To examine potential dose-response relationships with various non-small-cell lung cancer (NSCLC) SBRT fractionation regimens delivered with online CT-based image guidance.

505 tumors in 483 patients with clinical stage T1-T2N0 NSCLC were treated with SBRT using on-line cone-beam-CT-based image guidance at 5 institutions (1998-2010). Median maximum tumor dimension was 2.6 cm (range 0.9-8.5 cm). Dose fractionation prescription was according to each institution's protocol with the most common schedules of 18-20 GyX3, 12 GyX4, 12 GyX5, 12.5 GyX3, 7.5 GyX8 (median = 54 Gy, 3 fractions). Median prescription (Rx) BED10 = 132 Gy (50.4-180). Median values (Gy) of 3D planned doses for BED10 were GTV(min) = 164.1, GTV(mean) = 188.4, GTV(max) = 205.9, PTV(min) = 113.9, PTV D99 = 123.9, PTV(mean) = 164.7, PTV D1 = 197.3, PTV(max) = 210.7. Mean follow-up = 1.6 years.

26 cases (5%) had local recurrence (LR) for a 2-year rate of 6% and 3-year rate of 9%. All BED10 GTV&PTV endpoints were associated with LR as continuous variables on univariate analysis (p<0.05). Rx and PTV(mean) dose appeared to have the highest correlation with LR with area under ROC curve of 0.69 and 0.65 respectively and optimal cut points of 105 and 125 Gy, respectively. 2-year LR was 4% for PTV(mean)>125 vs 17% for <125 Gy (p<0.01) with sensitivity = 84% and specificity = 57% for predicting LR. 2-year LR for Rx BED10>105 was 4% vs 15% for <105 Gy (p<0.01). Longer treatment duration (⩾ 11 elapsed days) demonstrated a 2-year LR of 14% vs 4% for ⩽ 10 days (p<0.01). GTV size was associated with LR on univariate analysis as a continuous variable (p = 0.02) with 2-year LR = 3% for <2.7 cm vs 9% for ⩾ 2.7 cm (p = 0.03). BED10 (p = 0.01) and elapsed days during RT (p = 0.05) were independent predictors on multivariate analysis as continuous variables.

There is a substantial dose-response relationship for local control of NSCLC following image-guided SBRT with optimal PTV(mean) BED10>125 Gy. Shorter treatment duration was also associated with better local control in this dataset.