WCLC 2025 - Posters & ePosters-P3.13.14 Lab Data-Based Prediction for Clinical Efficacy of Immunochemotherapy in ES-SCLC Using Statistical and Machine Learning Models

P3.13.14 Lab Data-Based Prediction for Clinical Efficacy of Immunochemotherapy in ES-SCLC Using Statistical and Machine Learning Models

Back to course

Pdf Summary

This study developed and validated statistical and machine learning models using readily available laboratory data to predict clinical efficacy of immunochemotherapy (PD-L1 inhibitors plus chemotherapy) in extensive-stage small cell lung cancer (ES-SCLC). Traditional biomarkers such as PD-L1 expression and tumor mutational burden have shown inconsistent predictive value, and RNA-seq gene expression signatures are impractical for routine clinical use. The study analyzed retrospective data from six Japanese institutions involving patients treated with atezolizumab or durvalumab combined with chemotherapy.<br /><br />The primary outcome was prolonged progression-free survival (PFS) defined as ≥5 months, with 213 patients eligible; 150 were used for training and 63 for testing the model. A non-linear logistic regression model incorporated baseline laboratory values—C-reactive protein, lactate dehydrogenase, neutrophil-to-lymphocyte ratio, and albumin—as predictors. Model prediction was considered indicative of prolonged PFS if probability exceeded 0.58. A similar approach was applied for overall survival (OS) ≥12 months using 194 patients.<br /><br />Model performance indicated moderate discriminative ability, with area under the curve (AUC) ~0.66 in training and ~0.64 in testing sets, suggesting good generalization without overfitting. However, validation on an independent cohort (n=46) showed poorer performance (AUC 0.468), likely due to sample size or cohort differences. Using a cutoff probability of 0.58 optimized specificity, reducing false positives crucial for prognostic reliability. Kaplan–Meier analyses showed clear separation of high- versus low-risk groups for PFS across datasets, supporting clinical relevance of predictions. OS model results were less consistent.<br /><br />Limitations include modest predictive performance and limited sample size preventing evaluation of longer-term cutoffs. Despite this, the study demonstrates that models based on standard laboratory parameters can stratify prognosis in ES-SCLC patients undergoing immunochemotherapy, potentially guiding clinical decision-making. Larger cohorts are needed to improve robustness and to develop models for longer-term survival outcomes.

Asset Subtitle

Shuichiro Hara

Meta Tag

Speaker Shuichiro Hara

Topic Small Cell Lung Cancer and Neuroendocrine Tumors

Keywords

immunochemotherapy

PD-L1 inhibitors

extensive-stage small cell lung cancer

progression-free survival

overall survival

machine learning model

laboratory biomarkers

C-reactive protein

non-linear logistic regression

Kaplan–Meier analysis