Causal Mediation Analysis of Time-to-Event Data in the Context of Intersectionality in Breast Cancer

Background and objectives: Health disparities in breast cancer contribute to needlessly poor health outcomes. This research aimed to 1) synthesize the evidence on variations in time to breast cancer treatment initiation and survival across racial/ethnic groups using a DAG-based review method and 2) to investigate challenges and mitigation strategies stemming from sparse-data bias in intersectional analyses relevant to breast cancer. Methods: To achieve objective 1: A comprehensive systematic search was performed to describe implicit causal relationships between racial/ethnic group membership and time to treatment initiation and overall survival outcomes. For objective 2, a simulation study using mediation analysis with an accelerated failure time model was conducted to investigate the challenges that sparse data may present when estimating the causal effect of intersectional categories on survival. Results: In the systematic review, forty studies on cohorts of female patients diagnosed with stage I-III breast cancer were included. Studies reported significant disparities in delays in time to first treatment and subsequent impact on survival. The results of the simulation study demonstrated how sparse data bias can affect mediation effect measures. The scenarios with varying numbers of events per variables and cross-stratified groups showed that, despite a large sample size, challenges related to sparsity persist, often leading to unstable estimates across different scenarios. Conclusions: Persistent disparities in breast cancer treatment and mortality among racial/ethnic groups, especially between White and Black women, emphasize the need for equitable care and the elimination of these disparities to improve survival across diverse populations.