Study
PLCO
(Learn more about this study)
Project ID
PLCO-1615
Initial CDAS Request Approval
Jul 8, 2024
Title
Structural nature of time-related biases in cancer screening studies: Using PLCO Trial data to inform causal diagrams and simulation
Summary
While the threat of time-related biases, such as lead time bias, to validity in cancer screening studies has been well known for several decades, they continue to provide challenges to researchers seeking to estimate effectiveness of screening in increasing survival time. Since their adoption in epidemiologic research, causal diagrams have grown as a tool for understanding the structural nature of biases, often providing intuitive insights through their visual representation of biasing pathways. Proposed structures for immortal time bias have recently been published and debated. Causal diagrams representing other time-related biases may prove useful for researchers using observational data to estimate the effects of cancer screening, aiding in decision-making during analysis and demonstrating causal mechanisms of bias. This project aims to propose relevant causal diagrams and use diagram-based simulation to illustrate the biases. The PLCO trial data, as results from randomized trials, can be used to (1) inform simulation parameters, (2) illustrate how trials avoid certain time-related biases (though sometimes with limits on the types of estimands that are identifiable), and (3) demonstrate how analytical decisions can lead to time-related biases even in the context of a trial. Heterogeneity in various factors plays an important role in these biases and the ways in which they can be addressed, these diagrams and simulations can incorporate heterogeneity.
Aims
1) In the context of cancer screening examples, to create causal diagrams representing time-related biases (e.g., lead time bias, length bias, immortal time bias) and use formal causal inference frameworks (i.e., potential outcomes, structural causal model) to represent causal and statistical estimands and characterize how these biases violate necessary assumptions or misrepresent estimands
2) To demonstrate these biases through causal diagram-based simulations with simulation parameters informed by analysis of PLCO trial data (e.g., to estimate lead time or to analyze data in ways that lead to immortal time bias)
3) To demonstrate these biases through analysis of PLCO data (comparing analytical methods leading to this bias versus those avoiding these biases)
4) To characterize how heterogeneity in various factors (e.g., lead time, tumor growth rates) that may be associated with sociodemographic characteristics of interest can impact bias and how information from trials can be generalized to correct for bias in observational data
Collaborators
Zuo-Feng Zhang; University of California, Los Angeles
Onyebuchi Arah; University of California, Los Angeles
Chad Hazlett; University of California, Los Angeles
Baffour Adusei-Poku; University of California, Los Angeles
