Error and Bias

Question 1

Error

Answer 1

Assessment of error in a particlar study is first step in process of making any causal inference

Question 2

Accuracy

It is divided in

Answer 2

How do we evaluate whether an association is true?
1. Validity
2. Precision (Reliablility)

Question 3

Validity

Answer 3

Concerned with the degree of
SYSTEMATIC ERROR (BIAS) or
NON-RANDOM ERROR

• Influences the validity of the study
• Affects actual estimate of “risk” (e.g., OR, RR)
• Can bias association toward or away from the null
• Divided in Internal and external validity
  1. Divided in
  1.1. Selection Bias
  1.2 Measurement bias
  1.3.Confounding

Question 4

Precision (Reliable)

Answer 4

Concernes with the degree of
NON SYSTEMATIC ERROR or RANDOM ERROR

• Influences reliability of the study
• Chance
• Affects variability around “risk” (e.g., 95% CI)
  1. Divided in
  1.1.random measurement error
  1.2 sampling error

Question 5

Selection bias

From validity

Answer 5

1. Case control studies
   1.1 Sampling C&C is dependent or based on exposure status
2. Cohort studies
   2.1 Self selection (Volunteers differ systematically from non-volunteers.)
   2.2 Loss to follow up (attrition)
   2.3 Non-responders
   2.4 Healthy worker effects (Workers are healthier than the general population.)

May bias the effect estimate compared to true estimate:
* Away from the Null (Large or small OR,RR)
* Towards the Null (Close OR,RR =1)

Question 6

Measurement bias (aka Information bias)

From validity

Answer 6

1. Recall bias (Cases remember exposure differently from controls.)
2. Reporting bias (Systematic underreporting or overreporting.)
3. Interviewer bias (Probing differently based on preconceived notions.)
4. Suverillance bias (Uneven monitoring leads to differential detection.)

Can be:
* Differential: Away from the Null (Large or small OR,RR 5.7). Different only in control group.
* Non Differential: Towards the Null (Close OR,RR =1) Similar among exposed vs unexposed

Question 7

Internal Validity

Validity

Answer 7

Accuracy of the study results. Inference to target population.

• Appropiate measures
• Appropiate techniques
• Accurate estimation
• True OR =3, Estimated OR =3

Question 8

External validity

Answer 8

How well the results relate to the target population

• Estimates are applicable or may be generalized to the larger population

Question 9

Minimizing Bias (Systematic error/Validity)

Answer 9

Selection Bias
* Objective case definitions and analytic procedures
* Enroll all aprticipants in defined time and place
* Similar procedures for selection all participants
* Strive for high participation and incentives to avoid loss to follow up

Measurement/Information Bias
* Training staff properly
* Blind participants and staff to exposure/disease
* Memory aids to help participants recall
* Standarize data collection methods

Question 10

Impacts of Bias (Systematic error/Validity)

Answer 10

• Affects the validity of the study
• Creates systematic bias and differences from true results that persist across multiple studies
• Cannot typically be “fixed” during analysis phase

As we increase the study size, impacs of random errors decrease.

Question 11

Random Measurement Bias

Precision

Answer 11

• Reduced reliability of measure
• Results in non-differential misclassification

Precision

Inconsistent data collection, leading to non-differential misclassification.

Question 12

Sampling Error:

Answer 12

• Non representative sample
• Reduced by increasing sample size

Non-representative sample due to chance.

Question 13

Minimizing Random error (Non-systematic error/Precision)

Answer 13

• Increase sample size.
• Standardize measurement instruments and protocols.