Diagnostic test selection and Bayesian reasoning in paediatrics — branching viva

Station map

Branch A — High-risk neonate, reassuring marker

Examiner: A 14-day-old is poorly feeding and mottled. CRP is normal. The night team wants discharge because “the bloods are fine.” [7] [9]

Strong answer should include:

• Pre-test risk is high from age and physiology. [7] [9]
• Sensitivity/NPV language: a normal marker is not automatic rule-out. [1] [2]
• Post-test probability may remain above treatment or observation thresholds. [2] [4]
• Residual-risk plan: treat/observe, senior review, timed reassessment, family explanation. [4] [9]

Trap: equating one normal number with zero disease probability. [2] [9]

Branch B — Stable bronchiolitis and “just in case” imaging

Examiner: Classic bronchiolitis, stable. Registrar requests chest radiograph and full panel. [5] [8]

Strong answer should include:

• Clinical question already answered by history/examination for typical disease. [5]
• Routine radiograph often low value and can drive unnecessary treatment. [5] [8]
• Reopen testing if the question changes (focal signs, severe course, mismatch). [5]
• Stewardship without denying needed tests for true residual risk. [8]

Trap: protocol stacking as a substitute for Bayesian indication. [5] [8]

Branch C — Incidental adult-range laboratory flag

Examiner: Asymptomatic school-age child; adult reference interval flags a value high. [6]

Strong answer should include:

• Reinterpret with age-appropriate paediatric intervals. [6]
• Avoid cascade testing for a software-labelled “abnormal” that fits childhood physiology. [6] [8]
• Only investigate further if independent clinical concern exists. [8] [9]

Trap: treating laboratory flags as diagnoses. [6] [8]

Branch D — Appraisal probe

Examiner: A paper claims 99% sensitivity from ICU cases versus healthy controls. Can you use that number tonight? [10]

Strong answer should include:

• Spectrum bias and applicability concerns. [10]
• Need for STARD/QUADAS-style appraisal of selection, reference standard and flow. [10]
• Bedside use requires match to your patient spectrum and decision thresholds. [2] [3]

Closing synthesis the candidate should say

“I name the question and pre-test risk first. I order only discriminating tests. I update probability with the result. I always state residual risk, next action and safety-net.” [2] [4] [9]