Bias and Confounding Reading Flashcards
define bias
lack of internal validity or incorrect assessment of association between exposure and effect
how did Sackett and Choi classify biases
according to the stages of research
specification and selection of sample
execution of experiment
measurement of exposure
data analysis
what did Kleinbaum et al find
3 groups of biases
confounding
selection bias
information bias
explain competing risks
when 2 or more outputs are mutually exclusive and compete for same subject
e.g death only happens once and the risk factor for death can be affected by an earlier risk
what is healthcare access bias
patients admitted to institution do not represent cases originated in the community
either because admissions are determined by interest of health personnel (popularity bias)
or if patients are attracted to the prestige of a hospital (centripetal bias)
what is Neyman bias
e.g smoking and myocardial infarction
if smokers/non-smokers are interviewed after heart attack and smokers die more frequently then the sample left will show lower frequency of smoking
undervalues the association
what is spectrum bias
validity when researcher only included clear or definitive cases - dont represent conditions of normal diagnosis
specificity and sensitvity of a diagnostic test are wrongly increased
what is survivor treatment selection bias
those who live longer and more likely to receieve certain treatment
so retrospective analysis can yield positive association between treatment and survival
what is the healthy worker effect
lower mortality observed in employed population, tend to be healthier
what is Berkson’s bias
when probability of hospitalisation and cases and controls differs
influenced by exposure
why can matching sometimes cause selection bias
overmatching - when researchers match by non-confounding variables (associated with the exposure but not disease)
which underestimates the association
what is relative control bias
correlation in exposure status between cases and relative controls yields biased estimated of association between exposure and outcome
what does a non-random sampling bias yield
a non representative sample
what does loss to follow up affect
validity of statistical results
what are the 3 main types of information bias
misclassification bias
ecological fallacy
regression to the mean
what is misclassification bias
when sensitivity or specificity of procedure to detect exposure is not perfect so diseases subjects mistakenly classified as non diseased and vice versa
what is differential misclassification bias
when misclassification is different in the groups to be compared
what is non-differential misclassification bias
misclassification is the same across groups to be compared
in what studies does detection bias typically occur
its an information bias typically occurring in cohort/clinical trials
what is rumination bias
happens within recall bias when the presence of disease influences the perception of its causes
outline reporting bias
participants collaborating with researchers to give answers they believe are of interest
or underreporting bias is common with socially undesirable behaviours like alcohol consumption
what is ecological fallacy
bias when analyses realised in ecological (group) level are used to make inferences at individual level
what is the regression to the mean phenomenon
variable that shows an extreme value on first assessment will tend to be closer to centre of its distribution on later measurement
how can we neutralise regression to the mean phenomenon
use appropriate reference group
what is sick quitter bias
related to protopathic bias
people with risky behaviours quitting their habits as a consequence of disease - so studies label them as non-exposed and underestimate the true association
what is temporal ambiguity
when it cannot be established that the exposure precedes effect
common in cross sectional and ecological studies
outline confounding
when a variable is a risk factor for effect among non-exposed people and is associated with exposure of interest in pop from which the effect derives, without being affected by exposure or disease
at what stages can confounding be neutralised
at design stage - by matching or randomisation
analysis - if confounders have been measured properly
outline confounding by group
in ecological study when exposure prevalence of each community is correlated with disease risk in non-exposed of same community
can produce ecological fallacy
outline confounding by indication
when treatment is indicated by perceived high risk, poor prognosis or some symptoms
the confounder is the indication
mainly in retrospective observational studies analysing interventions
what is allocation of intervention bias
when intervention is differentially assigned to population
common in non-randomised trial
if sequence is known in advance may produce selection bias
what is compliance bias
degree of compliance influences efficacy assessment of the intervention.
what is contamination bias
when intervention-like activities find their way into control group
biases the estimate of intervention effect toward null hypothesis
