Class 13: Bias Flashcards
3 aspects to evaluate for internal validity?
Confounding or effect modification
Bias
Statistical significance
Define bias
Systematic (non-random) error in study design or conduct that leads to erroneous results.
Internal validity [based on deduction]
Valid relationship btw exposure and outcome
How can you fix a bias once it has already occurred?
YOu CANT. Nothing can be done to fix a bias after the study end.
3 elements of bias impact:
1-source/type
2-magnitude/strength
3-direction
2 and 3 are similar to confounding in the sense that it can skew magnitude and direction greatly
2 main categories of bias
Measurement-related
And
Selection-related
Measurement-related bias
Any aspect in the way the researcher collects info or measures/observes subjects which creates a systematic difference
Selection-related bias
Any aspect in the way the researcher selects or acquires study subjects which creates a systematic difference between groups
Commonly seen: subjects not coming from same population OR not representative of entire population.
What should you not do in a study
DONT DO ANYTHING THAT IS DIFFERENT OR CREATES A DIFFERENCE BETWEEN GROUPS
Types of selection bias
Healthy worker bias: cohort studies
Self-selection/participant [responder] bias: differences in those volunteering and those not volunteering.
Control selection bias: case-control studies
Bias in Cross-sectional studies
Newman bias: [selective survival]
More easily descriptive for longer lasting and more indolent diseases
Not effective for acute or rapidly fatal diseases.
Measurement:
Subject-related bias
1-recall bias: differential level of accuracy/detail [hawthorne effect]
2-contamination bias: control grp subject accidentally gets treatment
3-compliance/adherence bias:different compliances
4-Lost to follow up bias: witthdrawl or lost to F/u rates…differences btw those in study and those withdrawling.
Information/measurement:
Observer-related:
1-interviewer bias: interviewer influence on subject
2-diagnosis/surveillance (expectation) bias: different evaluation/diagnosis/observation btw groups.
Misclassification bias
Error in classifying either disease or exposure status…or both.
source of measurement (info/obs) bias
2 types: differential or non-differential
Non-differential Misclassification bias
Error in both groups equally…not as bad as differential-balanced error
Misclassification of exp. or disease is UNRELATED to the other
This moves the measure of asso. (RR/OR) Towards 1.0…reduces effect estimates of asso.
Moves closer to 1.0
Differential Misclassification bias
Error in one group differently than the other.
Misclassification of exp. or disease is RELATED to the other
Bias moves measure of asso. (RR/OR) either direction from 1.0–can be either attenuating or inflation
How to control for biases
Select the most precise,accurate and medically-appropriate measures of assessment and evaluation/obs
2 main process interventions to control for bias
Blinding/masking:
Random allocation of observers/interviewers
Also: multiple sources to gather info, build in many methods to minimize loss to f/u
