Derivation and validation of a 18F-FDG PET/CT scoring model to predict malignant pleural effusion.

ABSTRACT

OBJECTIVE: To develop an 18F-fluorodeoxyglucose PET/computed tomography (CT) scoring model based on metabolic and radiologic findings of the pleura and fluid to identify malignant pleural effusion. METHODS: The PET and CT findings from patients with pleural effusion in the derivation dataset were used to develop a scoring model. Then, the diagnostic accuracy of the predictive score was verified by the validation dataset. RESULTS: Eight parameters independently predicting malignancy were retained in the scoring model, including pleural nodules or masses (4 points), focal pleural thickening (2 points), absence of pleural loculation (2 points), thickness of mediastinal pleura involvement ≥0.5 cm (2 points), maximum standardized uptake value (SUVmax) of mediastinal pleura involvement ≥2.3 (2 points), thickness of nonmediastinal pleura involvement ≥0.5 cm (1 point), SUVmax of nonmediastinal pleura involvement ≥3.0 (1 point) and fluid SUVmax ≥1.6 (1 point). The operating characteristics of the PET/CT score were 0.958 area under the curve (AUC), 88.6% sensitivity, 91.2% specificity, 10.09 positive likelihood ratio and 0.13 negative likelihood ratio, with 6 points as the threshold. These values in the validation dataset were 0.947, 91.7%, 88.4%, 7.91 and 0.094, respectively. No difference was found in AUCs between the derivation and validation datasets (z = 0.517, P = 0.697). The negative predictive value was 99.4% in the score from 0 to 2, and the positive predictive value was 98.3% for patients with score between 9 and 15. CONCLUSIONS: The PET/CT scoring model is a valuable strategy to help physicians to distinguish malignant-benign pleural effusion and stratify patients who will benefit from invasive procedures.