Cardiac-dose model validated non-cancer death risk in one lung SBRT cohort

KEY POINTS

• The original model was developed in 803 predominantly inoperable patients with stage I–II NSCLC treated using SBRT. External validation used independent cohorts of 221 patients from UMCG and 217 from Memorial Sloan Kettering.
• The model combined age, performance status, lung function, baseline cardiac disease, left-atrial maximum dose, and the dose received by 90% of the superior vena cava to predict non-cancer death after SBRT.
• Two-year overall survival was 68% in the derivation cohort, 71% at UMCG, and 66% at Memorial Sloan Kettering. Median follow-up was approximately 35, 23, and 110 months, respectively.
• Model discrimination was preserved in both external cohorts. Prognostic-index slopes were 1.15 at UMCG and 0.97 at Memorial Sloan Kettering, neither significantly different from the expected value of one; Harrell c-indices were 0.670, 0.696, and 0.627 across the three cohorts.
• Calibration was good in the UMCG cohort, where non-cancer death was defined similarly to the derivation study. The observed-versus-expected analysis produced an R² of 0.946, and the poorest-risk group had approximately 3.8 times the non-cancer death risk of the reference group.
• The model did not calibrate adequately in the Memorial Sloan Kettering cohort, where non-cancer death had to be approximated using a different definition. A newly derived overall-survival model using the same predictors subsequently validated in both external cohorts.
• In the nomogram, reducing left-atrial and superior-vena-cava dose by 36% lowered estimated 24-month non-cancer death from 18.2% to 17.3% for a representative high-dose patient. Baseline cardiac status was missing in both validation cohorts, lung function was unavailable at Memorial Sloan Kettering, and the observational analysis cannot prove that cardiac sparing improves survival.

CLINICAL TAKEAWAY

SOURCE

Radiotherapy and Oncology