Evaluating the concordance between AI-based Large Language Model recommendations and Virtual Molecular Tumor Board expert consensus for solid tumor cases in India

This prospective, blinded, cross-sectional study evaluates the concordance between treatment recommendations generated by large language model (LLM)–based decision-support tools and expert consensus from the Tamil Nadu Medical and Pediatric Oncologist Society (TAMPOS) Virtual Molecular Tumor Board (vMTB) for precision oncology cases in India. Consecutive de-identified solid tumor cases discussed at vMTB sessions (September–December 2025) will be analyzed. AI tools will undergo extraction validation, reproducibility testing (two-run protocol with escalation to a third run if inconsistent), and blinded concordance scoring across pathway, therapy, and evidence level dimensions. The primary hypothesis is that LLM systems will meet predefined concordance thresholds (~90% for high-evidence and ~40% for low-evidence recommendations), consistent with international benchmarking studies (Sunami et al., 2024).