Bias in Research

OVERVIEW

Bias

• Bias is the systematic distortion of the estimated intervention effect away from the “truth”, caused by inadequacies in the design, conduct, or analysis of a trial
• cannot be reduced by sample size (which reduces the effects of chance/ random variation and improves the precision, but not the accuracy of a trial)
• Systematic bias makes a study inaccurate i.e. lacking internal validity

Bias can affect any stage of the clinical research process:

1. in reading-up on the field
2. in specifying and selecting the study sample
3. in executing the experimental manoeuvre (or exposure)
4. in measuring exposures and outcomes
5. in analyzing the data
6. in interpreting the analysis, and
7. in publishing the results

Over 50 types of bias affecting clinical research have been described.

MAJOR TYPES OF BIAS

Selection bias

• Occurs when the selection of subjects into a sample or their allocation to a treatment group produces a sample that is not representative of the population, or treatment groups that are systematically different
• prevented by random selection and random allocation

Detection bias

• Occurs when observations in one group are not sought as diligently as in the other
• prevented by observer blinding

Observer bias

• Occurs when the observer is able to be subjective about the outcome
• prevented by observer blinding and outcome measure design

Recall bias

• Occurs when patients know which group they have been allocated to, which influences the way they report past history and symptoms
• ie. if patient knows the are in the placebo group they may exaggerate their ‘untreated’ symptoms
• prevented by patient blinding

Response bias

• Occurs when patients who enroll in a trial may not represent those of the population as a whole
• ie. the obese patients who enroll in a weight loss medication trial may be more motivated than those in the general population
• prevention -> random sampling from population

Publication bias

• Occurs because negative studies less likely to be submitted and/or published than positive ones
• prevented by clinical trials registries and ensuring all well conducted studies are submitted and published (should be mandatory)
• in meta-analysis, the possibility of absent negative studies should be sought for by funnel plot analysis

Regression to the mean

• Occurs when random effects may cause a rare, extreme variation on a measurement
• if the measurement is repeated, the likelihood is that the measurement will be less extreme
• thus, if a treatment had been given after the first measurement, it would erroneously appear, on the basis of the second measurement, that it had had an effect
• prevented by having a control group

Hawthorne effect

• Occurs when the  process of studying and following up patients itself influences the outcome
• ie. chronic headache may improve in patients who are being studied and regularly followed up
• prevented by having a control group and masking the intention of study from patients and observers

References and Links

LITFL

• CCC — Interpretative biases

Journal articles

• Sackett, D. L. (1979). Bias in analytic research. Journal of Chronic Diseases 32 (1–2): 51–63. PMID: 447779.