Intro to study design Flashcards
What is an exposure give examples?
A causative factor e.g. physiological (height), lifestyle characteristic (exercise) etc.
What is an exposure also known as?
Explanatory variable, independent variable, X variable, risk factor, treatment group
What is an outcome?
A disease state e.g. periodontal disease
What is an outcome also known as?
Response variable, dependent variable, y variable, case/control group, disease group
What is a null hypothesis?
No association between exposure and outcome
Hypothesis: drinking fruit juice is not associated with dental erosion. In this hypothesis what is fruit juice?
The exposure
What are the two common types of studies?
Interventional and Observational
What are the two interventional studies?
Experiment and randomised controlled trial
What are the 5 types of observational studies?
Cohort, case-control, cross-sectional, ecological, descriptive
What is an interventional study?
Investigators test if changing something alters the outcome of the study
What is an observational study?
The investigator collects data associated with the occurrence/progression of an outcome with no intervention/manipulation of any kind
How is a randomised controlled trial designed?
Subjects randomly allocated to either treatment or control groups, all followed up then outcomes measured and compared
What are the two types of cohort study?
Perspective and historical
How is a perspective cohort study designed?
Exposure data is collected, then subjects followed up to see if exposed/unexposed develop outcome at different rate (only included if don’t present the outcome of interest)
How is a historical cohort study designed?
Exposure data is obtained from historical records then subjects followed to see if have experienced the outcome of interest at any time up until the resent
How is a case control study designed?
Cases (with outcome) identified first then controls (without outcome), then frequency of exposure measured retrospectively and compared
How is a case control study designed?
Cases (with outcome) identified first then controls (without outcome), then frequency of exposure measured retrospectively and compared
How is a cross-sectional study designed?
Exposure and/or outcome data collected at a particular point in time
How is an ecological study designed?
Units of analysis are populations or groups of people rather than individuals (e.g. to environmental exposures)
How is a descriptive study designed?
Exposure or outcome datant described in terms of time, place, or person
Which study design collects data on groups of people rather than individual?
Ecological
In which order do the different studies go on the heirachy of evidence?
- Systemic review & Meta Analysis
- Randomised controlled trial (eliminate confounding and bias is minimised)
- Cohort study (temporal sequence clear)
- Case-control study (attempt to asscertain exposure status before the onset of outcome)
- Cross-sectional study (measure exposure and outcome at same time)
- Ecological study (confounding cannot be excluded or accounted for as for population)
- Descriptive study (no comparison group = patterns difficult to interpret)
Before a relationship can be said to be causal what must be ruled out/considered?
Chance
Bias
Confounding
Reverse causality
What is chance?
variation that is due to random fluctations i.e. some untreated patients will get better, some treatment groups will still have the disease
Determined if chance by using p-values and confidence limits = assess evidence against the null hypothesis
What is Bias?
systematic departure from the true value = misleading results
What are the different types of bias (4)?
- Measurement (recall, interviewer, detection)
- Selection (of subjects)
- Loss to follow up
- Performance
What is confoudning?
association with third variable that provides an alternative explanation for observed association between exposure and outcome (may or may not have been measured = related to both exposure and outcome)
Which type of study is least likely to be affected by confoundng?
Randomised controlled trial (exposure allocated randomly so not associated with any other factors)
What is reverse causality?
Alternative explanation for observed association between exposure and outcome whereby outcome causes exposure rather than vice versa (e.g. oral cancer causes body weight)
Only an issue when data on both variables are collected at the same time so dont know which came first without temporal sequence
When can something be labelled causal?
- With consistency of evidence across studies in diff populations
- When there is a biologically plusible mechanism that links the two
Why is heirachy of evidence only a guide?
Quality also depends on how well the study has been designed (e.g. well designed case control may provide more valid evidence than a poorly designed cohort study)
What is a genetic study?
- Migrant, twin and adoption studies
- Sequence of human genome fully mapped (candidate genes/polymorphisms frequently identified & traditional epidemiological studies used to measure associations between genes and outcomes, assess gene-environment interactions and facilitate mendelian randomisation approach)
What is the medelian randomisation approach?
Observational -> esp. in cohort studies (need large no. of people)
= instrumental variable approach
Estimate of association between candidate gene (related to modifieable exposure) and outcome unconfounded and not affected by reverse causaulty
- genotypes randomly assigned from parent to offspring (no confoudners)
- Outcomes cant alter genotype (no reverse causality)
Which non-genetic approach to investigating assocations between modifiable exposures and outcomes has similarities to mendelian randomisation?
Randomised controlled trial
What are the similatrities beterrn mendellian randomisation and randomised control trials?
Write a null hypothesis:
No assocation between exposure and outcome in this population
In which type of study is data collected at one point in time?
Cross-sectional
What issues may there be with collecting information from subjects by questionarre or interview?
- Recall bias -> difficult to remember exact previous smoking habits
- Measurement bias -> Participants may be embarrassed about the questions so may be reluctant to answer or not answer truthfully
How could you deal with the issue of smoking as a confounder if designing a study to assess the association between periodontitis and coronary heart disease?
- Ensure detailed smoking data collected (adjusta analyses)
- Restrict study to non-smokers only (may reduce generalisability)
What is the main difficulty with using a mendelian randomisation approach in general?
Can only be used when it is possible to identify appropriate genetic variants
