Objectives: Diagnostic tools that stratify lung cancer (LC) risk can help prioritize care for patients at the highest risk and optimize time and procedures to achieve the final diagnosis. We have previously demonstrated that six tumour biomarkers (TBs) - CEA, CYFRA 21.1, CA 15-3, SCC Ag, ProGRP, and NSE - can help assess LC risk. We developed expert software that combines these TBs with clinical and imaging data to estimate LC risk.
Methods: The diagnostic accuracy of this expert software was evaluated in a multicentre study.We prospectively recruited 2,005 individuals referred to 12 reference hospitals in Spain and Portugal for suspicion of LC. The six TBs were determined and the expert software was applied to all patients and correlated with the final diagnosis.
Results: A final diagnosis of LC was made in 1,392 patients. The expert software yielded 87.7% sensitivity, 75.5% specificity, 89.0% positive predictive value and 73.0% negative predictive value. Sensitivity increased with tumour size and extension. The software also provides histological information, correctly predicting cancer in 98.4% of small-cell LC and 93.2% of non-small-cell LC, which correlates with the histological diagnosis of 90% and 91.2%, respectively.
Conclusions: The expert software developed provides excellent diagnostic accuracy for diagnosing LC. Accordingly, this software can help stratify the risk of LC and prioritize the evaluation of patients at higher risk, optimizing procedures based on risk and knowledge of the most likely histological type, and providing a valuable tool for risk stratification and clinical decision support, particularly in Rapid Diagnostic Units.







