Bias 2

Learning Objectives

• Define bias (systematic error) and differentiate between the three different types of errors:

  • Selection bias

  • Information bias

  • Confounding (to be covered in the next week)

• Define and identify different types of selection biases

  • Control selection bias

  • Self-selection bias

  • Healthy worker effect

  • Loss to follow-up

• Define and identify different types of information biases

  • Recall bias

  • Interviewer bias

  • Misclassification bias

• Identify the effect a particular bias can have on a study

• Identify which types of studies and which study features are prone to various types of bias

Information Bias

• Definition: Information bias refers to errors that occur in the information collected about study participants.

  • It is specifically related to the accuracy of the data concerning participant classification (exposed vs. unexposed, diseased vs. non-diseased).

  • Information bias arises after participants have been entered into a study.

• Consequences: It leads to an observed association that may significantly diverge from the true association if all participants were classified correctly.

Occurrence of Information Bias in Studies

• Information bias can occur:

  • In case-control studies where different techniques are used to collect information from cases vs. controls

  • In cohort studies where different procedures are used to collect information from exposed vs. unexposed groups

• Bias is exacerbated in case-control or retrospective cohort studies where exposures and outcomes are already established prior to participant selection.

Illustrated Example: Two-by-Two Table with No Information Bias

• Categories:

  • Diseased: Yes/No

  • Exposed: Yes/No

• Classification is accurate, leading to a proper understanding of exposure and disease without bias.

  • Categories:

  • Exposed case of disease

  • Exposed non-case

  • Unexposed case of disease

  • Unexposed non-case

Illustrated Example: Two-by-Two Table with Information Bias

• Categories:

  • Diseased: Yes/No

  • Exposed: Yes/No

• Classification is inaccurate, reflecting biased assignments in disease exposure.

Types of Information Bias

1. Recall Bias

2. Interviewer Bias

3. Measurement (Misclassification) Error

Recall Bias

• Example Research Question: Are birth defects associated with the use of the anti-nausea drug Bendectin in pregnancy?

• Problem: Mothers of affected infants may more accurately recall exposures to Bendectin than those without affected infants.

• This bias arises when participants with a disease (cases) recall or report their exposure differently than participants without the disease (controls).

• Hypothetical Data (Case-Control Study):

  • Cases (Birth Defects): 100 recalled using Bendectin; controls: 60% accurately recall usage.

  • Actual exposure recall distribution:

    • Cases (Bendectin+): 100%

    • Controls (Bendectin+): 60% (40% forgot)

• True vs. Observed Odds Ratios:

  • True OR = 1.0

  • Observed (Biased) OR = 2.3

Recall Bias: Solutions

• Use controls who are ill to ensure comparable recall.

• Utilize standardized, closed-ended questionnaires.

• Examine pre-existing data or biological measurements to determine exposure.

Interviewer Bias

• Research Question: Are birth defects associated with Bendectin usage during pregnancy?

• Design: Case-control study with birth defect cases vs. non-malformed controls.

• Problem: Interviewers may probe cases more than controls, leading to systematic differences in information collection.

• This bias influences how interviewers solicit, record, and interpret information based on participant case status.

Interviewer Bias: Solutions

• Implement blinding/masking to prevent interviewers from knowing case/control status.

• Utilize high-quality standardized questionnaires to enhance consistency.

• Validate data against existing records and provide rigorous training for interviewers.

Measurement (Misclassification) Error

• Definition: Occurs when participants are classified incorrectly concerning exposure or disease, often manifesting as the most common form of bias present in all study types.

• Sources of Error:

  • Self-report inaccuracies (e.g., high blood pressure, smoking)

  • Errors in medical records or death certificates

  • Data entry mistakes

  • Ambiguity in disease or exposure definitions

• Effects of Misclassification:

  • Non-differential misclassification: Biases results toward the null hypothesis.

  • Differential misclassification: Can bias results toward or away from the null hypothesis.

Non-Differential Misclassification

• Occurs when the extent of misclassification is equal across exposed and unexposed groups.

  • Classification categories are maintained:

  • D + (diseased), No D (not diseased)

  • Exposure: a, b (exposed), c, d (not exposed)

• Example: In a case-control study of bladder cancer where misclassification of smoking status is similar for cases and controls, resulting in both groups biased toward the null.

Differential Misclassification of Exposure

• Occurs when misclassification impacts groups differently.

  • Example of bladder cancer study: 95% of cases recall they smoke versus 75% of controls.

• The resulting Odds Ratios are affected due to the inconsistency in recall, leading to potential bias away from the null due to more accurate case reporting.

Poor Recall vs. Recall Bias

• Poor recall can happen generally, for example, unable to recall yesterday's breakfast or doctor's visits over the last year.

• Poor recall represents a form of non-differential bias, while recall bias signifies a difference in recall ability between diseased and non-diseased populations.

• E.g., 90% of cases vs. 70% of controls accurately recall suggests differential memory accuracy that biases results away from the null.

Measurement (Misclassification) Error Solutions

• Improve the accuracy of collected information:

  • Utilize the most accurate sources, multiple measures of exposure and disease, and validate data through corroboration.

• Note that it is challenging to rectify information bias after it has occurred; this type of bias must be avoided through careful study design.

Summary of Information Bias Effects and Prevention

Case Study: Coffee and Pancreatic Cancer

• Brian MacMahon (1923–2007) conducted a pivotal study on coffee consumption and its potential link to pancreatic cancer.

• Published in 1981, the conclusion suggested significant associations prompting widespread public and media reaction.

• Study methods involved interviews querying smoking habits, and coffee consumption details, with the lack of blinding potentially impacting results.

• Odds Ratios:

  • Coffee consumption showed variable associations across studies, with some suggesting weak links between cigarette smoking and pancreatic cancer.

Assessment of Information Bias

1. Identify possible sources of bias.

2. Analyze the likely impact on study results (both direction and magnitude).

3. Propose feasible solutions to minimize bias.

Recap: Types of Bias

• Selection Bias: Related to the selection of participants

• Information Bias: Concerns the accuracy of data collected from participants

• Note: Both bias types can co-occur within the same study, reflecting the complexities inherent in research design and analysis.

Quick Recap of Bias Types Across Study Designs

Questions and Discussion

• Engage with questions during the class and support hours for clarity and understanding of biases discussed.