Lung nodule characterization: can the combination of CT and FDG PET parameters improve ability to predict malignancy?

1328 Objectives: Characterization of malignant risk within lung nodules remains an uncertain science with both computed tomography (CT) and fluorodeoxyglucose (FDG) positron emission tomography (PET). While schema exist for CT and PET alone for lung nodule risk stratification, few studies have modeled the predictive value of combining CT and PET data. Existing CT predictive models may not be applicable in this setting, as patients referred for PET imaging are at higher risk for cancer than an unselected population. Our objective was to assess the value of combining CT parameters with PET parameters toward quantifying risk of malignancy in indeterminate lung nodules. Methods: 303 subjects without known history of cancer were included in this retrospective observational study, all of whom had both chest CT and FDG PET/CT scan showing >1 solid lung nodule 10-30 mm in size, with definitive characterization of the nodule either via biopsy (histology showing either malignant neoplasia or alternative, e.g., inflammatory etiology) or ≥ 2 years of stability on CT imaging. CT and PET images were analyzed to assess location and size of dominant lung nodule, overall nodule count, spiculation, FDG uptake (assessed both by standardized uptake value [SUVmax] and 5-point Likert scale), and presence of other metabolically active lesions outside the lungs. Electronic medical records were reviewed for patient characteristics including age, sex, presence or absence of emphysema and family history of lung cancer. K-fold cross-validation was used to identify the model most predictive of malignancy. Multivariable logistic regression was used to assess predictive value of clinical plus CT imaging variables vs. clinical, CT imaging, plus PET imaging variables. Receiver operating curves (ROC) were compared across models. A model was determined to be superior where the area under the curve / c-statistic was higher and p was <0.05. Subanalyses were performed by FDG PET uptake score. Results: In multivariable analyses using clinical and CT variables, only nodule size (OR 1.21, 95% CI 1.05-1.40) and patient age (OR 1.15, 95% CI 1.11-1.31) were predictive of malignancy. In a model employing clinical, CT and PET data, FDG uptake score (OR 2.32, 95% CI 1.85-2.92) and presence of other metabolically active lesions (OR 3.23, 95% CI 1.65-6.34) were strongly predictive of malignancy; CT variables that had been significant in CT-only models did not remain significant in a model that included PET variables. The model using PET, CT and clinical data was significantly more predictive than the model with CT and clinical variables only (c-statistics: 0.877 vs. 0.731; p C-statistic comparing PET plus CT vs. CT only: 0.828 vs 0.709, p=0.004. Conclusions: For characterization of lung nodules with respect to malignancy, PET parameters (FDG uptake score and presence of other metabolically active lesions) are strongly predictive of malignancy. Addition of CT variables (other than nodule size) to PET variables adds little predictive value, as these cease to be statistically significant when combined with PET variables. Absence of FDG uptake (score=1) indicates very low (less than 10%) risk of malignancy. High FDG uptake (score=4-5) indicates high risk of malignancy, and further refinement of this risk is possible with addition of age, SUVmax and presence of other metabolically active lesions outside the lungs.