Predicting clinical benefit of immunotherapy in veterans

Though the introduction of immune checkpoint inhibitors (ICI) has improved outcomes for patients with stage III non-small-cell lung cancer, most patients do not have robust tumor responses to ICI and real-world patients discontinue ICI due to toxicity at a higher rate than clinical trial populations. There is an unmet need to develop predictive models to identify patients at risk of poor ICI response, high risk of ICI toxicity, and high risk of non-cancer mortality, as these patients are unlikely to benefit from ICI and may require treatment intensification or de-intensification, selection of alternative therapies, or enrollment on clinical trials. In this proposal we seek to leverage the richness and national scale of Veterans Affairs (VA) clinical data to develop statistical models to predict ICI efficacy, toxicity, and non-cancer mortality. We will evaluate specific hypotheses generated from preclinical research and use machine-learning approaches to identify new predictors from high-dimensional VA healthcare data. These models will be integrated into a unified multistate model, allowing prediction of the individualized overall survival benefit of ICI for each patient. Clinical decision-making based on model predictions will be compared to other decision making strategies using decision-curve analysis. This work will identify novel biomarkers of ICI efficacy and toxicity, validate putative ICI effect modifiers from the preclinical literature, and develop the first unified model for ICI clinical benefit. If successful, this tool can be used for patient counseling, physician decision-making, and identification of patients likely to benefit from alternative therapies, treatment de-escalation, or clinical trial enrollment.