Journal club epidemiology critique — teaching/CASC-style station · Psych CASC / OSCE

Prompt

You are the psychiatry registrar facilitating a 10-minute teaching station with a foundation doctor and a medical student. Station material is a one-page abstract: 'Cross-sectional household survey (n=1,200, response rate 48%). Lifetime depression by self-report checklist 28%. Cannabis use associated with depression OR 1.9 (95% CI 1.2–3.0). Authors conclude cannabis causes nearly one-third of depression (PAF 30%) and recommend universal school cannabis screening using a 2-item tool (Sn 80%, Sp 80%) to prevent depression.' Your job is to teach correct measure choice, bias threats, PAF assumptions, and why universal screening PPV will fail — without humiliating juniors or dismissing all observational evidence.

Station brief

Format. Teaching/communication station, approximately 8–10 minutes. Facilitate; do not monologue. [1]

Candidate instructions. Help the team correctly name the frequency measure, list bias threats, explain why PAF 30% is not proved, and show why universal school screening is epidemiologically weak at low prevalence. Close with a collaborative rewrite of the abstract’s conclusion. [2][3][4]

Candidate scenario

Abstract claims: lifetime self-report depression 28%; cannabis–depression OR 1.9; causal PAF 30%; recommend universal 2-item school screening (Sn 80%, Sp 80%) to prevent depression. Response rate 48%. Cross-sectional design. [1]

Marking domains

Identifies lifetime period-style burden estimate, not current point/12-month service need; notes self-report checklist limits [1]
Names non-response bias (48%), cross-sectional lack of temporality, possible confounding and recall/information bias [2]
Separates OR association from causal PAF; states Rockhill-style assumptions [3]
Explains low PPV / false-positive burden for school screening at low prevalence; mentions programme criteria beyond Sn/Sp [4][5]
Collaborative teaching tone; invites calculation intuition; avoids nihilism about all observational data
Bottom line: rewrite conclusions; do not implement universal screening from this abstract
Time management

Reveal assessor key

Open. "Before we argue about cannabis, what measure is 28% — and what question does it answer?" Board: lifetime self-report caseness ≠ current treated need. [1]

Threats. Response 48% → selection; cross-section → no temporality (depression may increase cannabis use); confounding (shared vulnerability, socioeconomic factors); checklist ≠ structured diagnosis. [2]

OR vs PAF. OR 1.9 is an association measure. PAF 30% requires causal RR, valid exposure prevalence, and minimal bias — not automatic from an OR in a weak design. [3]

Screening. Even “80/80” tests drown in false positives when prevalence is low; screening needs pathways, acceptability, and capacity (Wilson–Jungner tradition), not a 2-item app alone. [4][5]

Close. Collaborative rewrite: “Associated in a low-response cross-sectional survey; causal fraction unproven; screening not justified from these data. Next step: longitudinal designs, better measurement, STROBE-complete reporting.” Thank the team. [6][7]

References

[1]Grimes DA, Schulz KF An overview of clinical research: the lay of the land Lancet, 2002.PMID 11809203
[2]Grimes DA, Schulz KF Bias and causal associations in observational research Lancet, 2002.PMID 11812579
[3]Rockhill B, Newman B, Weinberg C Use and misuse of population attributable fractions Am J Public Health, 1998.PMID 9584027
[4]Altman DG, Bland JM Diagnostic tests 2: Predictive values BMJ, 1994.PMID 8038641
[5]Andermann A, Blancquaert I, Beauchamp S, Déry V Revisiting Wilson and Jungner in the genomic age Bull World Health Organ, 2008.PMID 18438522
[6]von Elm E, Altman DG, Egger M, et al. STROBE statement: guidelines for reporting observational studies Lancet, 2007.PMID 18064739
[7]Hill AB The environment and disease: association or causation? Proc R Soc Med, 1965.PMID 14283879

Station brief

Format. Teaching/communication station, approximately 8–10 minutes. Facilitate; do not monologue. [1]

Marking domains

Identifies lifetime period-style burden estimate, not current point/12-month service need; notes self-report checklist limits [1]

Names non-response bias (48%), cross-sectional lack of temporality, possible confounding and recall/information bias [2]

Separates OR association from causal PAF; states Rockhill-style assumptions [3]

Explains low PPV / false-positive burden for school screening at low prevalence; mentions programme criteria beyond Sn/Sp [4][5]

Collaborative teaching tone; invites calculation intuition; avoids nihilism about all observational data

Bottom line: rewrite conclusions; do not implement universal screening from this abstract

Time management

Reveal assessor key

Open. "Before we argue about cannabis, what measure is 28% — and what question does it answer?" Board: lifetime self-report caseness ≠ current treated need. [1]

OR vs PAF. OR 1.9 is an association measure. PAF 30% requires causal RR, valid exposure prevalence, and minimal bias — not automatic from an OR in a weak design. [3]

References

[1]Grimes DA, Schulz KF An overview of clinical research: the lay of the land Lancet, 2002.PMID 11809203
[2]Grimes DA, Schulz KF Bias and causal associations in observational research Lancet, 2002.PMID 11812579
[3]Rockhill B, Newman B, Weinberg C Use and misuse of population attributable fractions Am J Public Health, 1998.PMID 9584027
[4]Altman DG, Bland JM Diagnostic tests 2: Predictive values BMJ, 1994.PMID 8038641
[5]Andermann A, Blancquaert I, Beauchamp S, Déry V Revisiting Wilson and Jungner in the genomic age Bull World Health Organ, 2008.PMID 18438522
[6]von Elm E, Altman DG, Egger M, et al. STROBE statement: guidelines for reporting observational studies Lancet, 2007.PMID 18064739
[7]Hill AB The environment and disease: association or causation? Proc R Soc Med, 1965.PMID 14283879