Development of a haematological indices-based nomogram for prognostic prediction and immunotherapy response assessment in primary pulmonary lymphoepithelioma-like carcinoma patients.

ABSTRACT

Background: Primary pulmonary lymphoepithelioma-like carcinoma (PPLELC) is a rare yet aggressive malignancy. This study aims to investigate a deep learning model based on hematological indices, referred to as haematological indices-based signature (HIBS), and propose multivariable predictive models for accurate prognosis prediction and assessment of therapeutic response to immunotherapy in PPLELC. Methods: This retrospective study included 117 patients with PPLELC who received immunotherapy and were randomly divided into a training (n=82) and a validation (n=35) cohort. A total of 41 hematological features were extracted from routine laboratory tests and the least absolute shrinkage and selection operator (LASSO) algorithm were utilized to establish the HIBS. Additionally, we developed a nomogram using the HIBS and clinical characteristics through multivariate Cox regression analysis. To evaluate the nomogram's predictive performance, we used calibration curves and calculated the time-dependent area under the curve (AUC). Kaplan-Meier survival analysis was performed to estimate progression-free survival (PFS) in both cohorts. Results: The proposed HIBS comprised 14 hematological features and showed that patients who experienced disease progression had significantly higher HIBS scores compared to those who did not progress (P<0.001). Five prognostic factors, including HIBS, tumor-node-metastasis (TNM) stage, presence of bone metastasis and the specific immunotherapy regimen, were found to be independent factors and were used to construct a nomogram, which effectively categorized PPLELC patients into a high-risk and a low-risk group, with patients in the high-risk patients demonstrating worse PFS (7.0 vs. 18.0 months, P<0.001) and lower overall response rates (22.2% vs. 52.7%, P<0.001). The nomogram showed satisfactory discrimination for PFS, with AUC values of 0.837 and 0.855 in the training and validation cohorts, respectively. Conclusions: The HIBS-based nomogram could effectively predict the PFS and response of patients with PPLELC regarding immunotherapy and serve as a valuable tool for clinical decision making.