Prediction of pneumonia following neoadjuvant chemoradiotherapy in patients with oesophageal cancer.

Abstract

Pneumonia is a frequent, severe complication following neoadjuvant chemoradiotherapy (nCRT) and esophagectomy for oesophageal cancer, adversely affecting outcomes. We aimed to develop a model to accurately predict this risk. This multicentre, retrospective study included oesophageal cancer patients (cT1-4N0-3M0) undergoing nCRT +/- esophagectomy (CROSS regimen) treated between 2015-2021. The endpoint was grade ≥2 pneumonia (CTCAE v5.0) within six months post-nCRT. To handle high dimensionality, principal component analysis (PCA) was used to condense lung and heart DVH patterns into interpretable dose patterns. A logistic regression model was developed and validated using internal-external cross-validation to assess discrimination, calibration, and heterogeneity across centres. A total of 1,459 patients across five centers were included for the final model development; 314 (22%) of which developed pneumonia. The developed model included pre-existing lung disease, diabetes, esophagectomy, and three PCA-derived dose patterns (overall heart/lung dose, heart-versus-lung dose, and low-dose areas). The model showed low heterogeneity (I² = 0% for all measures), fair discrimination (pooled AUC 0.68; 95% CI, 0.63-0.72), and excellent calibration (slope 0.91; 95% CI, 0.56-1.26; calibration-in-the-large 0.02; 95% CI, -0.14-0.18). We developed and validated a generalizable NTCP model for pneumonia prediction in oesophageal cancer patients. This multicentre pulmonary NTCP model showed good calibration and homogeneous performance between centres. It offers a promising tool to personalize treatment by facilitating radiotherapy plan optimization and treatment/technique selection.