■ SAFE RISK ■ Healthcare
No. Machine tools are getting genuinely good at reading scans and matching mutations to trials, and none of that touches the part of the job where you sit with two parents and change their lives.
“AI helps with treatment plans. But telling parents their child has cancer needs a human heart.”
Our AI replacement risk score — how we score jobs
The specialty combines dense technical work with sustained relationship. A paediatric oncologist reviews pathology and molecular profiling, calculates chemotherapy doses by body surface area with narrow margins, enrols patients on cooperative group trial protocols, manages febrile neutropenia at 2am, and runs tumour boards with surgeons, radiation oncologists and radiologists. Because most childhood cancers are treated on trials, protocol adherence and deviation decisions are constant. Alongside it runs longitudinal care: the same family for years, through relapse, survivorship and late effects, or through palliative transition and bereavement.
Automation is advancing on the analytical side. Imaging models assist tumour detection and response measurement, digital pathology aids classification, genomic pipelines match variants to targeted therapies and trial eligibility far faster than manual review, and dosing calculators and toxicity-prediction models are in use. Ambient documentation is drafting clinic notes. These reduce cognitive load and catch things humans miss — meaningful in a field where rare tumours mean nobody has seen many cases.
The resistant core is thick. Communication is a clinical skill here, not a courtesy: breaking a diagnosis, negotiating goals of care with parents who disagree with each other, explaining a phase I trial without offering false hope, and knowing when a child has had enough. Rarity also limits data — many paediatric cancers have too few cases to train reliable models. Physical procedures, from lumbar punctures to bone marrow aspirates, are hands. Hence our score of 7. And the relationship is longitudinal in a way that changes the clinical work. You know this family across three years of protocol, relapse and survivorship, which shapes how you present each decision — a context window no consultation model has, and one that patients explicitly value.
Automatability: our editorial assessment of current and near-term AI capability
Resilient for the foreseeable future. Diagnostic and trial-matching support will be standard across major centres within this decade and will improve rare-tumour decision-making, which is exactly where human experience is thinnest. Paediatric oncology remains a workforce-shortage specialty in most countries. Expect augmented decision-making, less documentation burden, and no reduction in the number of oncologists families need to see.
It can assist substantially. Imaging and digital pathology models help detect and classify tumours, and genomic pipelines identify actionable mutations and trial eligibility much faster than manual review. The limitation is data: paediatric cancers are rare and heterogeneous, so training sets are small and model performance is uneven across tumour types. Diagnosis remains a multidisciplinary human decision informed by these tools.
Partly, and it already is protocol-driven. Most paediatric cancer treatment follows cooperative group protocols, and software handles dosing calculations, toxicity monitoring and eligibility checks well. What requires an oncologist is the deviation: the child whose kidney function will not tolerate the standard regimen, the family declining a component, the relapse with no protocol. Judgement lives in the exceptions, and in paediatric oncology there are many.
Very. It is a workforce-shortage specialty in most health systems, with long training, high emotional demand and modest compensation relative to adult subspecialties. Demand is stable and survivorship care is growing as cure rates rise, creating decades-long follow-up needs. Nothing about automation reduces the number of paediatric oncologists required; the constraint is how many people choose the field.
Mostly by removing friction. Ambient scribes draft clinic notes, trial-matching tools surface options that used to require manual chart review, molecular boards run on faster pipelines, and toxicity models flag risk earlier. That returns time to the clinic room. It also raises the standard: with better trial matching available, missing an eligible study becomes harder to excuse.