Tumour shape features are not the prognostic factors they appear to be

Abstract

Introduction: Tumour shape features have shown promise as prognostic markers, linking irregular tumour shape with worse prognosis. This study investigates tumour sphericity as a predictor of overall survival (OS) in lung cancer patients. Importantly, we consider the alternative that sphericity is correlated with known prognostic factors. Methods: 487 lung cancer patients with a range of histological subtype were treated with curative-intent radiotherapy (55Gy over 20 fractions). These patients had a single gross tumour volume (GTV) delineated on planning CT. Sphericity was calculated as the ratio of the surface area of a sphere with equivalent volume to the GTV, to the surface area of the GTV. Relationships between sphericity and common patient and tumour characteristics were investigated. A multivariate Cox regression was performed to assess sphericity as prognostic for OS. A subset analysis was performed for 300 patients with N-stage N0 to remove potential influence of nodal involvement on sphericity. Results: Sphericity is significantly correlated with known prognostic factors: tumour volume, mean lung dose, N-stage, and T-stage. Sphericity is a significant predictor of OS in a univariate Cox regression, but not significant in multivariate analysis (Table 1). For N0 subset analysis, sphericity is higher on average (0.648 vs 0.620) and is no longer significant in univariate analysis (p=0.088, HR=0.267). Conclusion: Sphericity is heavily influenced by the GTV and correlates with known prognostic factors: volume, stage, and mean lung dose. This is not considered in many radiomic studies. In addition, this study emphasises the importance of distinction between primary tumour and nodes for radiomic analysis. Without separation, the inclusion of nodes within the GTV will bias analysis, leading to more irregular shapes for higher N-stage. Issues identified are likely relevant for other radiomic features. Future work will investigate independent characteristics descriptive of the tumour border as oppose to the GTV volume.