Integrative Nomogram of CT-based Radiomics and Clinical Features for Predicting Oligometastases at Recurrent after Definitive Chemoradiotherapy for Locally Advanced Non-Small Cell Lung Cancer

Abstract

To develop and validate a radiomics-based nomogram for predicting oligometastases at recurrent after definitive chemoradiotherapy (CRT) for locally advanced non-small cell lung cancer (LA-NSCLC). The study consisted of 118 LA-NSCLC patients treated with definitive CRT between January 1, 2014 and December 31, 2017. 96 and 22 patients were allocated to the primary cohort and the validation cohort, respectively. Baseline clinical features data were collected from the medical records. Computed tomography (CT) radiomics features were quantitatively extracted by the 3D slicer software and “pyradiomics” package. Least absolute shrinkage and selection operator (LASSO) Cox regression model was applied for date dimension reduction, feature selection and developing the radiomics signature. The discrimination of the radiomics signature was calculated by the area under the curve (AUC). Univariate and multivariate Cox regression analysis was utilized to evaluate independent predictors of oligometastases at recurrent. Finally, the model was developed and presented as the nomogram that incorporated both radiomics signature and independent clinical risk factors. Nomogram performance was assessed via the Harrell concordance index (C-index), calibration and clinical usefulness. With a median follow-up of 21.5 months (range: 6.5 to 58.8 months), oligometastases at recurrent was detected by follow-up imaging in 26 patients (21 patients were assigned to the primary cohort). For those who experienced oligometastases, the first metastases were in lung (38.5%), followed by brain, bone, and adrenal (19.2% each). The median time to oligometastases was 11.3 months (range: 2.0 to 41.3 months). The radiomics signature, which was based on selected 9 selected radiomics features, showed good discrimination in the primary cohort (AUC = 0.865) and the validation cohort (AUC = 0.659). Multivariate Cox proportional hazards testing revealed that histology subtype was an independent clinical risk factor (p-value < 0.05). After uni- and multivariate analysis, a nomogram was built based on histology subtype and radiomics signature. The model showed good discrimination in both the primary cohort and the validation cohort, with a C-index of 0.869 (95% CI, 0.776 to 0.962) and 0.797 (95% CI, 0.664 to 0.930), respectively. The calibration curve showed agreement between predicted and actual values for the probability of oligometastases of recurrent at 1, 2, or 3 years. Decision curve analysis demonstrated that the radiomics nomogram was clinically useful. The signature based on CT radiomics features can be used as an imaging marker for predicting oligometastases at recurrent after definitive CRT for LA-NSCLC. Nomograms combining radiomics signature and the independent clinical risk factor may constitute a usefully clinical tool to guide subsequent personalized treatments.