Lecture 10: Bias Part 1 Flashcards
What is an association?
-Identifiable relationship b/w an exposure and outcome
-In epi we measure the relationship using measures of association
-Implies that exposure might cause outcome
-The presences of an association does not necessarily mean the relationship is causal
What is bias?
-Bias refers to systematic errors that result in an incorrect estimate of the association between exposures and outcomes
-Error= deviation from “true” values
-When bias is present it reduces the validity of our findings
What are the 2 types of errors?
Random error: fluctuations around a true value due to chance (solution is to increase the sample size)
-Inherent in all measures
Systemic error: deviations that disproportionately affect the data not due to chance (solution is to carefully study design, analytical techniques)
-Over or underestimate in one direction only
-Ex everytime take a measurement overestimate by 10%
What are the 3 main types of biases?
- Selection bias
- Information bias
- Confounding
What is selection bias?
-Occurs in sampling phase
-Arises from the way subjects are selected, agree to participate, or agree to remain in the study
-The relationship b/w E and O among those in the study differs from that amount those who were potentially eligible for the study but did not participate
(when selection bias is present you are now biasing the association based on how you sampled those participants)
What are the 3 types of selection bias?
-Non-response/volunteer bias
-Detection/surveillance bias
-Loss of follow up
What is non-response/volunteer bias?
-Bias from differences in characteristics b/w those who choose to participate & those who dont
-“volunteers” tend to be more highly educated, more health conscious more compliant with physician orders (ie different than those who dont mostly general public)
ie who didn’t/did respond and the differences b/w them
What is detection/surveillance bias?
-Occurs when probability of detecting the outcome differs by exposure status (bc searching for disease you are more likely to find in that area)
-If exposure Is NOT associated with the outcome, but leads to a symptom that results in an increase in the search for the disease, the associated will be OVER-ESTIMATED (oral contraceptives: oral contraceptives use may lead to breakthrough bleeding and then PAP requirements which will increase cancer of testing and detecting uterine/ cervical cancer)
What is loss to follow up?
-occurs when participants who withdraw from a study are systematically different (in terms of the exposure and outcome) from those who remain
-Has to be related to BOTH E and O for bias to occur
(es of obese individuals drop out may see fewer ppl in the study that have both low physical activity and hypertension so the effect between then will be UNDERESTIMATED/’masked’
How can we look for selection bias?
Assessing selection bias
-Careful exam of study design/sampling method
-Explore non-responders (if possible)
Preventing selection bias
-Attention to good study design
-Obtain randomized sample for source population
-Reduce likelihood of selection biases (increased incentives, different recruitment strategies)
What is the second type of selection biased: information bias?
-Happens once they are already in study
-Incorrect classification or measuring of exposure, outcome, or other factors of interest can have:
-Measurement error
-Misclassification bias (non-differential or differential)
What is misclassification bias?
-Error in obtaining and classifying categorical info
-Can affect classification of exposure or outcome
may be due to
-Inaccurate diagnostic tests (miss class outcome O status)
-Poor questionnaire design (miss class E orO status)
-Non-compliance in a ransomed controlled trial (miss class treatment status)
-Use of surrogate interviews (asking parents what child did last night and give false info)
-Interview bias (not blinded feelings can impact results)
What is non-differential misclassification?
1 of 2 of the types of misclassification bias which is a type of info bias which is one of the 3 types of bias
-Error is not related to exposure or outcome-rather its an inherent problem with the data collection methods
-Estimate gets biased towards the null value-we are less likely to detect an associated, even if one exists (ex if 2lbs was added to everyones weight would make association look WEAKER than actually is, has to be to both groups)
What is differential misclassification?
-Magnitude and direction of the misclassification (E or O) is different in the 2 groups being compared
-Direction of bias can be either way (ie towards or away from null)
-If towards the null (weaker association than actually existed)
-If away from the null (stronger the association actually present)
-Depends on situation and how worded
What is recall bias? Differential or non-differential?
-An individual may not rememberer completely occurred something that occurred in the past
-ex mother whose child had birth defects (notable significance stick in head longer and therefore would remember what they ate over a person without sickness)
-Differential
