Performance of each prediction model in diagnosing minimally invasive adenocarcinoma and invasive adenocarcinoma for pure ground-glass nodules
| Sensitivity | Specificity | Accuracy | PPV | NPV | AUC | |
| Brock model 1* | 32.2 | 90.5 | 56.4 | 82.6 | 48.7 | 0.671 (0.571–0.762) |
| Brock model 2† | 64.4 | 64.3 | 64.4 | 71.7 | 56.3 | |
| Lee model (nodule size‡) | 76.3 | 42.9 | 62.4 | 65.2 | 56.3 | – |
Data are in percentages except for AUC. There were 42 preinvasive lesions and 59 invasive lesions (minimally invasive adenocarcinomas and invasive adenocarcinomas).
↵*A cut-off of 10% predicted probability was used as suggested by the British Thoracic Society.
↵†A cut-off of 4.29% predicted probability, the optimal threshold based on the Youden Index, was used.
↵‡A cut-off of 10 mm nodule size was used.
AUC, area under the receiver operating characteristic curve; NPV, negative predictive value; PPV, positive predictive value.