Learning Triggers for Heterogeneous Treatment Effects

The causal effect of a treatment can vary from person to person based on their individual characteristics and predispositions. Mining for patterns of individual-level effect differences, a problem known as heterogeneous treatment effect estimation, has many important applications, from precision medicine to recommender systems. In this paper we define and study a variant of this problem in which an individuallevel threshold in treatment needs to be reached, in order to trigger an effect. One of the main contributions of our work is that we do not only estimate heterogeneous treatment effects with fixed treatments but can also prescribe individualized treatments. We propose a tree-based learning method to find the heterogeneity in the treatment effects. Our experimental results on multiple datasets show that our approach can learn the triggers better than existing approaches.

Download Full-text

Estimating Causal Treatment Effects of Femoral and Tibial Derotational Osteotomies on Foot Progression in Children with Cerebral Palsy

10.1101/2021.03.04.21252476 ◽

2021 ◽

Author(s):

Michael H. Schwartz ◽

Hans Kainz ◽

Andrew G. Georgiadis

Keyword(s):

Cerebral Palsy ◽

Treatment Effects ◽

Causal Effect ◽

Machine Learning Algorithms ◽

Significant Heterogeneity ◽

Tibial Torsion ◽

Heterogeneous Treatment Effects ◽

Causal Treatment ◽

No Gold Standard

AbstractBackgroundFoot progression deviations are a common and important problem for children with CP. Tibial and femoral derotational osteotomies (TDO, FDO) are used to treat foot progression deviations, but outcomes are unpredictable. The available evidence for the causal effects of TDO and FDO is limited and weak, and thus modeling approaches are needed.MethodsWe queried our clinical database for individuals with a diagnosis of cerebral palsy (CP) who were less than 18 years old and had baseline and follow up gait data collected within a 3-year time span. We then used the Bayesian Causal Forest (BCF) algorithm to estimate the causal treatment effects of TDO and FDO on foot progression deviations (separate models). We examined average treatment effects and heterogeneous treatment effects (HTEs) with respect to clinically relevant covariates.ResultsThe TDO and FDO models were able to accurately predict follow-up foot progression (r2 ∼0.7, RMSE ∼8°). The estimated causal effect of TDO was bimodal and exhibited significant heterogeneity with respect to baseline levels of foot progression and tibial torsion as well as changes in tibial torsion at follow-up. The estimated causal effect of FDO was unimodal and largely homogeneous with respect to baseline or change characteristics.ConclusionsThis study demonstrated the potential for causal machine-learning algorithms to impact treatment in children with CP. The causal model is accurate and appears sensible – though no gold-standard exists for validating the causal estimates. The model results can provide guidance for planning surgical corrections, and partly explain unsatisfactory outcomes observed in prior observational clinical studies.

Download Full-text

A Unified Survey of Treatment Effect Heterogeneity Modelling and Uplift Modelling

ACM Computing Surveys ◽

10.1145/3466818 ◽

2022 ◽

Vol 54 (8) ◽

pp. 1-36

Author(s):

Weijia Zhang ◽

Jiuyong Li ◽

Lin Liu

Keyword(s):

Treatment Effect ◽

Treatment Effects ◽

Causal Effect ◽

Personalised Medicine ◽

Online Marketing ◽

Heterogeneous Treatment Effects ◽

Source Codes ◽

Outcome Framework ◽

Treatment Effect Heterogeneity ◽

Effect Heterogeneity

A central question in many fields of scientific research is to determine how an outcome is affected by an action, i.e., to estimate the causal effect or treatment effect of an action. In recent years, in areas such as personalised healthcare, sociology, and online marketing, a need has emerged to estimate heterogeneous treatment effects with respect to individuals of different characteristics. To meet this need, two major approaches have been taken: treatment effect heterogeneity modelling and uplifting modelling. Researchers and practitioners in different communities have developed algorithms based on these approaches to estimate the heterogeneous treatment effects. In this article, we present a unified view of these two seemingly disconnected yet closely related approaches under the potential outcome framework. We provide a structured survey of existing methods following either of the two approaches, emphasising their inherent connections and using unified notation to facilitate comparisons. We also review the main applications of the surveyed methods in personalised marketing, personalised medicine, and sociology. Finally, we summarise and discuss the available software packages and source codes in terms of their coverage of different methods and applicability to different datasets, and we provide general guidelines for method selection.

