Construction and analysis of the invasive prediction model for pulmonary nodules: based on clinical, CT image and DNA methylation characteristics.

Abstract

Background: Accurately identifying whether pulmonary nodules are microinvasive adenocarcinoma or invasive carcinoma (MIA or IC) is clinically significant. This study aims to construct a predictive model for this.

Methods: Clinical, computed tomography (CT) image, and peripheral blood methylation data of 294 patients were collected. Based on postoperative pathology, they were divided into invasive (MIA or IC) and non-invasive groups. A quarter of the data was randomly selected as the validation set, and the rest was the training set. Screened significant indicators in training set and divided into three groups: clinical and image features, methylation features, and comprehensive features combining both. Logistic regression analyses were conducted respectively to construct models, and the model effect was verified in the validation set.

Results: There were six indicators in the comprehensive model (proportion of solid components, maximum CT value, SH3BP5_338_ CpG 4, PNPLA2_329_CpG 1, PNPLA2_329_CpG 4, and ARHGAP35 476_CpG_5). The area under the curve (AUC) of the training set and the validation set were 0.90 and 0.87, respectively. Prediction accuracies were 82% and 82%, sensitivities were 82% and 80%, specificities were 82% and 84%. The predictive effect of comprehensive model was better than that of the clinical and image feature model and the methylation feature model.

Conclusions: The invasiveness predictive model for pulmonary nodules constructed by combining clinical, CT image, and methylation features in this study has a relatively satisfactory effect and is worthy of further exploration and improvement.