Fifty marks of Paper B are critical review, and candidates treat it as the hard part. It is not the hard part. It is the easy part that everyone leaves until it is too late to make it easy. Clinical psychiatry is unbounded: there is always more to know. Critical review is a closed set of concepts and a single repeatable procedure, and once you have the procedure, every study you are shown yields to the same five questions asked in the same order. Learn the procedure, practise it on studies, and a third of the paper stops being a threat.
Key takeaways
- Apply the same five steps to every study: question, design, bias, effect, conclusion.
- Most marks are lost at step five, because the conclusion frequently does not follow from the data.
- The statistical content is small, finite and recurs every diet, which makes it exceptional value.
- Practise on studies, not on textbooks, because this is a procedure rather than a body of knowledge.
- Twenty minutes twice a week from the start beats a fortnight of reading at the end.
Step one: what was the question?
Before anything else, state what the study was actually trying to find out, in one sentence.
This sounds trivial and it is the step that prevents most errors, because a great many critical review questions turn on a mismatch between the question the study asked and the conclusion the authors drew. A study that measured a surrogate outcome cannot tell you about a clinical one. A study that asked about association cannot answer a question about causation. A study in one population cannot be extrapolated to another without argument.
If you cannot state the question in a sentence, you cannot judge whether it was answered.
Step two: what was the design, and was it the right one?
Name the design, and then ask whether it was capable of answering the question asked.
Randomised controlled trials answer questions about efficacy and can support causal claims, because randomisation deals with confounding you have not thought of. Cohort studies follow exposure forward and can establish temporality but cannot fully exclude confounding. Case-control studies are efficient for rare outcomes and are vulnerable to recall and selection bias. Cross-sectional studies measure prevalence and cannot establish temporality at all, which means they cannot support a causal claim, however tempting.
The exam constructs questions in which the design cannot bear the weight of the conclusion, and asks whether you noticed.
Within trials, ask the specific questions: was randomisation genuine and was allocation concealed, was blinding possible and was it done, was the analysis by intention to treat, and was the follow-up adequate and were losses accounted for.
Step three: what could have gone wrong?
Now interrogate the study for the systematic errors, and know their names because the exam uses them.
Selection bias, where the people studied are not representative of those the conclusion is applied to.
Information and measurement bias, including recall bias in case-control studies and observer bias where blinding failed.
Attrition bias, where those lost to follow-up differ systematically from those who remained, which is common in psychiatric trials and important.
Confounding, where a third variable explains the apparent association, and where the question is not merely whether the authors adjusted for it but whether they could have known to.
Publication bias, which belongs to the appraisal of a body of literature rather than a single study, and which is examined in the context of meta-analysis.
For each, do not merely name it. Say what direction it would push the result in, because the exam asks that.
Step four: what does the effect actually mean?
Here is where the statistics live, and the statistics are finite.
You need to be able to interpret and distinguish: relative risk and odds ratios and when the two diverge, absolute risk reduction and why it matters more clinically than the relative version, number needed to treat and number needed to harm, confidence intervals and what it means when they cross the line of no effect, p-values and what they do and emphatically do not tell you, and the concepts of statistical power and type one and type two error.
Add the diagnostic test statistics, which recur every diet: sensitivity, specificity, positive and negative predictive value, and the crucial fact that the predictive values depend on prevalence while sensitivity and specificity do not.
That is close to the whole of it. It is a fortnight of short sessions, and it never stops being worth marks.
Step five: does the conclusion follow?
This is the step the exam loves and the one candidates skip, having exhausted themselves on the statistics.
Read what the authors concluded, and ask whether the data support it. The recurring failures are consistent and recognisable.
Claiming causation from an observational design. Extrapolating beyond the population studied, which in psychiatry frequently means claiming an effect in a group who were explicitly excluded from the trial. Emphasising a statistically significant result whose effect size is clinically trivial. Emphasising a subgroup finding that was not pre-specified. Ignoring a wide confidence interval that includes both benefit and harm. Concluding no effect from a study that was underpowered to detect one, which is not the same as demonstrating absence of effect.
That last distinction, between evidence of absence and absence of evidence, is examined repeatedly and misunderstood widely.
Practise on studies, not on textbooks
The final and most important point about how to prepare.
Critical review is a procedure, and procedures are learned by executing them. Reading a chapter about bias produces recognition of the word and no ability to spot it in a paper you have never seen.
So practise on studies. Take an abstract, run the five steps, and write down your verdict before you read anyone else's. Do it with practice questions. Do it with papers from your journal club. The volume required is modest, and it is the doing rather than the reading that builds the skill.
Twenty minutes, twice a week, from the beginning of your preparation, will comfortably secure fifty marks that most of your competitors will be scrambling for in the final fortnight.
Where iatroX fits
iatroX's MRCPsych Paper B bank tracks critical review as its own domain rather than folding it into a clinical percentage, which is the separation a fifty-mark methodological component demands, and its questions present studies to be appraised rather than definitions to be recalled, which is how the exam tests it. Missed questions can be opened in the Socratic Tutor, which asks you to work the steps before it explains and names which of the five you skipped, and spaced repetition holds the statistical concepts that are pure recall and decay fastest. Try it with free sample questions at iatroX. For running this alongside the clinical two-thirds, see parallel revision tracks.
Frequently asked questions
How much of MRCPsych Paper B is critical review? Roughly a third of the paper, about 50 marks. It is a methodological skill rather than a psychiatric one, and because it is finite and procedural it is the best return on revision time in the paper.
What are the five steps for appraising a study? State the research question, name the design and ask whether it can answer that question, identify the bias and confounding and their direction, interpret the effect size and its uncertainty, and finally judge whether the authors' conclusion actually follows from their data.
Where do candidates lose the most critical review marks? At the final step. The recurring failures are claiming causation from observational data, extrapolating beyond the population studied, emphasising a statistically significant but clinically trivial effect, and concluding no effect from an underpowered study.
How should I practise critical review? On studies, not on textbooks. It is a procedure, and procedures are learned by executing them. Twenty minutes twice a week applying the five steps to an abstract or a practice question, from the start of your preparation, is enough.