Read a study abstract as an argument, not as a collection of statistical keywords. For MRCPsych Paper B revision, identify the question, reconstruct the comparison, calculate what the numbers permit and decide how far the result can travel. Reading the explanation first removes the opportunity to discover which of those steps you cannot yet perform.
This article is published by iatroX and discusses MRCPsych Mentor, SPMM and iatroX as learning resources. The abstract below is entirely fictional, created for this article in September 2026. Its participants, intervention, results and numbers are not evidence about any real treatment.
The fictional abstract
An imaginary research team recruits 200 adult volunteers from two outpatient clinics to test a new educational support programme alongside usual care. Participants are randomly allocated equally between the programme and usual care alone. The primary outcome is a pre-specified symptom-improvement criterion at the end of follow-up.
The abstract reports improvement in 60 of 100 programme participants and 45 of 100 control participants. It describes the programme as "effective for all adults with the condition". The abstract does not state the allocation-concealment method, whether outcome assessors knew the assignments, how volunteers differed from non-participants, or whether harms and burdens were measured.
For the calculation exercise, assume outcome data are available for every randomised participant. Later, we will deliberately remove that assumption. No other numerical result should be imagined: there is no supplied confidence interval, significance test, cost estimate or longer-term outcome.
First identify the actual question
Write a neutral version: among the adults recruited from these clinics, did adding this educational programme change the specified short-term improvement outcome compared with usual care alone? That is narrower than asking whether the intervention works for every adult or improves every clinically important outcome.
Separate the intervention from the background care. The comparison does not establish what would happen if the programme replaced usual care. Equally, an effect on the stated outcome does not prove an effect on admission, employment, long-term relapse or quality of life unless those were actually assessed.
This is an original appraisal exercise rather than a report of an existing trial. Its deliberately overbroad conclusion is there to test whether the reader challenges the claim rather than simply accepting the authors' wording.
Calculate the result without changing its meaning
Using the fictional numbers, the improvement proportions are 60% and 45%. The absolute difference is 15 percentage points. The relative ratio of improvement proportions is 60 divided by 45, approximately 1.33. These calculations describe the supplied data; they do not establish precision, causality beyond the design's limitations, or applicability elsewhere.
Calling the absolute difference a "15% relative improvement" would mix two different quantities. Calling the ratio a reduction in risk would also change the direction and meaning of the outcome: this example measures improvement, not an adverse event.
Do not invent statistical significance. A point estimate without its uncertainty is an incomplete account. In an examination, use the supplied interval or test result when provided; in a real appraisal, consult the full report and methods rather than manufacturing the missing analysis.
What randomisation does not answer by itself
The word randomised does not tell you whether allocation was concealed before enrolment, whether missing outcomes differed between groups or whether assessors could have been influenced by knowing the intervention. It also does not guarantee that volunteers from these clinics represent every population named in the conclusion.
The CONSORT reporting guidance, checked on 27 September 2026, provides an established framework for examining how randomised trials are reported. Use reporting questions to locate missing information, not to assume that an unreported method was necessarily absent.
Write "not stated in the abstract" where appropriate. That phrase preserves a distinction between inadequate reporting and demonstrated methodological failure. A full paper could resolve the uncertainty; a confident guess cannot.
Change the denominator and reconsider the answer
Now imagine that the abstract's percentages were calculated only among participants who returned for follow-up, with different numbers missing from each group. Do not reuse the earlier calculations as if the underlying dataset were unchanged.
The next task is to establish the randomised denominators, the analysed denominators, reasons for missingness and how missing outcomes were handled. An intention-to-treat label does not remove the need to understand those details. This changed version is intentionally incomplete, so the correct exercise response is to identify what is needed before recalculating.
That distinction separates an arithmetic error from an appraisal error. A learner may divide correctly but use the wrong population. Another may identify the correct comparison but calculate the difference incorrectly. They need different revision tasks.
A one-page appraisal worksheet
Complete this before revealing a question-bank explanation. Keep each answer short enough to expose uncertainty rather than bury it.
| Question | Entry for the fictional study |
|---|---|
| Population | Volunteers from the stated clinics, not every adult automatically |
| Intervention and comparator | Added programme plus usual care versus usual care |
| Outcome | The defined improvement measure at the stated follow-up |
| Effect | Absolute difference and relative measure, labelled separately |
| Precision | Not supplied in this abbreviated exercise |
| Validity questions | Concealment, assessment, missingness and other relevant methods |
| Applicability | Whether the patient and setting resemble those studied |
| Claim to reject | A conclusion extending beyond the population or outcomes assessed |
Add a final question of your own: what result or methodological detail would change your interpretation most? That makes the worksheet a guide to further reading rather than a checklist completed for its own sake.
Read the protocol before accepting the headline
For the original complete-data exercise above, keep the reported 60 of 100 and 45 of 100 improvement results. Before moving further, identify the prespecified primary outcome, its time point, the planned analysis population and any changes made after recruitment. A favourable secondary outcome cannot silently replace an unfavourable primary one.
The CONSORT 2025 statement, published in April 2025 and checked on 27 September 2026, provides updated reporting guidance for randomised trials. Its purpose here is to identify the information needed for appraisal, not to assign a validity score from the number of boxes an abstract appears to tick.
Separate randomisation, concealment and blinding
Random sequence generation determines the allocation sequence. Allocation concealment protects the forthcoming assignment before enrolment. Blinding concerns knowledge of assignment after it occurs. A study can perform one of these well and another poorly.
In an original methods variant, an unpredictable computer-generated sequence is kept on an open list visible to the recruiting clinician. The sequence may be random, yet foreknowledge can influence who enters each group. In a different variant, central concealed allocation is used but an assessor knows the intervention when rating a subjective outcome. That raises a different concern.
