Bradford Hill Criteria
If study was replicated in different time + place, same association?
If exposure to cause goes up, does effect follow/vice versa if considering treatments
If the cause is removed, does the effect go away?
- Amount of variation from the mean value
- Standard deviation of means from multiple samples
- No sample fully representative
- STANDARD ERROR: measure of deviation of sample's mean from true popn mean
- Higher sample sizes reduce SE
- Describes the pattern
- PREDICTS, DESCRIBES, ADJUSTS
- Product of case-control
- Odds of event occuring in one group compared to odds of another
- Thing over other thing
- If odds ratio >1, there may be an association between risk factor and event
- <1, factor may be protective against event
- = 1 factor makes no difference
- Measure of statistical significance
- Probability than an effect is not due to chance alone
- p<0.01: very strong presumption against null hypothesis
- p<0.05: Strong presumption against null hypothesis
Type 1 error
- Opposite to what you are testing
Type 2 error
- Null hypothesis incorrectly rejected:OPTIMIST
- Null hypothesis incorrectly accepted: PESSIMIST
Stronger methodology, Less Bias, Controls for comparison, Fewer studies
Randomized controlled trials
- Systematic reviews
- One intervention group, one control group
- Participants truly randomised to one of two groups
- Blinding to remove bias
- Best but expensive
Case control Studies
- Longitudinal, prospective study
- Starts before disease onset
- Two groups: exposed to risk factor and not exposed
- Groups are followed and incidence of disease is monitored
- Produces relative risk
- Expensive and time consuming
- Observer bias but reduced recall and selection bias
Case reports/ Case series
- Exposure to risk factor assessed in group of cases (people with disease) compared to a group of controls (people without disease)
- Matching is employed as control of confounding risk factors
- Only proves association, not causation
- Produces an odds ratio
- Recall and selection bias
Animal and lab research
- Individual case
- Number of same/similar case
Cross Sectional Study
- Correlates two or more variables to investigate relationship
- (Prevalence study): snap shot of popn, good indicator of prevalence
- Risk of disease in exposed relative to unexposed. Expresses how many more times likely even is to occur in one group compared to another
- Thing over total
- Probability of event occuring
- Criteria should be applied before implementing potential popn screening programme
- Is condition an important health problem?
- Is there suitable exam/test?
- Is there accepted/effective treatment?
- Diagnostic and treatment facilities available?
- Disease have latent/early symptomatic stage?
- Natural record of condition known?
- Agreed policy on who is case + treatment
- Economically viable?
- Continuing process?
- People identified as having disease, further tests they do not
- People identified as having disease, further tests confirmed
- People identified as not having disease, further tests they do
- People identified as not having disease, further test confirms
Sensitivity and specificity: Measures of accuracy of screening test
Sensitivity: True +ves/ (True +ves + False -ves)
Specificity: True -ves/ (True -ves + False +ves)
- Everyone that has disease
- Everyone that doesn't have disease
PPV: positive predicitive values
NPV: Negative predictive values
- The chance, if you test +ve, you will have disease
- The chance, if you test -ive, that you will not have disease
- How good a test is at identifying correct answer