Game 6 · Study designs, risk and bias
What went wrong in this study?
Selection, information and screening biases, and confounding.
Read with: Bias and validity49 DNB questions
Play
Find what went wrong
Some rounds describe a made-up study and ask you to name the bias, or to choose the step that would have prevented it. Others ask you to split a table by a third factor and judge whether the association survives, or to drag a screening diagnosis earlier and separate lead time from real benefit. You have three lives.
Loading the first round…
Where bias enters
Name each bias by the stage at which it enters
Bias is a systematic error in the design, conduct, analysis or reporting of a study that moves the result away from the true value in one direction. Unlike random error, it does not become smaller when the sample grows. A larger biased sample only gives the wrong answer with a narrower confidence interval.
| Stage | Biases | Prevention |
|---|---|---|
| Selecting participants | Berkson's, healthy worker, volunteer, non-response, attrition, prevalence-incidence (Neyman), healthy user | Same source population, probability or consecutive sampling, high response and follow-up |
| Measuring | Recall, interviewer, observer, detection, misclassification, Hawthorne effect, social desirability | Blinding, standard tools, records made before the outcome |
| Analysing | Confounding (often classed separately), immortal time | Randomisation, restriction, matching, stratification, regression |
| Reporting | Publication, outcome reporting, language | Registration, prespecified outcomes, search of registries |
| Screening | Lead-time, length-time, overdiagnosis | Randomised trial with deaths from the disease as outcome |
| Diagnostic accuracy | Spectrum, partial verification (work-up), differential reference, incorporation | Reference standard for every patient, the intended range of patients |
Sackett (1979) listed 56 biases grouped by stage. The full catalogue with figures is in the bias and validity chapter.
Confounding
A third factor mixed into the comparison
Confounding occurs when the effect of the exposure is mixed with the effect of a third factor. A confounder must meet three conditions: it is associated with the exposure, it is a risk factor for the outcome even in unexposed people, and it is not on the causal pathway (the exposure does not cause it).
To check for it, split the data into strata of the factor and calculate the odds ratio in each. If the stratum odds ratios agree with each other but differ from the crude odds ratio, the factor was confounding, and the pooled Mantel-Haenszel odds ratio is the adjusted estimate. If the pooled comparison points the opposite way from every stratum, that is Simpson's paradox.
OR(MH) = Σ(a × d ÷ n) ÷ Σ(b × c ÷ n)
- a, b
- exposed people with and without the outcome, in one stratum
- c, d
- unexposed people with and without the outcome, in the same stratum
- n
- total number of people in that stratum
| Feature | Confounding | Effect modification |
|---|---|---|
| What it is | A third factor that distorts the exposure-outcome association | A third factor across whose levels the association truly differs |
| Stratum estimates | Similar to each other, but different from the crude estimate | Different from each other |
| What you do | Adjust, and report the pooled estimate | Report each stratum separately, and test for interaction |
| Removed by design? | Yes: randomisation, restriction or matching | No: it is a property of the disease or population |
Only measured confounders can be adjusted for. Randomisation is the only method that also balances unmeasured ones. With many confounders, use regression or propensity scores.
Screening
Why screen-detected disease seems to survive longer
Survival is counted from the date of diagnosis. Screening moves that date earlier, so survival after diagnosis grows by the lead time even if death is not delayed at all. This is lead-time bias. Slow-growing disease stays longer in the phase before symptoms, so a screening round finds it more often than fast-growing disease, and screen-detected cases do better whatever the treatment. This is length-time bias. Overdiagnosis is the detection of disease that would never have caused symptoms or death in the person's lifetime. These people are counted as survivors and raise survival figures further.
A screening programme is tested fairly by a randomised trial that compares deaths from the disease in the group invited to screening with the control group, with time measured from randomisation.
Writing the answer
How to write this in a 10-mark DNB answer
- Define bias as systematic error, and contrast it with random error in two lines.
- Draw the stages of a study (select, measure, analyse, report) and place each bias under its stage.
- Give one clinical example for each named bias, and the design step that prevents it.
- Treat confounding separately: the three conditions, an example such as coffee, smoking and MI, and the methods of control at design and analysis.