Download Full-text

Estimating heterogeneous treatment effects by balancing heterogeneity and fitness

10.1101/333278 ◽

2018 ◽

Author(s):

Weijia Zhang ◽

Thuc Le ◽

Lin Liu ◽

Jiuyong Li

Keyword(s):

Treatment Effects ◽

Recursive Partitioning ◽

Superior Performance ◽

Breast Cancer Patients ◽

Computationally Efficient ◽

Gene Expressions ◽

Heterogeneous Treatment Effects ◽

Multi Objective ◽

Biological And Medical Applications ◽

Better Than

AbstractEstimating heterogeneous treatment effects is an important problem in many medical and biological applications since treatments may have different effects on the prognoses of different patients. Recently, several recursive partitioning methods have been proposed to identify the subgroups that with different responds to a treatment, and they rely on a fitness criterion to minimize the error between the estimated treatment effects and the unobservable true effects. In this paper, we propose that a heterogeneity criterion, which maximizes the differences of treatment effects among the subgroups, also needs to be considered. Moreover, we show that better performances can be achieved when the fitness and the heterogeneous criteria are considered simultaneously. Selecting the optimal splitting points then becomes a multi-objective problem; however, a solution that achieves optimal in both aspects are often not available. To solve this problem, we propose a multi-objective splitting procedure to balance both criteria. The proposed procedure is computationally efficient and fits naturally into the existing recursive partitioning framework. Experimental results show that the proposed multi-objective approach performs consistently better than existing ones.Author summaryThe effects of a treatment are often not the same for different individuals with different gene expressions. Learning to predict the heterogeneous treatment effects from clinical and expression data is an important step towards personalized medical treatment. Existing computational methods are not ideal for the task because they do not address the interpretability of the model and do not consider the limited sample sizes in biological and medical applications. Our method addresses these issues and achieves superior performance in analyzing the treatment effects of radiotherapy on breast cancer patients.

Download Full-text

Constrained Instruments and their Application to Mendelian Randomization with Pleiotropy

10.1101/227454 ◽

2017 ◽

Cited By ~ 3

Author(s):

Lai Jiang ◽

Karim Oualkacha ◽

Vanessa Didelez ◽

Antonio Ciampi ◽

Pedro Rosa ◽

...

Keyword(s):

Instrumental Variables ◽

Genetic Variants ◽

Instrumental Variable ◽

Genetic Variant ◽

Mendelian Randomization ◽

Causal Effect ◽

Naive Method ◽

Effect Estimation ◽

Selection Of ◽

Better Than

AbstractIn Mendelian randomization (MR), genetic variants are used to construct instrumental variables, which enable inference about the causal relationship between a phenotype of interest and a response or disease outcome. However, standard MR inference requires several assumptions, including the assumption that the genetic variants only influence the response through the phenotype of interest. Pleiotropy occurs when a genetic variant has an effect on more than one phenotype; therefore, a pleiotropic genetic variant may be an invalid instrumental variable. Hence, a naive method for constructing instrumental variables may lead to biased estimation of the causality between the phenotype and the response. Here, we present a set of intuitive methods (Constrained Instrumental Variable methods [CIV]) to construct valid instrumental variables and perform adjusted causal effect estimation when pleiotropy exists, focusing particularly on the situation where pleiotropic phenotypes have been measured. Our approach includes an automatic and valid selection of genetic variants when building the instrumental variables. We also provide details of the features of many existing methods, together with a comparison of their performance in a large series of simulations. CIV methods performed consistently better than many comparators across four different pleiotropic violations of the MR assumptions. We analyzed data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) Mueller et al. (2005) to disentangle causal relationships of several biomarkers with AD progression. The results showed that CIV methods can provide causal effect estimates, as well as selection of valid instruments while accounting for pleiotropy.

