Bias and Misclassification Flashcards
Bias
A systematic ( non-random) error in study design or conduct leading to errounous results….distorts the relationship btw exposure and outcome
Bias
A source of error for assesing internal validity
What can be done to fix a bias
Nothing after the study has ended, bc it has already occured. ANy consideration must be done before hand.
What 3 components do researchers look for when assesing for a bias?
Sourse, type… magnitude, strength… direction
for magnitutde can account entirely for wk association, but not likely to acount for entire very strong association… Large OR/RR
The direction of a bias
can over or under estimate the true measure of association
Information/observation/ measurement related
Any aspect in the way the researcher collects inforamtion or measure/observes subject which creates systematic differences btw groups in the quality/accuracy of their information
Selection related bias
any aspect in the way the reasearcher selects subjects which creates a systemeat difference in the composition btw groups…. not reprenting the full population… P value tell if groups are equal
Selection bias
Selecting study subjects that are not representative of your primary population of interest or that generates differences in groups being compared
Healthy worker bias
can easily be seen in prospective cohort studies… workers usually exhibit lower overall death rates than the general pop bc severly ill and disablesd ppl ar eexcluded form employment
Self-Selection/ Participant (responder) Bias
those that wish to participate(volunteer) may be differnt in some way to those that don’t bolunteer or self-select(refusal/non repsonse) to partcipate
Control selection bias
can easily be see in case control studies
It’s hard to find controls. they have to be as close as possible without ever having disease.
Recall Bias
A differential level of accuracy in provided information btw study groups. People who have had a bad experience have a much clearer memory of detail
Hawthorne effect
The nobelty of being research subjects and the increased attention from such coudl lead to temporary increases in study examples responses
Can loss to followup happen in a cohort study?
Yes if it is prosepective
Contamination bias
interventional study…. members of the control gourp accidently or outside of the study protocol receive the treatemnt or are exposed to the intervention being studided
compliance bias
interventional studies, groups being interventinoally studdied have difference compliences
lost to followup bias
groups being studied have different withdrawal or lost to followup rates or there are other differences btw those that stay in the study and those that withdraw or are lost to fllow up
differential vs not differential
What are selection related biases?
healthy worker biasas, self selection bias, control selection bias
What are subject related variation?
Recall, contamination bias, complaince bias, lost to followup bias,
What are observer related variation?
Interviewer, diagnosis/surveillance,
interview bias
a systematic difference in soliciting, recording or interpreting on the part of the reasearcher, interviewers knowledge may influence response, and not apply equally btw groups. why a double blind study is good
diagnosis/surveillance(expectation) bias
different evaluation/classification/diagnosis/observation btw study groups. observer may have preconceived expectiations of what they should find… hawthorne like effect for researchers perspective
kappa statistic
allows statistics to be done on evaluators to see if there is any differene btw them.
Controlling for bias
Select most precise, accurate, medically-appropriate measures of assessment and evalution/observation… used published and past utilized techniques and process if possible
ways to control for bias
blinding/masking, randomly allocate observers/interviewers for data collection, build in as many methods necessart ro minimize loss to followup, start with more people then needed for loss to follwo up
Misclassification Bias
error in classifying either disease or exposure status or both… putting ppl in the wrong box
non-differential
error in both groups equal…. Misclassification of exposure or disease which is unrelated to the other, depending on study design
non-differential effect
For dichotomous 2 category variables, bias can move the measure of association TOWARD THE NULL HYPOTHESIS….. MOVES IT TOWARD 1.0… conservatie error
Differntial
misclassification of exposure or disease is related to the other, depending on study design
differential effect
bias can move the measure of association IN EITHER direction… it can inflate or attenuate your effect… differntial is related to the opposite element
I unknowingly misclassified 10 cases of exposure in ONLY the cases
this is differential, bc it only effected the one side,
There is a misclassification on both sides, but not in equal amounts. 10% on one side, 40% on other
this is differential, bc they are not balanced… still shifting the measurements unequally
Misclassified both groups by the same amount
non differential… pushes the OR towards nutrality, and more likely for us to say there is no difference so the error is conservative and less painful
Controlling for misclassifications
same as information bias
Neyman bias
fatal cases being excluded, only looking at the living.
Lead-time bias
amount of time a person has to see the disease present, cancer vs strep throat
length bias
…
