Abstract:
|
Preinvasive or precancerous conditions are important indicators of increased cancer risk. These conditions are actively treated to maximize survival rates, which has resulted in challenges when analyzing data. To estimate incidence rates of invasive cancer, conventional methods either combine these conditions with invasive cancer to analyze a composite outcome, or treat them as censoring events, both of which can lead to biases. We aim to provide a guide on how to handle this issue in practice. We use a marginal cumulative incidence function (MCIF) to characterize the risk of invasive cancer. We proposed two approaches to approximate this function: one that decomposes MCIF into two different pathways and estimates the risk of each pathway based on assumptions about progression, and one that provides risk bounds corresponding to special cases of no progression or complete progression. We compared the two conventional approaches with the two proposed approaches through simulation studies and re-analyzed Sister Study data in the context of DCIS and invasive breast cancer.
|