Causal inference for recurrent events via aggregated marginal odds ratio

Area of Science:

• Epidemiology
• Biostatistics
• Psychology

Background:

• Causal effect measures are underdeveloped for time-varying treatments and recurrent events.
• Recurrent events and time-varying treatments are common in psychological and medical research.
• Existing measures are primarily designed for single, one-time treatments.

Purpose of the Study:

• To propose a novel causal measure for quantifying the impact of time-varying treatments on recurrent events.
• To develop robust estimators with standard errors for various weight models.
• To compare the performance of different weight models in estimating causal effects.

Main Methods:

• Development of a new causal estimand for time-varying treatments and recurrent events.
• Utilizing inverse probability weight models, including stabilized versions, for estimation.
• Comparison of estimation results across different treatment settings and weight models.
• Demonstration of consistent estimation for moderately long study periods.

Main Results:

• The proposed causal estimand can be consistently estimated.
• Stabilized inverse probability weight models show advantages in certain settings.
• The method is effective for both absorbing and nonabsorbing treatments.
• The approach was successfully applied to the 1997 National Longitudinal Study of Youth.

Conclusions:

• The novel causal measure effectively quantifies the effect of time-varying treatments on recurrent events.
• The proposed estimation methods provide robust standard errors and are suitable for various weight models.
• The approach is applicable to diverse treatment scenarios, including those in psychological research.