Clinical-CT predictive modeling of stage I micropapillary adenocarcinoma components

<bold>Background:</bold> Early assessment of adenocarcinoma and prompt identification of micropapillae are imperative to enhance the prognosis and reduce the mortality in these patients. However, discrepancies between intraoperative frozen sections and paraffin permanent sections contribute to limited sensitivity of intraoperative frozen sections in diagnosing micropapillary components. Therefore, preoperative prediction of MPC is meaningful and helps to guide subsequent treatment. <bold>Method:</bold> Clinical and imaging features specifically associated with MPA patients were identified through univariate analysis. Utilizing multifactorial logistic regression analysis, we calculated odds ratio along with 95% confidence intervals for each parameter, discerned independent risk factors, established preliminary prediction models and scales, then tested their efficacy. <bold>Result:</bold> The clinical-CT prediction model is represented by the equation logit(p) = -4.430 + 0.812 × smoking history + 0.809 × CT diameter ≥22.5mm - 1.937 × GGO + 1.734 × lobulation + 1.325 × burr + 0.788 × vacuole + 1.348 × volume doubling time ≤310 days, where p denotes the probability of the patient's pathologic component containing MPC. The AUC value of the model was 0.867 (Figure A). After optimization for simplified clinical use (Figure B), the AUC value remained 0.867. In the validation population comprising 110 MPA and 115 non-MPA individuals, the model exhibited an AUC of 0.793. Furthermore, in patients with CT manifestations of ground-glass nodules, the model demonstrated an AUC of 0.801. <bold>Conclusion:</bold> The high area under the curve (AUC) value, as well as its consistency in the validation population, indicates the model's robust testing efficacy.