Choosing and defending study designs in psychiatry (MEQ)

(A) CBT adjunct for relapse prevention. (i) Parallel-group RCT (ideally assessor-blind; full double-blind of therapy content often impossible) randomising remitted adults to CBT plus medication management versus medication management alone, with pre-specified 6-month relapse as primary outcome.[1][3] (ii) Main threats if weak: performance/detection bias (unblinded supportive contact and unblinded relapse ratings) and attrition with completer-only analysis.[3] (iii) CONSORT for trial reporting (flow, randomisation, outcomes).[3]

(B) Clozapine severe neutropenia incidence and predictors. (i) Prospective cohort of new clozapine starters with active haematological surveillance — exposure defined at inception, outcomes counted forward to estimate incidence and associations with baseline predictors.[2] A case-control design could study rare agranulocytosis associations but does not directly yield incidence without a defined source population and sampling fraction.[1] (ii) Main threats: loss to follow-up, incomplete blood monitoring data, and confounding of predictors (comorbidity, co-prescribing).[6][2] (iii) STROBE (cohort checklist).[4]

(C) Carer lived experience. (i) Qualitative design (e.g. semi-structured interviews with purposive sampling; thematic analysis or culturally adapted methodology with Indigenous governance).[5] (ii) Main threats: poor reflexivity, tokenistic sampling, analysis without credibility checks, extracting themes as if they were prevalence rates.[5] (iii) COREQ for interview/focus-group reporting.[5]

(iv) Why RCT is not default for B and C. Question B asks for incidence and predictors under real-world clozapine use, not experimental assignment of a harmful outcome; randomising people to clozapine solely to count neutropenia would be unethical and unnecessary when the exposure is already indicated clinically — a cohort matches the question.[2][1] Question C asks for meaning and experience, which quantitative trials do not answer; forcing an RCT would be a design–question mismatch.[5][1] Hierarchy rank does not override question fit or ethics.