Radiomic model to predict the expression of PD-1 and overall survival of patients with ovarian cancer

Background: Programmed cell death 1 (PD-1), encoded by programmed cell death protein 1 (PDCD1), is widely investigated in clinical trials. We aimed to develop a radiomic model to discriminate its expression levels patients with ovarian cancer (OC) and explore its prognostic value. Methods: Computed tomography (CT) images with the corresponding sequencing data and clinicopathological features were used. The volumes of interest were manually delineated. After extraction and normalization, the radiomic features were screened using repeat least absolute shrinkage and selection operator. A radiomic model for PD-1 prediction, radiomic score (rad_score), was developed using logistic regression and validated via internal 5-fold cross-validation. The Kaplan-Meier curves, COX proportional hazards model, and landmark analysis were used for survival analysis. Results: The mRNA level of PDCD1 significantly affects the overall survival (OS) of OC patients. The rad_score for PDCD1 prediction was based on four features and was significantly correlated with other genes involved in T-cell exhaustion and immune checkpoint molecules. The areas under the receiver operating characteristic curves reached 0.810 and 0.772 in the training and validation datasets, respectively. The calibration curves and decision curve analysis proved the model's fitness and clinical benefits. Patients with higher rad_score had poorer OS (P < 0.001, 0.031, 0.014, 0.01, and < 0.001, after landmark of 12 months, before and after landmark of 36 months, and before and after landmark of 60 months, respectively). Conclusions: The radiomic signature from CT images can discriminate the PD-1 expression status and OC prognosis, which is correlated with T-cell exhaustion.