Combined Modeling of CT Radiomics and Clinical Parameters for Risk of Cachexia in Extensive-Stage Small Cell Lung Cancer Patients Treated with First-Line Immunotherapy

Abstract

Cachexia occurs in about 60% of patients with extensive-stage small cell lung cancer (ES-SCLC), which may contribute to resistance to immunotherapy. This study integrating CT radiomics and clinical parameters aimed to identify patients at risk of developing cachexia that may be associated with prognosis in extensive-stage small cell lung cancer patients treated with first-line immunotherapy. This study retrospectively included 200 ES-SCLC patients treated with first-line immunotherapy. All patients were randomly divided into training cohort and validation cohort according to a ratio of 7:3. The clinical characteristics and CT images before treatment were collected. Univariable and multivariable logistic regression analyses were conducted to evaluate the predictive role clinical characteristics. 3D slicer was used to retrieve radiomics features from CT images at the level of the third lumbar vertebra (L3). The radiomics signature was conducted by using the least absolute shrinkage and selection operator (LASSO). The receiver operating characteristic (ROC) curve analysis and Delong test were performed to determine and compare the predictive performance of radiomics, clinical and combined models. All patients were divided into training (n = 140) and validation (n = 60) cohorts. Survival analysis indicated the inferior PFS (7.3 vs 8.7 months, p = 0.0053) and OS (7.9 vs 9.3 months, p = 0.00099) for patients with cachexia. The multivariate analysis indicated that pretherapeutic lactate dehydrogenase (LDH) and post-treatment advanced lung cancer inflammation index (pretreatment BMI×serum albumin/NL) were independent risk factors for cachexia. A total of 883 radiomics features were obtained from each patient's CT scan, and four CT features were selected to build radiomics signature. The radiomics signature achieved area under curves (AUCs) of 0.706 (95% CI 0.624-0.780) in the training cohort, and 0.610 (95% CI 0.475-0.733) in the validation cohort. The combined model achieved AUCs of 0.769 and 0.674 in the training and validation cohorts, respectively, which performed better than clinical model or radiomics model only. Kaplan-Meier survival analysis showed that for the training cohort, the PFS (p = 0.0012) and OS (p = 0.0007) were significantly longer among patients with low RS compared to patients with high RS. For the validation cohort, similar results could also be observed in training cohort for PFS (p = 0.029) and OS (p = 0.091) estimation, respectively. The combined model based on CT and clinical parameters improved the performance for predicting cachexia in ES-SCLC patients receiving first-line immunotherapy. It provided a novel strategy for clinicians to identify ES-SCLC patients who may suffer from the cachexia and early interventions would improve prognosis of response to immunotherapy.