Abstract
BACKGROUND AND PURPOSE: A recent study of NSCLC patients showed small residual setup errors (shifts) in the direction of the heart following image-guidance were significantly related to overall survival. This study of the dosimetric effects of these residual shifts investigates the hypothesis that observed survival differences were related to a change in heart dose. MATERIALS AND METHODS: Accumulated doses including shifts for each fraction were determined for 475 NSCLC patients. Planning CTs and corresponding dose distributions were deformed to a reference. Image-based data-mining techniques were then applied to the difference between the planned and accumulated dose (Δdose) to determine where Δdose relates to 1-year survival. The significance of Δdose in the identified region was assessed using multivariable Cox analysis. The cohort was then split into octiles, based upon planned dose to the region, and multivariable Cox analysis performed for each sub-cohort to explore the dose response relationship. The identified dose threshold for damage was then tested in an independent validation cohort of 1482 NSCLC patients from the same institution. RESULTS: Permutation testing identified a small region in the heart base where Δdose significantly correlated with 1-year survival. Δdose in this region showed no correlation with common clinical variables, and was significant in multivariable Cox regression (p < 0.001, hazard ratio 1.221/Gy), with increasing change in dose from plan resulting in greater risk of death. Octile analysis revealed Δdose to be significant only in the 7th octile, planning dose 16.2-23.4 Gy, suggesting a steep dose-effect relation for heart damage in this range. Taking 16.2 Gy as a conservative threshold dose, this result was successfully validated, with a significant difference being seen between patients with a region dose above or below 16.2 Gy. CONCLUSIONS: This study suggests the relation between residual set-up errors and survival is explained by changes in cardiac dose, and identifies an area at the heart base where dose is correlated with survival. Our results suggest the dose threshold for cardiac damage is between 16.2 and 23.4 Gy in the base of the heart, which was validated in an independent cohort. However, the dose effect in other regions of the heart should also be investigated.