Abstract
BACKGROUND: The heterogeneous biology of cancer subtypes, especially in lung cancer, poses significant challenges for biomarker development. Standard model building techniques often fall short in accurately incorporating various histologic subtypes because of their diverse biological characteristics. This study explores a nested biomarker model to address this issue, aiming to improve lung cancer early detection. METHODS: The study included 337 patients from two clinical sites. Blood biomarkers were analyzed and various statistical methods employed to develop a nested model. This model was designed to account for the biological heterogeneity across histologic subtypes, compared against traditional logistic regression models. RESULTS: The patient cohort included a range of malignant and benign nodules and included different cancer subtypes reflecting lung cancer heterogeneity. The nested model had comparable performance overall with the Mayo Clinic model and a standard logistic regression model with an AUC of 77.6 (95% confidence interval, 72.2-83.0) in training and 77.3 (95% confidence interval, 65.8-88.9) in testing. The nested subtype versus benign model had the best performance in the training set overall and had a particular advantage for small cell subtype prediction. CONCLUSIONS: This study highlights the challenges cancer heterogeneity present for biomarker development and the potential for nested biomarker models to improve early cancer detection. Validation of this approach in larger cohorts is essential to prove its predictive benefit in biologically diverse cancers. IMPACT: This work addresses the challenge of biological heterogeneity in biomarker development. A nested modeling approach may assist in developing more effective multicancer early detection strategies.