Multimodal Artificial Intelligence System for Risk-Adapted Cancer Survivorship Surveillance: A Multicenter Target Trial Emulation.
AI can identify which treated nasopharyngeal cancer patients truly need intensive monitoring versus those unlikely to relapse, cutting unnecessary clinic visits by over 90% for low-risk survivors.
Using 2,148 nasopharyngeal carcinoma patients across five centers, this study first validated via target trial emulation that treatment de-intensification (omitting concurrent chemotherapy for stage II NPC) achieves comparable survival, then developed a Transformer AI model with near-perfect discrimination (AUC 0.986 external validation) to predict individualized failure timing and tailor follow-up schedules. The resulting risk-adapted strategy reduces surveillance visits by >90% for failure-free patients while concentrating monitoring resources on high-risk individuals, offering a scalable solution to the growing burden of cancer survivorship care.
What the study was
- Study design
- Multicenter target trial emulation with prospective AI validation
- Population
- 2,148 patients with stage II nasopharyngeal carcinoma across 5 centers
- Sample size
- 2148
- Category
- Diagnostics
- Maturity
- Validated
- Journal
- International journal of radiation oncology, biology, physics
Why it surfaced
Well-validated (5-center, n=2148) multicenter AI framework for cancer survivorship surveillance with near-perfect AUC and >90% reduction in unnecessary follow-up visits; target trial emulation methodology strengthens causal inference; generalizable framework complementing existing NCCN guidelines.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.