Background: Low-dose computed tomography (CT) screening reduces lung cancer mortality but costs are prohibitive for most healthcare budgets due to high false positive rates. An adjunctive test able to distinguish mali...Background: Low-dose computed tomography (CT) screening reduces lung cancer mortality but costs are prohibitive for most healthcare budgets due to high false positive rates. An adjunctive test able to distinguish malignant from benign pulmonary nodules would be hugely beneficial. EarlyCDT-Lung measures serum autoantibodies to tumor-associated antigens and has found clinical acceptance to aid early detection of lung cancer for high risk patients. However performance was optimized for screening. The construction of a receiver-operating characteristic (ROC) curve would enable optimization of performance for alternative settings, including nodule malignancy. Methods: A Monte-Carlo search method was used to construct a ROC curve using a case-control cohort, enabling high and low specificity versions of EarlyCDT-Lung to be determined. These were used for a theoretical evaluation of a nodule cohort, and positive predictive value (PPV) was calculated under the assumption of independence of risk source. Patients or their nodules are typically classified into three risk groups: low (0% - 10%), intermediate (10% - 65%) and high (>65%) risk of malignancy. The predicted shift in risk group by application of the high and low specificity versions, along with the current commercial EarlyCDT-Lung, was then estimated. Results: The ROC curve, with an area under the curve of 0.743, was constructed. The high specificity (98%), low specificity (49%) and current commercial (91% specificity) versions of EarlyCDT-Lung re-classified 27%, 23% and 26% of intermediate nodules, respectively, to either a higher (10%, 8% and 10%) or lower (17%, 15% and 16%) risk group. Conclusion: A ROC curve was constructed to allow performance prediction of EarlyCDT-Lung at different specificities in the indeterminate nodule setting. This enabled risk re-classification of intermediate risk nodules, and could therefore facilitate alternative more appropriate intervention. We have shown how a multivariate biomarker test can add to the interpretation of pulmonary nodules and therefore aid patient management.展开更多
文摘Background: Low-dose computed tomography (CT) screening reduces lung cancer mortality but costs are prohibitive for most healthcare budgets due to high false positive rates. An adjunctive test able to distinguish malignant from benign pulmonary nodules would be hugely beneficial. EarlyCDT-Lung measures serum autoantibodies to tumor-associated antigens and has found clinical acceptance to aid early detection of lung cancer for high risk patients. However performance was optimized for screening. The construction of a receiver-operating characteristic (ROC) curve would enable optimization of performance for alternative settings, including nodule malignancy. Methods: A Monte-Carlo search method was used to construct a ROC curve using a case-control cohort, enabling high and low specificity versions of EarlyCDT-Lung to be determined. These were used for a theoretical evaluation of a nodule cohort, and positive predictive value (PPV) was calculated under the assumption of independence of risk source. Patients or their nodules are typically classified into three risk groups: low (0% - 10%), intermediate (10% - 65%) and high (>65%) risk of malignancy. The predicted shift in risk group by application of the high and low specificity versions, along with the current commercial EarlyCDT-Lung, was then estimated. Results: The ROC curve, with an area under the curve of 0.743, was constructed. The high specificity (98%), low specificity (49%) and current commercial (91% specificity) versions of EarlyCDT-Lung re-classified 27%, 23% and 26% of intermediate nodules, respectively, to either a higher (10%, 8% and 10%) or lower (17%, 15% and 16%) risk group. Conclusion: A ROC curve was constructed to allow performance prediction of EarlyCDT-Lung at different specificities in the indeterminate nodule setting. This enabled risk re-classification of intermediate risk nodules, and could therefore facilitate alternative more appropriate intervention. We have shown how a multivariate biomarker test can add to the interpretation of pulmonary nodules and therefore aid patient management.
