Abstract
BACKGROUND: Immune checkpoint inhibitor (ICI) therapy is an emerging type of treatment for lung cancer (LC). However, hyperprogressive disease (HPD) has been observed in patients treated with ICIs that lacks a prognostic prediction model. There is an urgent need for a simple and easily implementable predictive model to predict the occurrence of HPD. This study aimed to establish a novel scoring system based on a nomogram for the occurrence of HPD. METHODS: We retrospectively identified 1473 patients with stage III-IV LC or inoperable stage I-II LC (1147 in training set, and 326 in testing set), who had undergone ICI therapy at the Shanghai Chest Hospital between January 2017 and March 2022. Available computed tomography (CT) data from the previous treatment, before ICI administration, and at least 2 months after the first the course of ICI administration is collected to confirm HPD. Data from these patients' common blood laboratory test results before ICI administration were analyzed by the univariable and multivariable logistic regression analysis, then used to develop nomogram predictive model, and made validation in testing set. RESULTS: A total of 1,055 patients were included in this study (844 in the training set, and 211 in the testing set). In the training set, 93 were HPD and 751were non-HPD. Multivariate logistic regression analyses demonstrated that lactate dehydrogenase [LDH, P<0.001; odds ratio (OR) =0.987; 95% confidence interval (CI): 0.980-0.995], mean corpuscular hemoglobin concentration (MCHC, P=0.038; OR =1.021; 95% CI: 1.003-1.033), and erythrocyte sedimentation rate (ESR, P=0.012; OR =0.989; 95% CI: 0.977-0.997) were significantly different. The prediction model was established and validated based on these 3 variables. The concordance index were 0.899 (95% CI: 0.859-0.918) and 0.924 (95% CI: 0.866-0.983) in training set and testing set, and the calibration curve was acceptable. CONCLUSIONS: This model, which was developed from a laboratory examination of LC patients undergoing ICI treatment, is the first nomogram model to be developed to predict HPD occurrence and exhibited good sensitivity and specificity.