Performance of a multiplexed dual analyte immunoassay for the early detection of non-small cell lung cancer

Abstract

Objectives: "PAULA's" test (Protein Assays Utilizing Lung cancer Analytes) is a novel multiplex immunoassay blood test that incorporates both tumor antigens and autoantibodies to determine the risk that lung cancer (LC) is present in individuals from a high-risk population. The test's performance characteristics were evaluated in a study using 380 retrospective clinical serum samples.

Methods: PAULA's test is performed on the Luminex xMAP technology platform, and detects a panel of 3 tumor antigens (CEA, CA-125, and CYFRA 21-1) and 1 autoantibody marker (NY-ESO-1). A training set (n = 230) consisting of 115 confirmed diagnoses of non-small cell lung carcinoma (NSCLC) cases and 115 age- and smoking history-matched controls was used to develop the LC predictive model. Data from an independent matched validation set (n = 150) was then used to evaluate the model developed, and determine the ability of the test to distinguish NSCLC cases from controls.

Results: The 4-biomarker panel was able to discriminate NSCLC cases from controls with 74% sensitivity, 80% specificity, and 0.81 AUC in the training set and with 77% sensitivity, 80% specificity, and 0.85 AUC in the independent validation set. The use of NY-ESO-1 autoantibodies substantially increased the overall sensitivity of NSCLC detection as compared to the 3 tumor markers alone. Overall, the multiplexed 4-biomarker panel assay demonstrated comparable performance to a previously employed 8-biomarker non-multiplexed assay.

Conclusions: These studies confirm the value of using a mixed panel of tumor antigens and autoantibodies in the early detection of NSCLC in high-risk individuals. The results demonstrate that the performance of PAULA's test makes it suitable for use as an aid to determine which high-risk patients need to be directed to appropriate noninvasive diagnostic follow-up testing, especially low-dose CT (LDCT).

Figure 1

ROC curve analyses of training data set ( n  = 230) (A and B) and validation set ( n  = 150) (C and D). Panel A shows ROC curves for prediction of LC using multiple logistic regression weightings for each individual biomarker and a model combining all 4 biomarkers. The AUC values are 0.60, 0.79, 0.70, 0.69, and 0.83 for NY-ESO-1, CEA, CA125, CYFRA 21–1, and the combined panel, respectively; Panel B is a ROC curve generated from the 4-biomarker panel using the MoM scoring method (AUC value = 0.81); Panel C shows a ROC curve of the validation set for prediction of LC using the MLR method (AUC = 0.82), and Panel D shows the same data using MoM transformation (AUC = 0.85).

Figure 2

Distribution of PAULA’s test scores by NSCLC stage (A) and by histological types (B) using MoM model. The red horizontal lines show the PAULA’s test cut-off of 6.4 derived from training set results.