A novel predictive model for distinguishing mediastinal lymphomas from thymic epithelial tumours

Abstract

Abstract OBJECTIVES We recently reported a high rate of nontherapeutic thymectomy. Mediastinal lymphomas (MLs) are the malignancies most likely to be confused with thymic epithelial tumours (TETs). This study aimed to establish a predictive model by evaluating clinical variables and positron emission tomography to distinguish those diseases. METHODS From 2018 to 2021, consecutive patients who were pathologically diagnosed with TETs or MLs were retrospectively reviewed. Univariable and multivariable analyses were used to identify association factors. The Akaike information criterion was used to select variables. A nomogram was developed and validated to differentiate MLs from TETs. RESULTS A total of 198 patients were included. Compared with TETs, patients with MLs were more likely to be younger with higher metabolic tumour volume (154.1 vs 74.6 cm3), total lesion glycolysis (1388.8 vs 315.2 g/ml cm3), SUVmean (9.2 vs 4.8), SUVpeak (12.9 vs 6.3) and SUVmax (14.8 vs 7.5). A nomogram was established based on the stepwise regression results and the final model containing age and SUVmax had minimal Akaike information criterion value of 72.28. Receiver operating characteristic analyses indicated that the area under the curve of predictive nomogram in differentiating MLs from TETs was 0.842 (95% CI: 0.754–0.907). The internal bootstrap resampling and calibration plots demonstrated good consistence between the prediction and the observation. CONCLUSIONS Combination of age and SUVmax appears to be a useful tool to differentiate MLs from TETs. The novel predictive model prevents more patients from receiving nontherapeutic thymectomy.