CT-derived spatial intratumoural heterogeneity and KRAS mutation status in a predominantly lung adenocarcinoma cohort: an exploratory study of association and incremental discrimination

Abstract Background To assess whether CT-derived spatial intratumoural heterogeneity (ITH) is associated with KRAS mutation status after accounting for tumour burden and clinical covariates, and whether it adds predictive information beyond clinical and volume-based reference models. Methods In this retrospective single-centre study, 271 candidate records from an intentionally KRAS-enriched cohort underwent final eligibility review; after excluding 71 patients, 200 were included (101 KRAS-positive, 99 KRAS-negative), and the original 140/60 train–test split was preserved without re-partitioning. Separate 2D and 3D ITHscore associations were assessed with adjustment for log2 tumour volume, T category, age, sex and smoking history, with restricted and overlap-weighted sensitivity analyses. Six logistic-regression reference models were fitted in the training cohort and compared using paired bootstrap confidence intervals in the previously examined internal-test partition. Results Tumour volume correlated strongly with both ITH scores (Spearman ρ = 0.658 for 2D and 0.679 for 3D). Fully adjusted odds ratios per SD increase were 3.84 (95% CI, 2.23–6.59) for the 2D score and 3.45 (95% CI, 1.98–6.00) for the 3D score. Within the overlapping tumour-volume range ( n = 102) the estimates attenuated to 2.39 (95% CI, 1.14–5.00) for 2D and 2.28 (95% CI, 0.999–5.22) for 3D, whereas under overlap weighting, which balanced the modelled covariates to absolute standardised mean differences below 0.001, they were 3.94 (95% CI, 2.53–10.91) and 3.07 (95% CI, 1.96–6.57). Attenuation occurred but the association did not disappear; the restricted cohorts were not matched, however, and residual imbalance in sex and smoking history persisted. The clinical-plus-volume model achieved an internal-test AUC of 0.844 (95% CI, 0.726–0.941), versus 0.833 (95% CI, 0.725–0.925) after adding both ITH scores; the paired ΔAUC was −0.011 (95% CI, −0.099 to 0.077). Conclusions Spatial ITH was associated with KRAS mutation status after adjustment for measured covariates. Estimates attenuated within the overlapping tumour-volume range (2D, 2.39; 3D, 2.28) but persisted under overlap weighting (2D, 3.94; 3D, 3.07), so independence from measured tumour burden was neither confirmed nor excluded. Incremental discrimination beyond clinical and volume-based references was not established. These exploratory findings do not establish molecular specificity or clinical utility and require validation in representative external cohorts.