Bias and Confounding Flashcards
how did the Cochrane Handbook define bias?
systematic error, meaning multiple replications of the same study would reach the wrong answer on average
what is bias in terms of validity
lack of internal validity or incorrect assessment of the association between exposure and effect in target population
what is external validity
generalisation of the results observed in one population to others
how can bias arise in review and publication
biases of rhetoric
positive results bias (especially peer-reviewed journals)
‘hot stuff’ bias
‘alls well’ literature bias
what is selection bias
arises from procedure of selecting subjects and factors affecting participation
what is the consequence of selection bias
relation between exposure and disease is DIFFERENT for study population and target population
what is the target population
those eligible for inclusion in the study but not included
what are the 4 types of selection bias
- ASCERTAINMENT BIAS = inappropriate definiton of study population
- NON-RANDOM SAMPLING = lack of accuracy in sampling frame
- uneven diagnosis in target population
- during study implementation e.g withdrawal, loss to follow up,
what is attrition bias (selection)
systematic differences between groups on withdrawals from a study
how can we try to prevent selection bias?
chose study pop with extreme care - representative of the target population
use validated disease registers
maximise TAKE UP and FOLLOW UP among subjects using standardised protocol
compare characteristics of study population to people who didnt respond/others
when does information bias arise
during data collection
focus is misclassification bias
what is misclassification bias (information)
bias in classifying either exposure or outcome
what is observer (interviewer bias) bias
(information)
an interviewers knowledge may influence structure of questions and manner of presentation
which may influence responses
what is recall bias - info bias
those with particular outcome/exposure may remember events more clearly or amplify their recollections
what is reporting bias - info bias
participants may give answers they perceive to be of interest to investigators
what is the Hawthorne effect - info bias
people act differently if they know they are being watched
what is detection bias - info bias
if greater effort is given to diagnosing disease in exposed group than the unexposed group
what is protopathic bias
exposure is influenced by early stages of disease
e.g drug is prescribed for early manifestation of disease that has not yet been diagnosed
how can we deal with information bias
BLINDING - participants/those assessing exposure/ those assessing disease status
use standardised definition of exposure and disease
under standardised data collection instruments
maximise response to questionnaires
what is ‘mistaken identity bias’
where strategies meant to alter patient behaviour may also alter clinician behaviour
what is cognitive dissonance bias
belief in given mechanism may increase rather than decrease in face of contradictory evidence
what is significance bias
statistical vs. clinical significance
how to avoid analysis bias
state hypotheses to be tested in advance
follow analysis plan
how to avoid interpretation bias
treat all interpretations with caution
look for evidence in systematic reviews
what is confounding
the effect of extraneous factor is mistaken for or mixed with effect of the exposure under investigation
what is the Simpsons paradox
correlation present within different groups may disappear or even be reversed when the groups are combined
how can we control for confounding factors in the study design
restriction (exclude people)
matching (select people to minimise correlations with confounders)
randomisation
how can we control for confounding factors in study analysis
stratification (make comparisons within specific levels of confounders e.g gender, age)
standardisation
what is the low birth weight paradox an example of
children of smoking mothers more likely to be LBW, and LBW children have higher infant mortality rate
BUT
LBW children of smoking mothers have LOWER infant mortality rate
because of Simpsons paradox
what studies is confounding particularly important for
non-randomised studies
