Artificial intelligence and clinical decision support in paediatrics — OSCE

Station brief (candidate)

• Manage the child whose deterioration score reads low but who looks unwell: state that the bedside assessment overrides the score, escalate to the higher acuity, and treat the child, because automation bias has caused missed deterioration. [2]
• Frame the drug-interaction alert override culture as alert fatigue and a patient-safety failure, and outline the stewardship response: measure the override rate, retire or convert the lowest-value alerts to passive display, and tune the thresholds with the prescribers. [7] [8] [9]
• Manage the deep-learning retinal image flag as a prompt to escalate rather than a diagnosis, arrange the formal ophthalmology review by telehealth where the ophthalmologist is off site, and hold the clinician over-read and accountability principle. [5] [6]

Setting

A general paediatric ward at night. The night-shift registrar and a senior nurse brief the candidate, who is the on-call general paediatric registrar. Nine minutes for the encounter, one minute for the examiner's marking. [2]

Encounter script

Nurse: "Doctor, the deterioration score on bed six reads low, but he looks dreadful to me — he's working harder to breathe, his hands are cool, and he's not responding like he was. The score says we're fine, but I'm not fine with him." [2]

Candidate (deterioration score): I would thank the nurse and act on the child, not the score. The bedside assessment overrides the model, so I escalate to the higher acuity now — I call my senior and the rapid-response team and begin the structured airway, breathing, circulation, disability and exposure assessment. The most dangerous AI failure is the reassuring number in a sick child, and automation bias has caused missed deterioration when a normal-looking number silenced a worried clinician. The score is a prompt to look again, not a verdict, and the nurse's concern is itself a deterioration signal. After the child is safe I would document the discrepancy and the over-ride so the event is auditable and the tool is reviewed. [2]

Registrar: "Separately, we override basically every drug-interaction alert that pops up. There are so many of them that none of us reads them anymore. Is that a problem?" [7]

Candidate (alert fatigue): Yes — that is alert fatigue, and it is a patient-safety failure, not an inconvenience. An alert overridden more than nine times in ten has lost its value, and the unmanageable burden drives an override culture in which the one actionable alert is lost in the noise. The stewardship response is to measure the override rate, retire or convert the lowest-value interruptive alerts to a passive display — the Fallon study replaced a burdensome interruptive alert with passive clinical decision support — and tune the thresholds with the prescribers, as the Simpao paediatric drug-interaction study did with a visual analytics dashboard. The Chaparro framework sets the best-practice cycle of monitoring and improving interruptive alerts. I would flag this to the clinical decision support stewardship team in the morning and treat the override rate as a signal, not a habit. [7] [8] [9]

Nurse: "And the deep-learning tool just flagged a retinal image on the premature baby in the nursery. There's no ophthalmologist on site tonight. What do we do?" [5] [6]

Candidate (retinal image flag): I treat the tool's flag as a prompt to escalate, not a diagnosis. The deep-learning evidence for retinopathy of prematurity is anchored by the Taylor quantitative severity scale and the Young smartphone-telescreening study, which brought AI-assisted screening to a setting where the ophthalmologist is the limiting factor. I assess the baby's gestational and postnatal risk factors, arrange the formal ophthalmology review by telehealth tonight since the ophthalmologist is off site, and ensure the image and the algorithm's output are over-read by the clinician responsible for the baby. The tool extends screening into a low-resource setting; it does not remove the over-read obligation, and we remain accountable for the follow-up. [5] [6]

Examiner marking domains

• Clinical reasoning (3): States that the bedside assessment overrides a disagreeing score and escalates the sick child; frames the override culture as alert fatigue and names the stewardship response; treats the retinal flag as a prompt to escalate with telehealth ophthalmology review rather than a diagnosis. [2] [7]
• Accountability and over-read (3): Holds the clinician over-read and accountability principle across all three scenarios — the score is a prompt not a verdict, the alert burden is a safety signal to steward, and the image flag is an assist to the reader not a replacement for the reader.
• Communication and shared decision (2): Acknowledges the team's concern, names the principles in plain registrar-level language, and converts each scenario into a concrete, documented plan the team can execute tonight.
• Safety-netting (2): Documents the discrepancy and the over-ride for audit, flags the alert fatigue to the stewardship team, and arranges the telehealth ophthalmology review with a clear follow-up pathway for the baby.