A novel multimodal model integrating CT-based deep learning and clinical information for predicting 1-year OS of first-line immunotherapy in ES-SCLC patients.

Background: Extensive-stage small cell lung cancer (ES-SCLC) is a rare yet highly aggressive malignancy with severely limited treatment options. The efficacy of immunotherapy for ES-SCLC remain limited, with only 14% of patients deriving additional benefit from immune checkpoint inhibitors (ICIs), underscoring significant heterogeneity in patient response. This study aimed to develop a model for the accurate prediction of prognosis and therapeutic response to optimize treatment strategies. Methods: 99 and 25 ES-SCLC patients were enrolled as training and independent internal validation cohorts. Baseline CT images, clinical data, and laboratory markers were collected. Prognostic factors were screened via Cox regression. A mediastinal window CT-based deep learning model (MWDL) was developed and combined with Progastrin-releasing peptide (log(ProGRP + 1)) and bone metastasis to establish the novel MWGB-Cox model. Results: ProGRP and bone metastasis were independent prognostic factors with area under the curves (AUCs) of 0.705 and 0.666 respectively for predicting 1-year OS. Among the pure deep learning models, the MWDL model achieved better performance with a mean AUC of 0.813 than the deep learning model based on lung window CT images (LWDL, AUC 0.756). Whereas the multimodal MWGB-Cox model not only achieved the highest AUC (0.902) for 1-year OS prediction, but also effectively overcame the limitation of insufficient sensitivity. The predictive performance of the MWGB-Cox model was further validated in the independent internal validation cohort. Conclusions: This novel, simple, and cost-effective multimodal MWGB-Cox model integrating CT-based deep learning, log(ProGRP + 1), and bone metastasis could robustly predict 1-year OS of ES-SCLC patients receiving first-line immunotherapy, providing a reliable tool for clinical decision-making. Significance: By integrating CT-based deep learning, log(ProGRP + 1) levels, and bone metastasis, the novel multimodal model achieved the highest predictive AUC (0.902) with improved sensitivity, serving as a simple, cost-effective tool for clinical decision-making.