Artificial intelligence and clinical decision support in paediatrics

A registrar on a night shift sees a deterioration score flash on a ward monitor for a child who looks well, and a drug-interaction alert pop up for a prescription the team has used safely for years, and a retinal image flagged by an algorithm for a premature baby in the nursery. Each is a clinical decision support tool, each offers a number or a flag, and each can either help or harm the child depending on whether the team can read it correctly. The registrar who treats every score as a fact will be caught; the registrar who knows how the model was built, whether it was validated in a population like theirs, and where it fails will act safely. This page teaches the four classes of paediatric AI, how a machine-learning model is trained and where it fails, the validation ladder that separates a deployable tool from a hazard, and the four safeguards — clinician over-read, a written escalation plan, alert stewardship and ongoing monitoring — that hold the clinician accountable for every output. [2] [7]

Overview & Definition

Artificial intelligence is a computer system performing a task that would otherwise require human intelligence — recognising a pattern in an image, predicting an outcome from a set of observations, or drafting a summary of a note. The term is broad and is often used loosely, so the candidate must pin it down. Machine learning is the branch of AI in which the system learns the relationship between inputs and an outcome from historical data, rather than being explicitly programmed with rules. Deep learning is a machine-learning method built on artificial neural networks with many layers, and it is the engine behind most image-interpretation tools. Clinical decision support is the delivery of knowledge and patient-specific information to a clinician at the point of care — and it may be rule-based, where a human wrote the if-then logic, or model-based, where a machine-learning model generates the output. The candidate who conflates a rule-based alert with a trained prediction model loses the marks, because the two are validated, governed and trusted differently. [7] [10]

The defining principle of paediatric AI is augmented intelligence — the framing, adopted by the American Medical Association and the World Health Organization, that AI is designed to enhance clinical judgement rather than replace it. The tool extends the clinician; it does not take the decision. The second principle is that children are under-represented in the data that train most models, and a child changes physiologically across age bands in a way an adult does not, so age-stratified external validation is a non-negotiable gate. The third principle is accountability: the clinician who acts on the output remains responsible for it, and a tool that cannot be over-read, questioned or refused is a tool that cannot be deployed safely. [10]

References12Show ledgerHide ledger

1. [1]Le S Pediatric Severe Sepsis Prediction Using Machine Learning Front Pediatr, 2019.PMID 31681711
2. [2]Mayampurath A Development and External Validation of a Machine Learning Model for Prediction of Potential Transfer to the PICU Pediatr Crit Care Med, 2022.PMID 35446816
3. [3]Kausch SL Cardiorespiratory signature of neonatal sepsis: development and validation of prediction models in 3 NICUs Pediatr Res, 2023.PMID 36593281
4. [4]Choi JW Deep Learning-Assisted Diagnosis of Pediatric Skull Fractures on Plain Radiographs Korean J Radiol, 2022.PMID 35029078
5. [5]Taylor S Monitoring Disease Progression With a Quantitative Severity Scale for Retinopathy of Prematurity Using Deep Learning JAMA Ophthalmol, 2019.PMID 31268518
6. [6]Young BK Efficacy of Smartphone-Based Telescreening for Retinopathy of Prematurity With and Without Artificial Intelligence in India JAMA Ophthalmol, 2023.PMID 37166816
7. [7]Chaparro JD Clinical Decision Support Stewardship: Best Practices and Techniques to Monitor and Improve Interruptive Alerts Appl Clin Inform, 2022.PMID 35613913
8. [8]Fallon A Addressing Alert Fatigue by Replacing a Burdensome Interruptive Alert with Passive Clinical Decision Support Appl Clin Inform, 2024.PMID 38086417
9. [9]Simpao AF Optimization of drug-drug interaction alert rules in a pediatric hospital's electronic health record system using a visual analytics dashboard J Am Med Inform Assoc, 2015.PMID 25318641
10. [10]Liu X Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension Nature Medicine, 2020.PMID 32908283
11. [11]Bianco A Use of machine learning in pediatric surgical clinical prediction tools: A systematic review J Pediatr Surg, 2023.PMID 36804103
12. [12]Abady E Artificial Intelligence-Driven Triage in Pediatric Emergency Departments: Accuracy, Bias, and Impact on Clinical Outcomes: A Narrative Review Sage Open Pediatr, 2026.PMID 42137483