Do not say that randomisation guarantees identical groups. It balances prognostic factors in expectation under the design, not necessarily in every realised sample. Cochrane's randomised-trial bias guidance distinguishes these mechanisms and the different effects an analysis may seek to estimate.
Follow participants through a fully specified missing-data variant
The following is a new fictional version, not extra hidden information about the original abstract. It deliberately replaces the original complete-outcome assumption. Each group still has 100 randomised participants, but follow-up is incomplete:
200 participants randomised
|
-----------------
| |
Programme: 100 Comparator: 100
| |
Outcome known: 80 Outcome known: 90
Outcome missing:20 Outcome missing:10
| |
Improved: 48 Improved: 40
Not improved: 32 Not improved: 50
Among observed outcomes, improvement proportions are 48 / 80 = 60% and 40 / 90 = 44.4%, an observed-case difference of approximately 15.6 percentage points. These denominators are not the randomised denominators.
Now perform a deliberately extreme sensitivity exercise. If all missing programme participants did not improve and all missing comparator participants improved, the proportions become 48% and 50%, a difference of -2 percentage points. Reverse those assumptions and the proportions become 68% and 40%, a difference of 28 percentage points.
These are mathematical bounds for the stated binary outcomes, not plausible forecasts or a recommended primary analysis. Their purpose is to show how much the conclusion can depend on unobserved outcomes. Reasons for missingness, available predictors and the planned method determine which more informative analyses may be appropriate.
| Fictional assumption | Programme improvement | Comparator improvement | Difference |
|---|---|---|---|
| Observed cases only | 48/80 = 60% | 40/90 = 44.4% | About 15.6 percentage points |
| Missing outcomes least favourable to programme | 48/100 = 48% | 50/100 = 50% | -2 percentage points |
| Missing outcomes most favourable to programme | 68/100 = 68% | 40/100 = 40% | 28 percentage points |
Calling every missing outcome a failure in both groups is another assumption, not a neutral act. It would produce 48% versus 40% in this variant, but the equal label does not establish that missingness has the same meaning in the two groups.
Compare analysis populations without trusting the label alone
An intention-to-treat effect concerns assignment to the intervention. A per-protocol effect concerns adherence to the intervention under a specified definition. Simply excluding people who did not adhere does not automatically provide an unbiased per-protocol effect, because adherence can be related to prognosis.
An intention-to-treat label also does not create measurements that were never collected. Ask whether all randomised participants were included, how missing values were handled and whether the method corresponds to the intended question. A heading in a results table is not enough.
For an educational programme, participants who attend every session may differ from those who cannot attend because of illness, work or caring responsibilities. A completer-only analysis can therefore change both the population and the question. The clinically useful issue is not which label sounds stronger, but which effect was estimated and under what assumptions.
Inspect subgroup, multiplicity and non-inferiority claims
Imagine that a fictional study reports benefit in one age subgroup but not another. That does not by itself demonstrate that age modifies the effect. Compare the subgroup effects directly using an appropriate interaction analysis, and check whether the subgroup question was prespecified.
Similarly, searching many outcomes and time points for a favourable p-value changes the interpretation of an apparently isolated positive result. Report what was planned and what was exploratory. An exploratory finding may be worth investigating without being treated as a confirmed treatment recommendation.
For non-inferiority, the study must specify an acceptable margin, its justification and the direction of the comparison. A non-significant superiority result does not establish non-inferiority. In a fictional adverse-outcome comparison, an interval that allows clinically unacceptable additional harm cannot support a reassuring conclusion merely because it also includes no difference.
Match claims to the evidence actually available
| Claim in a fictional report | What must be established before accepting it |
|---|---|
| Effective for every adult | Applicability beyond the recruited clinics and outcome actually studied |
| Randomisation removed all bias | Concealment, measurement, missingness and reporting must still be assessed |
| All participants were analysed | The analysis population and handling of missing outcomes must be explicit |
| The effect is larger in a subgroup | An appropriate comparison of subgroup effects, not separate p-values alone |
| No significant difference means equivalence | A suitable equivalence or non-inferiority design and margin |
For the missing-data variant, a defensible conclusion is that the observed participants show a higher improvement proportion with the programme, but the direction and size of the full-population comparison depend on the unobserved outcomes. The next reading task is to examine the missing-data methods and reasons, not to invent a decisive p-value.
Choosing the next learning activity
As checked on 13 September 2026, MRCPsych Mentor, SPMM and BMJ's Paper B resource advertise relevant examination preparation. Their existence does not establish identical teaching methods or that one is universally preferable.
Use calculation teaching for arithmetic errors, a worked methods explanation for design confusion, and further abstracts when the difficulty is applying familiar concepts. Do not paste a provider's paid abstract or explanation into an AI tool; use an original summary or an openly licensed source with appropriate attribution.
Per the September 2026 iatroX specification, Tutor follow-ups can investigate the reasoning behind an attempted question. The combined banks, Tutor, planner, simulations and CPD subscription costs £99 upfront annually, equivalent to £8.25 monthly billed annually, or £29 a month. Choose it for a defined learning need, not as a guarantee that generated statistical reasoning is correct.
Frequently asked questions
Are the results in this article from a real trial?
No: the study and every numerical result are fictional teaching material. They must not be cited as evidence for a clinical intervention.
Is a missing method in an abstract proof of a flawed study?
No: record that the abstract does not specify it and consult the full report. Missing reporting and a demonstrated methodological problem are different findings.
Should I calculate before reading the explanation?
Yes, for this practice method: commit to your own comparison, denominator and calculation first. That makes it possible to identify the exact step that needs revision.