Background: With many lung cancers being detected at early stages,some might be overtreated because of subjective follow-up strategies.We developed an accurate predictive model to assess the progression risk level(PRL)in pulmonary precursor glandular lesions(PGLs)and pre-stage IA2 adenocarcinomas to provide objective treatment strategies,including follow-up frequency,surgery timing,and surgical approach.Methods: We enrolled 6040 patients with PGL or early-stage lung adenocarcinoma from10 thoracic surgery institutions covering various regions of China.A nomogram and a machine learning model were developed based on data from 1432 patients who had received at least two CT screenings up to 3-6 months apart to predict the PRL and observable progress time(OPT).External validation was performed using another 366 patients from distinct institutions.Relationships between pathology and PRL were assessed in 5674 patients,excluding the external group.Result: Six variables were selected using multivariate logistic regression analysis to establish a nomogram for PRL.The AUC values(development cohort 0.962,validation cohort 0.942,external validation cohort 0.921)and GiViTi calibration curves indicated satisfactory discrimination,calibration,and generalisation of the nomogram.Random forest was chosen as the machine learning approach to determine the relationship between PRL and OPT because of its better goodness of fit and performance based on satisfactory R2,root mean square error,and mean absolute error.PRL values ≥60 and≥99 likely represent invasive adenocarcinoma cancer(IAC)and a high-risk subtype component of IAC,respectively.Conclusion: The PRL prediction model for PGL and pre-stage IA2 adenocarcinoma provides an objective indication to guide clinical management decisions for the entire follow-up period and stages from atypical adenomatous hyperplasia to IAC,facilitating the diagnosis,follow-up,and treatment of early adenocarcinomas. |