Download Full-text

Quasi-oracle estimation of heterogeneous treatment effects

Biometrika ◽

10.1093/biomet/asaa076 ◽

2020 ◽

Cited By ~ 2

Author(s):

X Nie ◽

S Wager

Keyword(s):

Objective Function ◽

Treatment Effects ◽

Penalized Regression ◽

Loss Minimization ◽

Minimization Method ◽

Heterogeneous Treatment Effects ◽

Marginal Effects ◽

Optimal Resource ◽

Data Adaptive ◽

Effect Estimation

Summary Flexible estimation of heterogeneous treatment effects lies at the heart of many statistical applications, such as personalized medicine and optimal resource allocation. In this article we develop a general class of two-step algorithms for heterogeneous treatment effect estimation in observational studies. First, we estimate marginal effects and treatment propensities to form an objective function that isolates the causal component of the signal. Then, we optimize this data-adaptive objective function. The proposed approach has several advantages over existing methods. From a practical perspective, our method is flexible and easy to use: in both steps, any loss-minimization method can be employed, such as penalized regression, deep neural networks, or boosting; moreover, these methods can be fine-tuned by cross-validation. Meanwhile, in the case of penalized kernel regression, we show that our method has a quasi-oracle property. Even when the pilot estimates for marginal effects and treatment propensities are not particularly accurate, we achieve the same error bounds as an oracle with prior knowledge of these two nuisance components. We implement variants of our approach based on penalized regression, kernel ridge regression, and boosting in a variety of simulation set-ups, and observe promising performance relative to existing baselines.

Download Full-text

Mendelian Randomization with Instrumental Variable Synthesis (IVY)

10.1101/657775 ◽

2019 ◽

Author(s):

Zhaobin Kuang ◽

Aldo Cordova-Palomera ◽

Fred Sala ◽

Sen Wu ◽

Jared Dunnmon ◽

...

Keyword(s):

Graphical Models ◽

Instrumental Variable ◽

Mendelian Randomization ◽

Causal Effect ◽

Uk Biobank ◽

Inference Method ◽

Real World Data ◽

Individual Level ◽

Artery Disease ◽

Effect Estimation

SUMMARYMendelian Randomization (MR) is an important causal inference method primarily used in biomedical research. This work applies contemporary techniques in machine learning to improve the robustness and power of traditional MR tools. By denoising and combining candidate genetic variants through techniques from unsupervised probabilistic graphical models, an influential latent instrumental variable is constructed for causal effect estimation. We present results on identifying relationships between biomarkers and the occurrence of coronary artery disease using individual-level real-world data from UK-BioBank via the proposed method. The approach, termed Instrumental Variable sYnthesis (IVY) is proposed as a complement to current methods, and is able to improve results based on allele scoring, particularly at moderate sample sizes.

Download Full-text

Intuitive Prosociality: Heterogeneous Treatment Effects or False Positive?

10.31219/osf.io/hrx2y ◽

2019 ◽

Cited By ~ 1

Author(s):

Eirik Strømland ◽

Gaute Torsvik

Keyword(s):

Treatment Effects ◽

Time Pressure ◽

Causal Effect ◽

Dictator Game ◽

Heterogeneous Treatment Effects ◽

Conditional Effects ◽

Norwegian Population ◽

General Test ◽

Intuitive Decision Making ◽

Effect Heterogeneity

Heterogenous treatment effects make it difficult to extrapolate from one research setting to another. However, what appears to be differences in effects across subpopulations may simply be false positives. This paper uses a representative sample of the Norwegian population (N = 1390) to systematically test for several proposed sources of heterogeneity in the literature on intuitive prosociality – a literature with large variation in results, which some researchers claim results from heterogeneity in the underlying effect. We use time pressure to induce intuitive decision making, and exogenously vary participants’ experience with the game. We find no overall effect of time constraints on dictator game for inexperienced subjects, and there is no evidence for an interaction effect between subject experience and the effect of time pressure. As a more general test of treatment effect heterogeneity, we consider the full distribution of treatment effects conditional on various proposed moderators in the literature. The distribution of conditional effects is consistent with no causal effect of time pressure on giving and no systematic heterogeneity in the underlying effect across subpopulations.

Download Full-text