Solid pulmonary nodule risk assessment and decision analysis: comparison of four prediction models in 285 cases

The aim of this study was to compare classification results from four major risk prediction models in a wide population of incidentally detected solitary pulmonary nodules (SPNs) which were selected to crossmatch inclusion criteria for the selected models.

A total of 285 solitary pulmonary nodules with a definitive diagnosis were evaluated by means of four major risk assessment models developed from non-screening populations, namely the Mayo, Gurney, PKUPH and BIMC models. Accuracy was evaluated by receiver operating characteristic (ROC) area under the curve (AUC) analysis. Each model's fitness to provide reliable help in decision analysis was primarily assessed by adopting a surgical threshold of 65 % and an observation threshold of 5 % as suggested by ACCP guidelines.

ROC AUC values, false positives, false negatives and indeterminate nodules were respectively 0.775, 3, 8, 227 (Mayo); 0.794, 41, 6, 125 (Gurney); 0.889, 42, 0, 144 (PKUPH); 0.898, 16, 0, 118 (BIMC).

Resultant data suggests that the BIMC model may be of greater help than Mayo, Gurney and PKUPH models in preoperative SPN characterization when using ACCP risk thresholds because of overall better accuracy and smaller numbers of indeterminate nodules and false positive results.

• The BIMC and PKUPH models offer better characterization than older prediction models • Both the PKUPH and BIMC models completely avoided false negative results • The Mayo model suffers from a large number of indeterminate results.