Information Bias

PubHlth 500 - Class Notes

Course & Date

  • Course: PubHlth 500

  • Date: Wednesday AM, 10/8

  • Institution: School of Public Health, University of Michigan

Confounding Practice

  • Example: Insurance advertisement suggests: "Does your auto have body damage? Fix it, and you’ll significantly reduce your probability of involvement in another traffic mishap."

  • Observation: Drivers of newly repaired autos tend to drive safer than those with unrepaired damage.

  • Question: What could confound this purported relationship between auto repair (exposure) and driving record (outcome)?

Learning Objectives

  • Differentiate between common types of information bias and their impacts on study results.

Mini Lecture: Information Bias

  • Definition of Bias: A systematic error in the design or conduct of a study that leads to an invalid estimate of the association between an exposure and an outcome.

  • Distinction from Confounding: Unlike confounding, bias distorts the true association rather than altering subject placement in groups.

  • Categories of Bias:

    • Information Bias

    • Selection Bias (to be covered next week)

Information Bias Defined

  • Systematic Error: Occurs in the measurement of exposure, outcome, or covariates leading to distorted estimates of truth.

Types of Error

  • No Error

  • Systematic Error

  • Random Error

  • Random & Systematic Error

Types of Information Bias

Misclassification

  • Definition: Measurement error where participants are misclassified on their exposure or outcome status.

  • Commonality: Most common form of bias in epidemiology.

    • Example: Classifying exposure or outcome status incorrectly using a 2x2 table.

  • Types of Misclassification:

    • Differential Misclassification: Error rates differ between groups (exposed vs. unexposed).

    • Non-differential Misclassification: Error occurs at similar rates across groups.

Specific Types of Information Bias

  • Recall Bias:

    • Definition: Differing levels of accuracy in information provided by participants.

    • In case-control studies, cases may recall exposures differently than controls.

    • In cohort studies, exposed participants may recall outcomes differently than unexposed participants.

    • Example: Mothers of children with ASD are more likely to recall medication use during pregnancy compared to controls.

    • Minimization Strategies: Utilize various questionnaire types, self-administered instruments, select diseased controls, and biological records.

  • Interviewer Bias:

    • Definition: Systematic differences in the solicitation, recording, or interpretation of information by interviewers.

    • Occurs when interviewers are unaware of case/exposure status and probe differently based on that awareness.

    • Minimization Strategies: Masking interviewers, utilizing closed-ended questions, and interviewer training on non-directive probing.

  • Reporting Bias:

    • Definition: Selective revealing or concealing of findings in study analysis, publication, etc.

    • Common Forms: Publication bias, where studies producing non-significant findings are less likely to be reported.

    • Impact: Magnified in meta-analyses based solely on reported findings, leading to interventions based on incomplete data.

    • Prevention: Registering studies and enhancing transparency through data availability and adherence to reporting guidelines.

Assessing Bias

  • Consequences of Bias:

    • Bias can either exaggerate associations (bias away from the null) or mask true associations (bias toward the null).

  • Differential vs. Non-differential Misclassification:

    • To determine if misclassification differs among groups:

    • If no, then it is non-differential (generally biases toward the null).

    • If yes, it is differential (can bias toward or away from the null).

Recall Bias Example:

  • Autism association studies show:

    • True Association: Affected individuals report 125 exposures versus controls reporting none.

    • Odds Ratios:

    • Exposed Cases (125 exposed, 500 non-exposed) produces OR = 2.33, while non-exposed produces OR = 1.00.

Activities: Assessing Bias

  • Analyze provided scenarios for type of bias and potential misclassification.

    1. Case-Control Study on ASD: Study staff may unconsciously delve deeper with case parents, causing discrepancies in reporting.

    2. Foodborne Outbreak Investigation: Attendees recall food consumed differently based on their health outcomes.

    3. Pesticide Exposure Effects on Parkinson’s: Use of visual cues potentially creates biases in reporting exposures between cases and controls.

  • Review Tasks:

    • Review the CONSORT checklist and STROBE checklist to understand their role in preventing reporting bias.

Final Thoughts

  • Recognize that misclassification occurs but can be minimized through best practices.

  • Important strategies include:

    • Not assuming non-differential misclassification.

    • Quantifying impact in future studies.

    • Continuously seeking ways to prevent bias.

Afternoon Session Focus

  • Today's PM session will concentrate on:

    • Building and assessing a misclassification matrix.

    • Identifying information bias in the literature.