Since 2022, I have been working with Dr. Yuanjia Wang, Professor of Biostatistics at Columbia University, on my doctoral dissertation. Our work is focused on estimating heterogeneous treatment effects (HTE) in the setting of observational clinico-genomic data for salient time-to-event endpoints in oncology. All work has been done jointly with Dr. Yuan Chen (MSKCC) and Dr. Katherine Panageas (MSKCC).
Publications:
Lavery JA, Chen Y, Panageas KS, Wang Y. Unveiling non-small cell lung cancer treatment effect heterogeneity: a comparative analysis of statistical methods. J Natl Cancer Inst. 2025 Oct 1;117(10):2062-2072. doi: 10.1093/jnci/djaf176. PubMed PMID: 40637678; PubMed Central PMCID: PMC12505139.
Presentations:
Assessing Heterogeneous Treatment Effects for Time-to-Event Outcomes with a Mixture of Survival Models (ENAR Contributed Session, 2025): Heterogeneous treatment effect (HTE) estimation enables identification of subgroups of patients that respond differently to treatment and can be leveraged to improve patient outcomes. We previously evaluated several methods for estimating HTE in the context of time-to-event endpoints. We compared log-linear mixture models with inverse probability of censoring weights to Cox models, accelerated failure time models, and causal survival forests, and demonstrated that mixture models are a promising method for detecting HTE. We now extend the mixture model approach to handle censoring more effectively by using a mixture of survival models. This method assigns patients to subgroups and estimates treatment effects using the classification expectation-maximization algorithm, adapted to incorporate a covariate in the classification step. We will present a robust simulation study evaluating the performance of this method under different settings, varying the underlying distribution, sample size, and extent of censoring. We apply this approach to a case study utilizing data from the American Association for Cancer Research Project GENIE BPC, a clinico-genomic database of patients with cancer.
Assessing Heterogeneous Treatment Effects with Time-to-Event Outcomes (ENAR Contributed Session, 2024): To identify heterogenous treatment effects, Cox models with interaction terms have traditionally been presented in the medical literature. Recent methodological developments, including causal survival forests and Bayesian mixture models, enable a more flexible approach to modeling. Utilizing the American Association for Cancer Research Project GENIE BPC data, a multi-institution clinico-genomic database of patients with cancer, we will present methods to evaluate heterogeneous treatment effects of patients with advanced non-small cell lung cancer with time-to-event outcomes. For patients with NSCLC, immunotherapy has become standard of care for patients lacking a targetable genomic alteration, and it is unknown whether the addition of chemotherapy improves survival. Cox models, causal survival forests, and Bayesian mixture models were applied to identify features associated with variation in progression-free survival following treatment with immunotherapy alone versus with chemotherapy. We will contrast the assumptions of modeling approaches, proper adjustments of confounding, the resulting conclusions that can be drawn from each model, and present estimates of average treatment effects and conditional treatment effects.