Abstract 9: Will our biomarkers have an impact? Estimating the utility of adding biomarkers to diagnostic risk models for lung cancer using intervention probability curves

Abstract Introduction: Assessing the potential clinical value of adding a new biomarker to a diagnostic risk model is a necessary step to improve clinical outcomes. The reclassification of patients across clinically relevant subgroups, such as low risk/intermediate risk/high risk, is considered the best metric to estimate potential utility. However, data from clinical decisions demonstrates that the “all-or-nothing” approach to threshold-based decisions is, in practice, incorrect. Here we present a novel approach to assessing biomarker utility, the change in intervention probability (IP). The intervention probability curve (IPC) models the likelihood that a provider will choose the intervention as a continuous function of the risk of disease. To assess the impact of adding a new biomarker, the change in likelihood of intervention is calculated for each patient based upon their change in risk. The summary statistics of the change in likelihood of intervention over the population can be descriptive of the expected clinical impact of the added biomarker. Methods: An IPC for lung cancer diagnosis based upon the Mayo Clinic Model was derived from the National Lung Screening Trial. Each patient’s baseline intervention probability to undergo a diagnostic bronchoscopy for an indeterminate pulmonary nodule (IPN) was calculated based upon their Mayo risk. Their post-test IP was calculated using a new biomarker strategy incorporating a blood test (CYFRA 21-1), a quantitative radiomics signature, and patient clinical history assessed in 457 patients with IPNs between 6-30 mm). The bias-corrected clinical net reclassification index (cNRI) was calculated as the comparator method for estimating biomarker utility. Results: Based upon ACCP risk thresholds of 0.05 and 0.65, the cNRI was 0.08 for the control population and -0.003 for the case population. Interestingly, despite greatly improved diagnostic accuracy (AUC of 0.754 improved to 0.855), the cNRI shows a very modest improvement in the control population and no improvement in the case population. However, the results of the IPC analysis show that over the entire population, there would be a net decrease in interventions among the control population of 8.3%, and a net increase in interventions among the case population of 0.8%. Conclusions: Analysis of the change in probability of intervention provides a more informative perspective of which patients would benefit from the addition of the combined biomarker method. Citation Format: Michael Nolan Kammer, Dianna J. Rowe, Stephen A. Deppen, Eric Grogan, Fabien Maldonado. Will our biomarkers have an impact? Estimating the utility of adding biomarkers to diagnostic risk models for lung cancer using intervention probability curves [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 9.