Causal inference in continuous time: an example on prostate cancer therapy.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Traditional marginal structural models (MSMs) use discrete time, which is a limitation for continuous-time survival processes.
• Continuous-time MSMs and weights were developed by Røysland (2011) to address this gap.
• This work demonstrates the practical application of continuous-time MSMs.

Purpose of the Study:

• To adapt and apply continuous-time marginal structural models for causal effect estimation in survival analysis.
• To compare the medium to long-term effectiveness of radical prostatectomy versus radiation therapy for prostate cancer.
• To account for competing risks in the analysis of treatment outcomes.

Main Methods:

• Utilized continuous-time marginal structural models with counting process notation.
• Calculated continuous-time weights, analogous to inverse probability weights in discrete time MSMs.
• Employed additive hazard models to derive causal effect estimates.
• Applied the methodology to data from the Norwegian Cancer Registry.

Main Results:

• Demonstrated the practical implementation of continuous-time MSMs.
• Found that the marginal cumulative incidence of treatment failure was similar between radical prostatectomy and radiation therapy.
• The analysis, which accounted for the competing risk of other causes of death, contrasted with naive analysis results.

Conclusions:

• Continuous-time MSMs provide a viable framework for causal inference in survival data.
• Radical prostatectomy and radiation therapy show comparable medium to long-term effectiveness for prostate cancer, considering competing mortality.
• The study highlights the importance of accounting for time-continuously and competing risks in treatment effect evaluation.