Causal Inference Methods Based on Pseudo-Observations: A Comparative Analysis of Treatment Types for Iranian...

This study investigates the effects of different treatments on the survival of patients with gastrointestinal cancer. One of the methods for causal inference, the doubly robust estimator, is easy to apply with complete data, but becomes complex with incomplete or censored data. To overcome these challenges, we used pseudo-observations.

This historical cohort study included 602 patients residing in the provinces of Mazandaran and Golestan. The patients were followed up over a period of approximately 11 years. To evaluate the effects of surgical, radiotherapeutic, and chemotherapeutic treatments on survival, we used inverse probability weighting and the doubly robust estimator using pseudo-observations. The survival and pseudo packages of the R software were used for model fitting. The efficacy of the doubly robust estimator and inverse probability weighting was assessed by comparing survival estimates derived from both methods across different treatment types.

In this study, there were 441 cases (73.26%) of deaths from gastrointestinal cancer. Our analysis included three time points: the first year, the eighth year and the fifteenth year. The results showed that the survival of patients with gastrointestinal cancer varied depending on the treatment method. In particular, surgical treatment showed a significant impact on survival at all three time points. In contrast, radiotherapy only showed a significant correlation at the first time point, while no significance was found at the two subsequent time points. A significant correlation was found for chemotherapy at all three time points.

Using a doubly robust estimator with pseudo-observations is more efficient and simpler. Causal inference methods can serve as evaluation tools for different treatments in patients with gastrointestinal cancer.