Association between dose to cardiac structures and overall survival: A multivariable analysis in a large, multi-institutional database of stage III NSCLC patients with external validation

Inconsistencies in identifying dose-limiting cardiovascular substructures for treating stage III non-small cell lung cancer (NSCLC) have hindered the implementation of cardiac sparing treatment planning guidelines. This study aims to address these inconsistencies by performing a multivariable survival analysis with overall survival as the endpoint using a large, multinational database, followed by external validation. Clinical and dosimetric parameters from 1587 stage III NSCLC patients treated at five institutes were analyzed. The whole heart, four cardiac chambers, great vessels and their combinations were considered. The dataset was divided into a training set (four institutes) and a test set (one institute). The optimal parameter set was identified through cross-validation, and the resulting multivariable Cox regression model was externally validated using the test set. Adjusted hazard ratios (aHRs) for all cardiovascular parameters were evaluated. The strongest associations were found for low Dx% parameters. However, their incremental contribution to model performance, compared to clinical and lung dosimetric parameters only, was low, with small effect sizes. Specifically, the cardiovascular parameter identified by parameter selection was Left Side D5% (aHR: 1.007 Gy-1, 95 % CI: 1.004 - 1.010 Gy-1, p < 0.0001), which provided a slight improvement in model concordance index of 0.0062 (95 % CI: 0.0000-0.0127) in the training set and 0.0037 (95 % CI: -0.0200-0.0280) in the test set. Although significant associations between cardiovascular parameters and survival were found, their small effect sizes should be considered when prioritizing cardiac sparing in stage III NSCLC treatment.