AI in Oncology Treatment Planning

Oncology treatment planning combines diagnosis, staging, pathology, tumor markers, previous therapy, comorbidities, patient preferences and current evidence. AI systems may help search a large evidence base, summarize records, identify possible options or rank them for discussion. Their output is clinical decision support: it can structure a conversation or make a candidate treatment easier to notice, but it does not select the right regimen by itself. Even a well-matched guideline option may be unsuitable for an individual because of toxicity, contraindications, availability, goals of care or new findings not represented in the system. IBM Watson for Oncology is a useful historical example. In a study of 638 breast cancer cases from one Indian cancer center, the system’s treatment recommendations were considered concordant with the center’s tumor board in 93% of cases when the recommendation was rated either recommended or for consideration. The study measured agreement, not whether patients lived longer or experienced fewer harms. Other studies used different cancers, populations and definitions of concordance, so their percentages cannot be treated as a universal performance score. Agreement with a panel also does not prove that either recommendation is best for a particular patient. A safe planning workflow brings AI suggestions to a qualified tumor board, checks each option against current evidence and local treatment availability, and documents why the final plan fits the patient. FDA guidance on clinical decision-support software emphasizes that clinicians should be able to independently review the basis for a recommendation. Patients and clinicians can then discuss benefits, risks and preferences together. AI may help make complex information more navigable, but responsibility for treatment remains with the care team and patient.