Bias and validity

Bias & Validity in clinical research

Mohammad Shihata

Clinical Epidemiology

Definition:

“ The study of the distribution and determinants of health-related states or events in a specified population, and the application of this study to control health problems ”

Cl inical Epidemiology

Creating Evidence

Choosing Evidence

Using Evidence

Clinical Epidemiology Is used to:

Describe the NATURAL HISTORY of diseases

Describe CAUSALITY and ASSOCIATION between exposure and outcome

Describe the PROGNOSIS of diseasesEvaluate and compare DIAGNOSTIC tests

Epidemiologic Studies

Compare the frequency of an OUTCOME (e.g. disease rate, death, stroke,.....) in two (or more) groups according to a characteristic of interest (e.g. smoking, surgery, drug, ......). The characteristic is called the EXPOSURE.

Epidemiologic Studies

When the study is completed, the investigators compare the frequency of the outcome of interestbetween the study groups.

This comparison is used to describe the strength of causal relationship (association) between the exposure and the outcome.

Association

Apparent Association could be:

TRUE: NON-CAUSAL or CAUSAL in nature.

SPURIOUS: (Alternative Explanation)

Random Error

Confounding

Bias

Validity

Internal Validity:

Indicates a good construct of the study. If bias, confounding, and random error, are ruled out (or properly handled), the study has a good internal validity.

External Validity:

Indicates generalizability of the study results to a wider target population. (representative study sample)

Internal Val idity( Qual ity Control )( Qual ity Control )

Random error

Random error leads to false association between

exposure and outcome that arises from “chance”

an uncontrollable force that seems to have no

assignable cause.

How to assess random error

Hypothesis testing is a statistical method to assess the role of random error in the observed outcome of an epidemiologic study.

- p value

- Confidence intervals

CONFOUNDING

A systematic error due to mixing effect between an

exposure, an outcome, and an additional variable(s)

Male Male GenderGender

StrokeStroke

SmokingSmokingHTNHTNA-fibA-fib

?

How to control for Confounding

Design stage

Randomization

Restriction

Matching

Unequal distribution of confounders between comparison groups leads to inaccurate conclusions

Analysis stage

Stratification

Adjustment (Multivariable Analysis)

BIAS

A systematic error in the design, conduct, analysis

and/or, interpretation of a study that leads to an

erroneous association between the exposure and the

outcome.

BIAS

The only way to control for bias is to AVOID it with careful design and conduct of the study.

If bias is present in a data set it can not be corrected or adjusted for by any “fancy” statistical test.

BIAS

Usually does not result from a prejudiced investigator.

Can pull an association either towards or away from the null (H0).

The amount of bias can vary in its impact on the study.

Types of Bias

Types of Bias• Selection Bias:

• Information Bias

• Reporting Bias

Selection BiasA systematic error due to differences in characteristics between those who take part in a study and those who do not

Selection Bias

Examples:- Self selection bias- Referral bias - Healthy-worker effect-Attrition

Information BiasA systematic error in measuring exposure or outcome data for the study groups.

Information Bias

Examples:- Misclassification- Recall bias- Interviewer bias- Measurement bias- Compliance bias

Reporting BiasSelective revealing or suppression of information

Examples:- Publication bias- Citation bias- Language bias

Precision vs. AccuracyPrecision Accuracy

Definition The degree to which a variable has the same value when measured repeatedly

The degree to which a variable actually represents what its supposed to

Best Way to assess

Comparison among repeated measures Comparison with a reference standard

Value to Study Increase power to detect effects Increase validity of conclusions

Threatened by Random error Systematic errors

Precision vs. Accuracy

Conclusion

• Any epidemiologic study should be internally valid before assessing external validity.

• Random and systematic errors can threaten the validity of a study.

• Random error has no direction and can be minimized with repetition and larger sample size.

Conclusion

• Known confounders can be controlled for during study design or analysis.

• The only way to assure equal distribution of confounders (known and unknown) is randomization.

• Bias can only be avoided to assure the validity of a study.

Thank you