- If the question is about screening, draw the lead-time time line with diagnosis moved earlier and death unchanged.
- End with how bias is reported and assessed: STROBE item 9, CONSORT 2025, RoB 2 for trials and QUADAS-2 for diagnostic studies.
The one thing to remember
A bigger sample cures random error, not bias. Bias is prevented at the design stage, and confounding is checked by splitting the data by the third factor.
Past papers
DNB questions this game helps you answer
Questions from NBE theory papers for DNB, DrNB and Diploma courses, most often asked first. The tag after each question is the specialty and the year it was set.
- ×2Publication bias.
- ×1Discuss biases in case control studies. [3]
- ×1Discuss the various types of bias and measures to minimize them. [5]
- ×1Discuss the various ways to eliminate bias in a scientific study. [5]
- ×1Types of bias in biostatistics. [5]
- ×1What are various biases in controlled trials? [3]
- ×1Discuss the common types of biases in a research study. Describe the strategies to reduce the respective biases.
- ×1Epidemiological Investigation of RA (EIRA) study was conducted to investigate the influence of environmental factors on the risk of developing rheumatoid arthritis. 679 patients with rheumatoid arthritis were compared with 847 individuals randomly selected from the population for smoking status. The incidence of RA in current smokers, ex-smokers and ever-smokers, respectively, was compared with that of never-smokers. a) What study design was it? b) What sampling technique would be appropriate for such a study? c) What are the potential biases? d) What is the measure of effect calculated in such a study? e) Enumerate the pros and cons of this study design.
- ×1Potential sources of bias in diagnostic accuracy studies.
- ×1What are various confounding factors in clinical trials? Briefly discuss the double-blind design.
- ×1Disadvantages and biases associated with screening.
- ×1Write the steps for designing a cohort study to assess effect of air pollution on health. What possible biases can occur and how can they be reduced.
- ×1Discuss about case control study. Define odds ratio & the confounders if any in these studies?
- ×1Bias and Blinding.
- ×1Bias and confounding.
- ×1Discuss the various Bias while conducting a randomized controlled trial and ways to minimize them.
- ×1Risk of biases in RCTs.
- ×1Types of bias in research and their implications.
- ×1What is sampling bias?
- ×1Define bias in epidemiology. Describe four important biases observed in a case control study?
- ×1Discuss the various biases in clinical trials.
- ×1Types of bias in scientific studies.
- ×1Write what you understand by bias in a research study. Discuss the reasons for the same and the measures that can be used to decrease bias in results.
- ×1Screening Tests - Lead Time, Sensitivity, Specificity, Positive and Negative Predictive Values.
- ×1Role of bias in a scientific study.
- ×1What steps can you take while formulating a study to remove/reduce various Bias?
- ×1Bias in Case Control studies, explain with examples.
- ×1Define bias.
- ×1Enumerate types of biases.
- ×1Methods to control bias in Case Control studies.
- ×1What is confounding factor? How will you control them in epidemiological study, explain with examples?
- ×1Define lead time bias, length bias and over diagnosis.
- ×1What are the types of ‘bias’ in a study?
- ×1What is meant by lead time bias? Explain with an example.
- ×1What are confounding factors in a study and how to avoid them?
- ×1Concept of Lead time in Screening.
- ×1Lead time.
- ×1Selection and information bias
- ×1What kind of bias is the researcher is likely to encounter in this study?
- ×1a) What are the types of prospective studies and their advantages? b) What are the methods to avoid bias in a study?
- ×1Berksonian bias.
- ×1a) Active vs passive follow up. b) Cross sectional analysis. c) Bias.
- ×1What is lead time?
- ×1What are bias and confounding in statistical analysis?
- ×1What is bias in medical research?
- ×1Explain confounding. How logistic regression analysis can be used to handle confounders?
- ×1Define bias. Enumerate types of biases. Explain Neyman fallacy with an example.
- ×1Enumerate bias in epidemiological studies. How will you prevent them?
- ×1Enumerate differences between Research, Clinical Audit and Service Evaluation. What is ‘bias’ in clinical research and how it can be minimized?
Read next