Open questions across the corpus

This page records cross-paper disagreements that the corpus does not resolve. Each bullet names the conflicting papers explicitly and the point of contention. New theme-synthesizer runs append to this list; conflicts are recorded, not resolved (see schema/CLAUDE.md §“Hard rules” #6).

Radiation biology

  • CT radiomic signature generalizability in head-and-neck cancer. PMID:24892406 (Aerts 2014) vs PMID:37397861 (Kazmierski 2023). Aerts reports a fixed 4-feature NSCLC-trained signature generalizes without retraining to two independent HNSCC cohorts (CI=0.69 in both) and retains performance in HPV-negative HNSCC. Kazmierski, benchmarking on RADCURE (n=2,552) with three external HNSCC cohorts (n=873), reports that no radiomics-only model beat any EMR-only model, the best deep-radiomics model added to EMR dropped AUROC vs EMR+volume, and external performance degraded on 2 of 3 sites with significant HPV/disease-site/outcome-prevalence distribution shift. Point of contention: whether HNSCC CT radiomics generalizes across institutions without retraining.
  • Added value of image features beyond tumor volume + EMR in HNSCC prognosis. PMID:24892406 (Aerts 2014) reports signature+volume beats volume alone in all NSCLC/HNSCC validation cohorts. PMID:37397861 (Kazmierski 2023) reports no radiomics-only model beat any EMR-only model on RADCURE, replacing tumor volume with deep CNN features lowered AUROC from 0.823 to 0.766, and the top model only modestly beat volume alone on MDACC external validation. Point of contention: whether volume-independent image information adds meaningful prognostic value on top of EMR+volume in definitive-RT HNSCC.
  • PD-L1 CPS as a predictor of benefit with RT + PD-1 blockade in HNSCC. PMID:38780927 (Saba 2024) reports PD-L1 CPS (22C3) <20 vs ≥20 does not stratify PFS (P=0.86) or OS (P=0.74) in IMRT reirradiation + nivolumab, explicitly divergent from recurrent/ metastatic HNSCC with single-agent PD-1 blockade. Point of contention: whether PD-L1 CPS retains predictive value when PD-1 blockade is combined with radiation in a previously irradiated field.
  • Direction of the early peripheral PD-1+Ki-67+ CD4+ T-cell surge under PD-1 blockade. PMID:38780927 (Saba 2024) reports a ≥1.5-fold surge at week 2–4 trends toward worse PFS (HR 2.09; median 16.0 vs 27.1 months) in HNSCC reirradiation + nivolumab, opposite in direction to prior reports in lung cancer and melanoma under PD-1 blockade alone (external literature cited by the paper, not present as a primary paper in this corpus). Point of contention: whether the peripheral PD-1+Ki-67+ CD4+ pharmacodynamic biomarker has opposite sign in RT-combined immunotherapy vs immunotherapy alone.
  • Ketogenic diet as a radiosensitizer vs radioresistance promoter. PMID:41941260 (ROBIN consortium, OligoMET Project 2) reports that ketogenic diets unexpectedly promoted radio-resistance in preclinical prostate-cancer models, while low-fat and calorie-restricted regimens enhanced radiation response. The paper explicitly flags this as contrary to prior expectation in the preclinical literature. Point of contention: the direction of the ketogenic-diet effect on RT response.

This page was processed by crosslinker on 2026-05-04.