place/time - May use group/place level exposures and individual outcomes - Findings may not always apply to individuals
What is a cross-sectional study?
A study that examines the relationship between health and ill-health and other variables of interest in a defined population at one particular time - The population is usually selected without regard to exposure or disease status - It is a "snapshot" measure of current exposure or health
What is bias?
A systematic error in study design or conduct that yields erroneous associations - it can be introduced at any stage - increases spurious associations - retrospective or cross-sectional studies are prone to bias - data cannot be repaired once bias is introduced
What is confounding and how is it controlled for in an experiment?
A variable assocation with a risk factor and with the outcome, but not a path between the factor and outcome. It is controlled via exclusion, matching, and analysis
How is exclusion used to eliminate confounding and what are some complications with exclusion?
Exclusion is restricted sampling - selecting animals with only one level of confounder (subsequently removing its effect on the overall outcome) - Samples selected may not be representative of the larger population and may not apply beyond current group - May be harder to find enough animals
How are cohort studies defined by population and exposures?
Fixed cohort: cohort formed via irrevocable event Open population: exposure can change over time
What are cohort studies?
Healthy subjects are defined by exposure status and followed over time for incidences of disease
How are cohort studies defined by control group source?
Internal - unexposed members of same cohort General - preexistiing population date Comparison - members of other cohort
How is matching used to eliminate confounding and what are some of the complications with matching?
Matching involves equalizing the frequency of confounders within the group - If greater than 1-2 variables are matched it may be harder to find study participants - may throw out some important subjects that don't match the categories - you cannot analyze matched factors - you cannot use matching in cross-sectional studies
When are Odds Ratios used (and for what study types)?
Useful in any study - must use when you don't have a representative sample of the population - Case control, cross-sectional, ecological
How is analysis used to eliminate confounding and what are some of the complications with analysis?
Stratified groupings based on presence/absence of the confounder are analyzed separately. Commonly done with a series of 2X2 tables - More that a few stratified groups can result - big sample sizes are needed - multiple tables with increased chance that the differences you see are due to sampling errors, not ture associations
What are some things that can be done that make a study NOT worthwhile?
- Confounding - Bias - Study design flaws - Allocation within the study flaws
What are the four types of Observational Studies?
1. Cross-sectional 2. Ecological 3. Case-control 4. Cohort
What are the sub-types of observation bias?
1. Recall bias 2. Loss to follow up 3. Interviewer bias 4. Misclassification
What are two categories of bias?
1. Selection - systematic differences in characteristics of participants or group 2. Observation - systematic error in estimating exposure/outcome in one group
What is ecological fallacy?
Observed association on an aggregate level does not necessarily represent the association existing at the individual level
What do Relative Risks mean?
Odds of disease in exposed is XX times higher than non-exposed
What do Odds Ratios mean?
Odds of exposure in diseased is XX times higher than non-diseased
When is Relative Risk used (and for what study types)?
Only valid when you have a representative sample of a population - Cohort, +/- cross-sectional
How are cohort studies defined by timing?
Prospective - grouped by exposure and followed into the future to observe outcome Retrospective - exposures and outcomes occur pre-study
What are some examples of residual confounding?
Residual confounding are the "things that are left" that could screw things up: - Data not collected - Risks change within a factor category - Measure error/misclassification
What are case-control studies?
Sampling a population by enrolling diseased subjects (case group) along with a sample of the population that gave rise to cases (control/referent group) - the goal is to ascertain exposure factors giving rise to cases - case definition and selection of cases/controls is fundamental