Estimating the Individual Treatment Effect on Survival Time Based on Prior Knowledge and Counterfactual Prediction

Abstract

The estimation of the Individual Treatment Effect (ITE) on survival time is an important research topic in clinics-based causal inference. Various representation learning methods have been proposed to deal with its three key problems, i.e., reducing selection bias, handling censored survival data, and avoiding balancing non-confounders. However, none of them consider all three problems in a single method. In this study, by combining the Counterfactual Survival Analysis (CSA) model and Dragonnet from the literature, we first propose a CSA–Dragonnet to deal with the three problems simultaneously. Moreover, we found that conclusions from traditional Randomized Controlled Trials (RCTs) or Retrospective Cohort Studies (RCSs) can offer valuable bound information to the counterfactual learning of ITE, which has never been used by existing ITE estimation methods. Hence, we further propose a CSA–Dragonnet with Embedded Prior Knowledge (CDNEPK) by formulating a unified expression of the prior knowledge given by RCTs or RCSs, inserting counterfactual prediction nets into CSA–Dragonnet and defining loss items based on the bounds for the ITE extracted from prior knowledge. Semi-synthetic data experiments showed that CDNEPK has superior performance. Real-world experiments indicated that CDNEPK can offer meaningful treatment advice.

A Survey of Deep Causal Models and Their Industrial Applications

“…From a Bayesian perspective, David [156,157] introduced a review and synthesis of the problem of causal inference in large scale educational assessments which requires the articulation of framework for causal effect estimation followed by a statistical approach that closely matches the framework and can yield the causal estimate of interest. Zhao et al [32] proposed the Residual Counterfactual Networks in an Intelligent Tutor System can decide which hint is more suitable for a speciőc student. However, the effectiveness of an intervention is necessarily multifaceted and complex effects differ between students, as a function of implementation [158], and, potentially, as a function of time and location.…”

“…Causal effect estimation is a process of drawing a conclusion about a causal connection based on the circumstances surrounding the occurrence of the effect and has a variety of applications in realworld scenarios [2]. For example, causal effect estimation of observational data in advertising[3ś6], developing recommender systems that are highly correlated with causal treatment effect estimates[7ś13], learning optimal treatment rules for patients in medicine[14ś16], estimation of ITE in reinforcement learning[17ś21], causal effect estimation tasks in natural language processing[22ś26], emerging computer vision and language interaction tasks[27ś31], education [32], and strategy resolutions[33ś37], etc.…”

Counterfactual Prediction Under Selective Confounding

“…Predictive algorithms that can extract valuable insights from rich case information have the potential to enable better decision-making and improve future outcomes. Counterfactual prediction algorithms have been widely adopted in healthcare [39] and are now increasingly used in other domains, such as child welfare [10] and education [47,48]. For example, in child welfare, practitioners often face challenges regarding whether to remove children from their homes and place them in foster care or keep them with their families [2].…”