top of page
Search

Bias & Confounding in Clinical Research (MRCP Part 1)

TL;DR: 

MRCP Part 1 regularly tests whether you can recognise bias and confounding, distinguish them from effect modification, and choose the correct control method. You are not expected to calculate statistics, but you must interpret study descriptions accurately. This article covers exam-favourite biases, confounding logic, common traps, and a short MCQ to consolidate marks.


Why bias and confounding matter in MRCP Part 1

Epidemiology and research methods are dependable scoring areas. Most candidates lose marks not because the topic is difficult, but because they confuse terminology or miss clues embedded in the stem. MRCP Part 1 questions are designed to assess whether you can think like a critical reader of medical literature, not whether you can design a trial.

Bias and confounding directly affect internal validity—whether the study’s conclusions are trustworthy. Understanding these concepts is essential not only for the exam, but also for safe clinical interpretation of evidence.


Scope of testing in MRCP Part 1

You should be able to:

  • Define bias, confounding, and effect modification

  • Identify common biases by study design

  • Predict the direction of distortion

  • Choose appropriate methods to control confounding

  • Avoid common exam traps (e.g. over-adjustment)

A full syllabus overview is available in the official MRCP(UK) guidance:https://www.mrcpuk.org/mrcpuk-examinations/part-1


Key definitions (non-negotiable for the exam)

Bias

A systematic error in the design, conduct, analysis, or reporting of a study that leads to an incorrect estimate of the association between exposure and outcome.

Bias cannot be corrected by statistical adjustment once it has occurred.

Confounding

A distortion of the true association caused by a third variable that:

  1. Is associated with the exposure

  2. Independently affects the outcome

  3. Is not on the causal pathway

Effect modification

A real biological interaction where the effect of an exposure differs between subgroups (e.g. by sex or age). This is not a bias.

High-yield biases you must recognise

1. Selection bias

Occurs when study participants are not representative of the target population.

Classic examples

  • Berkson bias – hospital-based case–control studies

  • Healthy worker effect – occupational cohorts appear healthier than the general population

Exam clue: “Patients recruited from a tertiary referral centre”

2. Information (measurement) bias

Systematic error in how exposure or outcome is measured.

  • Differential misclassification → bias towards or away from the null

  • Non-differential misclassification → usually biases towards the null

Exam clue: “Exposure was self-reported using an unvalidated questionnaire”

3. Recall bias

Cases recall past exposures differently from controls.

Classic MRCP scenario:Maternal drug exposure and congenital abnormalities in a retrospective case–control study.

4. Observer / interviewer bias

Assessment influenced by knowledge of exposure or outcome.

Control: Blinding, objective outcome measures

5. Attrition bias

Differential loss to follow-up related to outcome or exposure.

Exam clue: “30% of participants in the treatment arm were lost to follow-up”


Confounding: how MRCP expects you to think

Confounding is about baseline differences, not errors in measurement.

Example: An observational study finds that coffee drinkers have higher rates of pancreatic cancer. Smoking is more common among coffee drinkers and independently increases pancreatic cancer risk.

➡ Smoking is the confounder.


Flowchart of common clinical study designs tested in MRCP Part 1

How confounding is controlled (very examinable)

Design stage

  • Randomisation – controls known and unknown confounders

  • Restriction – limits variability (e.g. only non-smokers)

  • Matching – pairs participants by confounder

Analysis stage

  • Stratification

  • Multivariable regression

  • Propensity score methods

Exam pearl: Randomisation is the only method that controls unknown confounders.

Confounding vs effect modification (common trap)

Feature

Confounding

Effect modification

Nature

Bias

True biological interaction

Adjust away?

Yes

No

Reporting

Adjusted estimate

Stratified results

Exam clue

“After adjustment, effect disappears”

“Effect differs by subgroup”

Mini-MCQ (MRCP style)

Question: A cohort study reports that patients prescribed a new antihypertensive have lower stroke rates. Treated patients were younger and had fewer comorbidities. After age-adjusted analysis, the association disappears. What best explains the initial finding?

A. Selection biasB. Recall biasC. ConfoundingD. Effect modificationE. Random error

Correct answer: C. Confounding

Explanation: Age and comorbidity influenced treatment choice and independently affected stroke risk. Adjustment removed the distorted association.


Five most tested subtopics

  1. Differential vs non-differential misclassification

  2. Confounding by indication (drug studies)

  3. Randomisation vs regression

  4. Recall bias in case–control studies

  5. Effect modification vs confounding

Common MRCP Part 1 pitfalls

  • Calling effect modification “residual confounding”

  • Forgetting that matching requires matched analysis

  • Adjusting for mediators (over-adjustment bias)

  • Assuming all misclassification biases away from the null

  • Ignoring direction of bias when asked

Practical revision checklist

  • Memorise definitions word-for-word

  • Identify study design before reading the options

  • Ask: bias or confounding?

  • Predict the direction of distortion

  • Practise under time pressure using real MRCP-style questions

High-yield practice resources:


Frequently Asked Questions

Is bias the same as confounding?

No. Bias is systematic error from study methods, while confounding is distortion from a third variable.

Can confounding be completely eliminated?

Randomisation controls confounding best, but residual confounding may persist in observational studies.

Does regression remove bias?

No. Regression can adjust for confounding, not for bias.

Is recall bias possible in cohort studies?

Rarely. It mainly affects retrospective case–control studies.


Sources

 
 
 

Comments

bottom of page