CLASS SESSION 20 INFORMATION BIAS Epidemiology 503 Section

CLASS SESSION 20 INFORMATION BIAS Epidemiology 503, Section 2

Announcements 2 Exams will be returned at the end of class.

Challenges to Internal Validity 3 1. 2. 3. Random Error: Not a systematic error, but chance or luck of the draw (sampling error). Bias: A systematic error in the design or conduct of a study which leads to results to an inaccurate estimate of the truth. Confounding: A confounding variable (3 rd variable) distorts the association between exposure and outcome.

Selection Bias 4 Because of problems in the selection or retention of study participants, the measure of association is distorted When selection/retention is related to BOTH exposure and outcome. � Case Control Study: selection/retention different among cases vs. controls and related with exposure. � Cohort Study: selection/retention different among exposed vs. unexposed and related with outcome.

Information Bias 5 When the quality of exposure or outcome data differs between groups and distorts your estimate of the true association. � An error arising from systematic differences in the way information on exposure or disease is collected from study groups � Incorrectly classify people as exposed/unexposed, or diseased/not diseased

Types of Information Bias 6 Recall bias: inaccurate recall of past exposure Reporting bias: inaccurate reporting of data Surveillance bias: follow one group more closely Interviewer bias: interviewer may influence response Measurement error: noise in measurements Any other bias in the way that exposure or outcome information is ascertained Information bias occurs AFTER participants have entered the study.

7 Information bias = misclassification of exposure or outcome status Does the misclassification differ among: � Cases and controls? Exposed and unexposed? NO: Non-differential misclassification Misclassification is same among cases/controls or exposed/unexposed. � Makes groups more similar so bias measure of association towards the null. � YES: Differential misclassification Misclassification is different among cases/controls or exposed/unexposed. � Can bias towards the null or away from the null. �

8 Today: Examine different types of exposure misclassification Disease Exposed None

9 Today: Examine different types of outcome misclassification Disease Exposed None

Exam 2: Median = 87. 91% Mean = 85. 93% 10