Interpreting Epidemiological Findings Flashcards
Bradford Hill Criteria
helps infer what?
STRENGTH CONSISTENCY SPECIFICITY TEMPORALITY BIOLOGICAL GRADIENT PLAUSIBILITY COHERENCE EXPERIMENT ANALOGY
causation
strength?
strong association increases confidence that an exposure causes an outcome
but the 9 or 10-fold increase in risk was suggestive of causality
CONSISTENCY
repeatable findings by time, person, place is suggestive
as rules out errors/fallacies that befall one/2 studies§
most criticised criteria?
specificity as unusual for an effect to be related to only 1 cause
specificity can be informative but it’s absence doesn’t say much
temporality
exposure must precede outcome
cross-sectional approach determines presence of exposure and outcome
reverse causality?
He also asks whether in chronic diseases where diet X co-exists with disease Y, does diet precede disease, or do those with the disease begin to prefer a specific diet? This is reverse causality
BIOLOGICAL GRADIENT
we talk about deterministic effects where there is a continuous outcome (such as cataract or infertility) and stochastic effects where the outcome is discrete (cancer).
Quantification is often difficult! Like specificity, when present the biological gradient is suggestive, but it’s absence may be of no value.
coherance
does association keep up with existing science?
relationship may be deemed stronger where several elements of scientific understanding point in the same direction.
but if existing science is wrong?
however when most scientists are wrong, this criterion may support the status quo.
analogous findings
rubella is a viral illness- what rubella does other viral illness contracted during pregnancy may also do; deafness, eye and heart problem
thalidomide caused fetal abnormalities ; rational to suspect other antenatal drugs of doing the same
but reliant on exisiting knowledge
causation is not the same as association
T/F?
T
Correlation is a linear relationship between two variables
internal validity
The extent to which findings accurately describe the relationship between exposure and outcome in the context of the study
external validity is synonymous with?
generalisability
types of selection bias?
Healthy worker effect
non-response bias: is the process through which people who do not respond are systematically different to the people who do respond. This is a type of selection bias.
Lake Wobegon effect
illusory superiority
believe to be better than you are, when self reporting : information bias
Hawthorne effects
The precise definition is controversial, but is usually described as the consequence of participants realising they are being observed and therefore acting differently. For example, if you knew you were being studied, you might behave in a slightly different way to normal!
types of information bias
- recall bias
- response bias
- interviewer bias
- diagnostic bias
non-differential misclassification
Where cases and controls equally mis-report their exposures, this would more likely be described as
differential misclassification
if cases or controls were likely to report exposures differentially then it would be differential misclassification. But in this case, the statement says they are equally likely to mis-report their exposures
Where cases and controls unequally mis-report their exposures AND in such a way that the overall consequence is an association that tends away from the null, this is best described as:
information bias
differential diagnosis is always biased towards the null?
T/F?
false
can be biased towards/ away from null
non-differential misclassification always results in bias __ null
non-differential misclassification always results in bias towards null
selection bias
systematic difference in study
= systematic error in association / outcome
if participation in study sample is associated with both outcome and exposure then there is selection bias
Berkson’s bias
type of selection bias
hospital-based case control study and controls are selected amongst the hospital’s patients
healthy worker effect
Healthy worker effect: occupational studies where active workers are more likely to be healthy compared to those who have retired or stopped working
how to minimise selection bias?
Controls representative of target population
Minimise non-response
Compare respondents with non respondents
information bias
misclassification
y variables are not properly defined
Flaws in data collection
