Abstract
OBJECTIVE: This study aimed to construct a model that predicts invasive lung cancer using longitudinal radiological features from multiple low-dose computed tomography (LDCT) scans, thereby addressing overdiagnosis in lung cancer screening. METHODS: In this retrospective study, 628 patients with pulmonary nodules who underwent three LDCT scans followed by surgical resection were categorized into invasive carcinoma (n=155) and non-invasive nodule (n=473) groups on the basis of pathological diagnosis. This derivation aimed to identify risk factors and construct a multivariate logistic model. The predictive performance was externally validated in two independent cohorts (retrospectively designed, n=252; prospectively designed, n=269). The discrimination and calibration of the model were evaluated using area under the curve (AUC), and calibration plots. Decision curve analysis (DCA) was further performed to evaluate the net benefit in practical clinical scenarios. RESULTS: The model, termed multiple CTs-invasive lung cancer (MCT-ILC), incorporated eleven factors encompassing nodule features at baseline and feature variability during follow-up. The standard deviation of diameter variability (SD(diameter)) was the most reliable predictor, with an odds ratio [95% confidence interval (95% CI) of 7.35 (5.32-10.16) (P<0.001). AUCs with 95% CIs for the MCT-ILC model were 0.912 (0.864-0.960) and 0.906 (0.833-0.979) in the two testing cohorts and were superior to those for the model containing only features at baseline (P(Delong)=0.002 and 0.021, respectively). For calibration, the Brier scores of the MCT-ILC model were 0.091 (95% CI: 0.064-0.118) and 0.078 (95% CI: 0.055-0.101) in the two test sets. The decision curve image showed that the MCT-ILC model was the only model that maintained positive net benefits across the entire threshold range. Furthermore, the MCT-ILC model score could classify more than 90% of patients with invasive nodules into the high-risk group. CONCLUSIONS: The MCT-ILC model could assess pulmonary nodule invasiveness, potentially mitigating overdiagnosis in lung cancer screening.