Dealing with death when studying disease or physiological marker: the stochastic system approach to causality

Area of Science:

• Causal inference
• Stochastic processes
• Biostatistics

Background:

• Causality research often simplifies systems, neglecting the impact of death.
• Existing methods may struggle with non-negligible mortality risks and complex data types.

Purpose of the Study:

• To apply a stochastic system approach to causality in scenarios with significant mortality risk.
• To develop methods for estimating causal effects of factors on processes of interest, incorporating death.
• To unify analyses for both event (counting process) and quantitative outcomes.

Main Methods:

• Utilizing multivariate stochastic processes to model complex systems.
• Defining local influences on the random time horizon of death.
• Examining continuous and discrete time observations and the impact of data incompleteness.

Main Results:

• Demonstrates a unified approach for estimating causal effects across different outcome types (counting and quantitative).
• Highlights the importance of defining influences relative to the time of death.
• Investigates the conditions under which mechanisms causing incomplete data can be disregarded.

Conclusions:

• The stochastic system approach offers a robust framework for causal inference when death is a significant factor.
• The methods are applicable to diverse data structures and can handle data incompleteness.
• Provides a foundation for analyzing real-world scenarios, such as the effect of blood pressure on cognitive ability in the elderly.