Abstract:Causal inference plays an important role in explanatory analysis and decision making across various fields like statistics, marketing, health care, and education. Its main task is to estimate treatment effects and make intervention policies. Traditionally, most of the previous works typically focus on the binary treatment setting that there is only one treatment for a unit to adopt or not. However, in practice, the treatment can be much more complex, encompassing multi-valued, continuous, or bundle options. In this paper, we refer to these as complex treatments and systematically and comprehensively review the causal inference methods for addressing them. First, we formally revisit the problem definition, the basic assumptions, and their possible variations under specific conditions. Second, we sequentially review the related methods for multi-valued, continuous, and bundled treatment settings. In each situation, we tentatively divide the methods into two categories: those conforming to the unconfoundedness assumption and those violating it. Subsequently, we discuss the available datasets and open-source codes. Finally, we provide a brief summary of these works and suggest potential directions for future research.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is a review of methods for dealing with complex interventions in causal inference. Specifically, traditional causal inference methods mainly focus on binary intervention settings, that is, whether a unit receives a certain single intervention. However, in practical applications, interventions may be more complex, including multi - value, continuous or combined options. This review article systematically reviews causal inference methods for these complex interventions, aiming to provide a comprehensive framework covering all aspects from methodology to data sets and open - source code. ### Core problems of the paper 1. **Definition of complex interventions**: The article first formally redefines the problem of complex interventions, including basic assumptions and their variants under specific conditions. 2. **Methods for multi - value, continuous and combined interventions**: The article successively reviews relevant methods applicable to multi - value, continuous and combined intervention settings, and divides these methods into two categories: methods that conform to the unconfoundedness assumption and methods that violate the unconfoundedness assumption. 3. **Data sets and open - source code**: Discusses available data sets and open - source code, providing resources for practical applications for researchers. 4. **Future research directions**: Finally, the article summarizes the deficiencies of existing work and proposes potential directions for future research. ### Key challenges - **Controlling confounding bias**: When estimating the intervention effect, how to effectively control the influence of confounding factors is a core challenge. - **Dealing with unobserved confounding factors**: When there are unobserved confounding factors, how to adjust through methods such as proxy variables or instrumental variables. - **Diversity of complex interventions**: The diversity and complexity of multi - value, continuous and combined interventions require more flexible and powerful methods. ### Method overview - **Propensity Score (PS) method**: Used in binary intervention settings, it controls confounding bias by estimating the probability of receiving an intervention given covariates. - **Representation learning method**: Utilizes neural networks to learn the general representation of samples, and then predicts potential outcomes. - **Generative model method**: Uses Generative Adversarial Networks (GAN) or Variational Auto - Encoders (VAE) to generate potential outcomes in order to estimate the intervention effect. - **Tree - structured method**: Such as Classification And Regression Trees (CART) and Bayesian Additive Regression Trees (BART), divides the population into multiple subgroups through tree splitting, thereby estimating heterogeneous intervention effects. ### Conclusion The article provides a comprehensive reference framework for researchers by systematically reviewing causal inference methods for complex interventions. At the same time, it points out the limitations of current methods and future research directions, which is helpful to promote the further development of this field.

Causal Inference with Complex Treatments: A Survey

A Survey on Causal Inference

Causal inference for community-based multi-layered intervention study.

A Survey of Causal Inference Frameworks

Causal Inference in the Medical Domain: a Survey

Causal Inference and Related Statistical Methods

Causal Inference: A Missing Data Perspective

Causal Inference Meets Machine Learning

Causal Inference for a Hidden Treatment

Data Integration in Causal Inference

Causal Inference for Comprehensive Cohort Studies

Causality for Complex Continuous-time Functional Longitudinal Studies with Dynamic Treatment Regimes

Quantifying causal effects from observed data using quasi-intervention

Causal effect estimation for multivariate continuous treatments

An Automated Approach to Causal Inference in Discrete Settings

Causal Inference with Latent Variables: Recent Advances and Future Prospectives

Causal Inference Via Sparse Additive Models with Application to Online Advertising.

Causal Inference With Two Versions of Treatment

The Blessings of Multiple Treatments and Outcomes in Treatment Effect Estimation

Causal inference from treatment-control studies having an additional factor with unknown assignment mechanism

A flexible approach for causal inference with multiple treatments and clustered survival outcomes