MRCPsych Paper B Question Bank

A parallel group RCT allocates participants to one intervention for the entire study duration. In this design, each participant is randomised to receive either the experimental treatment or the comparator, and they remain in that group throughout. This is the most common RCT design and is particularly suitable when carryover effects would be problematic, as is often the case with antipsychotic medications where washout periods carry clinical risk.

A crossover trial (A) would require participants to receive both treatments sequentially. A factorial design (C) tests two or more interventions simultaneously. A cluster RCT (D) randomises groups rather than individuals. A pragmatic trial (E) describes the trial's orientation rather than its structural design.

Parallel group designs are preferred in schizophrenia research because the illness has a variable course, making crossover designs problematic. The CATIE trial is a well-known example of a parallel group RCT in schizophrenia research. Sample size calculations must account for between-group variability, which is typically larger than within-subject variability seen in crossover designs.

Remember: Parallel group RCTs are the gold standard design when carryover effects, disease progression, or ethical concerns preclude crossover designs.

A significant treatment-by-period interaction in a crossover trial indicates the presence of a carryover effect. This means the effect of the first treatment persists into the second treatment period, contaminating the results. In this GAD study, a 2-week washout may be insufficient given sertraline's half-life and the time required for neuroadaptive changes to reverse.

Carryover effects are a fundamental threat to the validity of crossover designs. They can be pharmacological (drug still present), pharmacodynamic (receptor changes persisting), or psychological (learned coping strategies). When detected, the standard approach is to analyse only first-period data, effectively converting the study to a parallel group comparison with reduced statistical power.

Inadequate randomisation (A) would produce baseline imbalances. A washout that is too long (C) would reduce carryover effects, not cause them. Insufficient sample size (D) affects power but does not cause interactions. Invalid outcome measures (E) would affect both periods equally.

The AB/BA crossover design assumes no carryover, stable disease, and that treatment effects are reversible. SSRIs pose particular challenges for crossover designs due to discontinuation effects and slow onset of action.

Remember: When a significant carryover effect is detected in a crossover trial, only first-period data should be analysed, sacrificing the efficiency advantage of the crossover design.

Cluster randomisation is primarily used to avoid contamination between groups when the intervention is delivered at a group level. If individual GPs within the same practice were randomised, trained GPs could share knowledge with untrained colleagues, diluting the treatment effect. By randomising entire practices, this contamination is prevented.

Cluster RCTs actually require larger sample sizes (A is incorrect) because observations within clusters are correlated, quantified by the intracluster correlation coefficient (ICC). The design does not eliminate selection bias (B) - in fact, it can introduce recruitment bias if clinicians know their practice allocation before recruiting patients. Generalisability (D) and blinding (E) are not inherent advantages of cluster designs.

The design effect in cluster RCTs is calculated as 1 + (m-1) × ICC, where m is the average cluster size. This inflation factor must be applied to standard sample size calculations. The UK Medical Research Council has published specific guidance on cluster RCTs, emphasising the need to analyse at the cluster level or use multilevel modelling to account for the hierarchical data structure.

Remember: Cluster randomisation is chosen to prevent contamination when interventions operate at a group level, but always requires inflation of sample size to account for intracluster correlation.