| dc.description.abstract | Objective: The purpose of this doctoral dissertation project was to 1) compare the effectiveness of first-line individual immune checkpoint inhibitor (ICI) agents and ICI in combination with chemotherapy (ICI-Chemotherapy) versus ICI monotherapy (ICI-Monotherapy) for patients with metastatic non-small cell lung cancer (mNSCLC), 2) develop regression and machine learning (ML) models for the prediction of immune-related adverse events (irAEs), and 3) evaluate the cost-effectiveness of first-line cemiplimab plus chemotherapy (CCT) compared to pembrolizumab plus chemotherapy (PCT).
Methods: For Aims 1 and 2, we used the 2014–2020 Surveillance, Epidemiology, and End Results (SEER)-Medicare-linked database. In aim 1, we designed a new-user cohort to compare the effectiveness of first-line atezolizumab, nivolumab, and pembrolizumab among older adults with mNSCLC. Overall survival (OS) was compared using unadjusted, multivariable-adjusted, and propensity score matching (PSM) Cox proportional hazards models. Additionally, we emulated two hypothetical target trials to compare OS between first-line ICI-Monotherapy and ICI-Chemotherapy, and 4 versus 6 cycles of chemotherapy within the first-line ICI-Chemotherapy regimens. The clone-censor-weight approach and discrete-time hazards model were applied to estimate the per-protocol treatment effects. OS was evaluated through survival curves, 72-week OS probabilities,72-week restricted mean survival time (RMST), and hazard ratios (HR). In aim 2, given the suboptimal performance of models in predicting overall irAEs, we systematically evaluated predictive performance across irAE subtypes, ultimately focusing on endocrine-related irAEs. Patients with mNSCLC treated with first-line ICIs were randomly divided into training (80%) and testing (20%) sets. We developed the Least Absolute Shrinkage and Selection Operator (LASSO) and ML approaches, including random forest, support vector machines, and XGBoost, to predict 3-month endocrine-related irAEs. Model performance was evaluated using accuracy, F1 score, and the area under the receiver operating characteristic curve (AUROC) in the test set. In aim 3, a three-state partitioned survival model with a 10-year time horizon was constructed. Clinical data were sourced from the EMPOWER-Lung 3, KEYNOTE-407, and KEYNOTE-189 trials. Costs and quality of life inputs were obtained from the 2024 U.S. Centers for Medicare & Medicaid Services drug price lists and published literature. We calculated the total cost, quality-adjusted life-years (QALYs) gained, and the incremental cost-effectiveness ratio (ICER).
Results: Pembrolizumab was associated with a significant survival benefit compared to atezolizumab (multivariable-adjusted HR=0.67; 95% confidence interval (CI)=0.53-0.84; PSM HR=0.69; 95% CI=0.54-0.87) and nivolumab (multivariable-adjusted HR=0.83; 95% CI=0.69-0.99). In the per-protocol analysis of emulated target trials, ICI-Chemotherapy provided a marked OS advantage over ICI-Monotherapy, including a 5.5% higher 72-week OS rate (95% CI=4.2%-6.7%), a 6.27-week additional RMST gain (95% CI=4.88-7.61), and HR=0.76 (95% CI=0.71-0.81). Patients receiving 6 versus 4 cycles of chemotherapy had a 5.1% (95% CI=–0.9%–11.0%) higher 72-week OS rate, a 2.19-week additional RMST gain (95% CI=0.54–4.92), and an HR of 0.83 (95% CI=0.65–0.95). The LASSO model outperformed other ML models to predict 3-month endocrine-related irAEs, achieving an accuracy of 0.791 (95% CI=0.789–0.793), an F1 score of 0.507 (95% CI=0.502–0.512), and an AUROC of 0.724 (95% CI=0.719–0.728). The most important predictors identified included pre-existing endocrine disorders, type 2 diabetes, use of atezolizumab, concurrent chemotherapy, and patient demographics. In the cost-effectiveness analysis, the total cost of PCT was $207,926 with 1.609 QALYs, whereas CCT had a total cost of $175,247 with 1.657 QALYs, indicating that CCT was a dominant first-line treatment strategy over PCT (ICER=−$675,304 per QALY) for patients with mNSCLC.
Conclusion: Pembrolizumab was associated with a significant OS benefit compared with atezolizumab and nivolumab among older adults with mNSCLC. Adding chemotherapy to first-line ICI therapy offered additional survival benefits over ICI-Monotherapy, and extending chemotherapy from 4 to 6 cycles was associated with additional improvements in survival outcomes. The LASSO model demonstrated the best performance in predicting endocrine-related irAEs using SEER-Medicare data. Compared with PCT, CCT was a dominant first-line treatment option for mNSCLC from a cost-effectiveness perspective. | en_US |
