2023 North America Conference on Lung Cancer (NACLC) - Posters and Abstracts-PP01.76 (Poster) Optimizing Biomarker Models for Biologically-Heterogeneous Cancers: A Nested Model Approach for Lung Cancer

Pdf Summary

This study aimed to investigate whether a nested algorithm approach could improve the performance of a biomarker prediction model for lung cancer, incorporating information from biomarkers associated with less common histologic subtypes. Currently, there is no non-invasive blood-based biomarker in widespread clinical use for early detection of lung cancer, and the heterogeneity of the disease makes identifying a single useful biomarker challenging.<br /><br />The study followed a prospective specimen collection, retrospective blinded evaluation design and included 337 patients from Vanderbilt University Medical Center and the Veterans Affairs Hospital - Tennessee Valley Healthcare System. Blinded serum specimens were sent to Abbott Laboratories for measurements of various biomarkers and clinical variables. Logistic regression models were developed using feature selection to determine the most important predictive features for each subtype of lung cancer.<br /><br />The nested algorithm approach involved fitting a model for each lung cancer subtype using the selected features and then using the predicted probabilities from the subtype models as inputs for an overall model for lung cancer malignancy versus benign subjects. The study compared the performance of the nested algorithm approach to other modeling approaches and found that it addressed the barrier of histologic heterogeneity, improving performance by selecting the most impactful biomarkers for each subtype.<br /><br />The study concluded that the nested algorithm approach improved performance by addressing the histologic heterogeneity of lung cancer. By training individual models on each subtype and using these model outputs as inputs to the overall nested model, the issue of losing data inputs more sensitive to less common subtypes was reduced.<br /><br />The results showed that the nested algorithm approach had comparable performance to other approaches, with the subtype versus benign model performing the best overall. It had high specificity, particularly for the small cell subtype.<br /><br />In summary, this proof-of-concept study demonstrated that a nested algorithm approach improved the performance of a biomarker prediction model for lung cancer by considering the heterogeneity of the disease and selecting the most impactful biomarkers for each subtype. The study suggests that this approach could potentially aid in early detection and diagnosis of lung cancer.

Asset Subtitle

Palina Woodhouse

Keywords

nested algorithm

biomarker prediction model

lung cancer

histologic subtypes

non-invasive blood-based biomarker

early detection

heterogeneity

prospective specimen collection

retrospective blinded evaluation

logistic regression models