What is Individualised Treatment Effect Inference?
Individualised treatment effects have been an area of significant focus for our lab’s researchers since 2016. Understanding how treatments affect individual patients is a crucial challenge in healthcare.
Traditional treatment guidelines often focus on the “average” patient, using insights derived from randomised control trials, but evidence shows that treatment responses can vary significantly from person to person. As we enter the era of precision medicine, the availability of vast amounts of data from electronic health records enables researchers to explore individualised treatment effects more deeply.
Our lab is dedicated to pioneering methods that estimate how specific treatments can impact outcomes for individual patients—such as assessing the survival benefits of weight loss for a 60-year-old with diabetes and heart disease. By moving away from one-size-fits-all approaches, we aim to empower clinical decision-makers with quantitative insights from machine learning, ensuring that every patient receives the most effective treatment tailored to their unique needs.
Current Research Highlights
Our Video Tutorials About ITE Estimation
Our latest Inspiration Exchange session
Prof Richard Peck and Alicia Curth talking about CATE
ODE Discovery for Longitudinal Heterogeneous Treatment Effects Inference (ICLR 2024)
Read how SyncTwin successfully reproduced the findings of a randomised controlled clinical trial (NeurIPS 2021)
Related Research Areas
Individualised treatment effect estimation is closely linked to clinical trials, causal deep learning, and digital twins. In clinical trials, ITE methods allow to refine insights beyond average treatment effects, tailoring findings to individual patient characteristics and enabling more personalised interventions.
Causal Deep Learning is a framework to categorise causal methods using a parametric and structural classification. Such a framework is useful to compare inference techniques in ITE, as well as the required assumptions for causal identification. This is necessary to avoid mistakes, and increase performance within a particular setting.
Digital twins, on the other hand, are dynamic computational models focused on generating multivariate patient trajectories to anticipate future states. While not centered on ITEs, they benefit from integrating ITE insights into their knowledge base to ensure causal relationships are respected, improving their predictive power and alignment with real-world treatment effects. These interconnected areas collectively drive more personalized and causally informed healthcare solutions.
Research Pillar on Clinical Trials
Research Pillar on Causal Deep Learning
Research Pillar on Digital Twins
Our Research
Estimating Heterogeneous Treatment Effects
To aid healthcare professionals in making treatment decisions, we need methods that can accurately forecast the effects of different treatments for individual patients.
However, most machine learning (ML) models are not designed for this. Traditional ML models make outcome predictions that are exactly in line with the distribution of the training data, but they struggle to reason about outcomes under interventions – key to understanding the effects of treatments. Unlike standard prediction, treatment effect estimation involves predicting what would happen under alternative treatment scenarios, making it a far more complex task.
Our lab develops ML algorithms specifically designed for the estimation of conditional average treatment effects (CATE), advancing these methods toward clinical use and addressing the unique challenges they present.
To Impute or not to Impute? Missing Data in Treatment Effect Estimation
J Berrevoets, F Imrie, T Kyono, J Jordon, M van der Schaar
ICLR 2023
Missing data in real-world scenarios complicates the accurate estimation of treatment effects, as traditional assumptions don’t account for the interaction between treatment variables and missing data. The authors introduce a new mechanism, mixed confounded missingness (MCM), showing that standard imputation methods can erase key information, while ignoring imputation leads to bias. They propose selective imputation, which determines which variables to impute, and demonstrate its effectiveness in improving treatment effect models across various algorithms.
Benchmarking Heterogeneous Treatment Effect Models through the Lens of Interpretability
J Crabbé, A Curth, I Bica, M van der Schaar
NeurIPS 2022
In this work, we use post-hoc feature importance methods to identify features that influence the model’s predictions. This allows us to evaluate treatment effect estimators along a new and important dimension that has been overlooked in previous work: We construct a benchmarking environment to empirically investigate the ability of personalised treatment effect models to identify predictive covariates – covariates that determine differential responses to treatment. Our benchmarking environment then enables us to provide new insight into the strengths and weaknesses of different types of treatment effects models as we modulate different challenges specific to treatment effect estimation – e.g. the ratio of prognostic to predictive information, the possible nonlinearity of potential outcomes and the presence and type of confounding.
Transfer Learning on Heterogeneous Feature Spaces for Treatment Effects Estimation
I Bica, M van der Schaar
NeurIPS 2022
Improving the estimation of conditional average treatment effects (CATE) for a specific group by using data from a different group with distinct features is a fascinating machine learning challenge. This transfer learning problem arises when we want to assess a treatment’s effectiveness for a new patient population with different clinical characteristics and limited data.
Our approach involves several key components:
- Representation learning: This helps handle the differences in feature spaces between the source and target domains.
- Multi-task architecture: We employ a flexible structure with shared and private layers to transfer information between potential outcome functions across domains.
By combining these building blocks, we develop transfer learning versions of standard CATE learners.
Identifiable Energy-based Representations: An Application to Estimating Heterogeneous Causal Effects
Y Zhang, J Berrevoets, M van der Schaar
AISTATS 2022
We propose an energy-based model (EBM) that learns a low-dimensional representation of the variables by employing a noise contrastive loss function. With our EBM we introduce a preprocessing step that alleviates the dimensionality curse for any existing learner developed for estimating CATEs. We prove that our EBM keeps the representations partially identifiable up to some universal constant, as well as having universal approximation capability.
These properties enable the representations to converge and keep the CATE estimates consistent. Experiments demonstrate the convergence of the representations, as well as show that estimating CATEs on our representations performs better than on the variables or the representations obtained through other dimensionality reduction methods.
Really Doing Great at Estimating CATE? A Critical Look at ML Benchmarking Practices in Treatment Effect Estimation
A Curth, D Svensson, J Weatherall, M van der Schaar
NeurIPS 2021
The machine learning (ML) toolbox for estimation of heterogeneous treatment effects from observational data is expanding rapidly — yet many of its algorithms have been evaluated only on a very limited set of semi-synthetic benchmark datasets. In this paper, we investigate current benchmarking practices for ML-based conditional average treatment effect (CATE) estimators, with special focus on empirical evaluation based on the popular semi-synthetic IHDP benchmark. We identify problems with current practice and highlight that semi-synthetic benchmark datasets, which (unlike real-world benchmarks used elsewhere in ML) do not necessarily reflect properties of real data, can systematically favour some algorithms over others – a fact that is rarely acknowledged but of immense relevance for interpretation of empirical results. Further, we argue that current evaluation metrics evaluate performance only for a small subset of possible use cases of CATE estimators, and discuss alternative metrics relevant for applications in personalised medicine. Additionally, we discuss alternatives for current benchmark datasets, and implications of our findings for benchmarking in CATE estimation.
On Inductive Biases for Heterogeneous Treatment Effect Estimation
A Curth, M van der Schaar
NeurIPS 2021
We investigate how to exploit structural similarities of an individual’s potential outcomes (POs) under different treatments to obtain better estimates of conditional average treatment effects in finite samples. Especially when it is unknown whether a treatment has an effect at all, it is natural to hypothesize that the POs are similar – yet, some existing strategies for treatment effect estimation employ regularization schemes that implicitly encourage heterogeneity even when it does not exist and fail to fully make use of shared structure. In this paper, we investigate and compare three end-to-end learning strategies to overcome this problem – based on regularization, reparametrization and a flexible multi-task architecture – each encoding inductive bias favoring shared behavior across POs. To build understanding of their relative strengths, we implement all strategies using neural networks and conduct a wide range of semi-synthetic experiments. We observe that all three approaches can lead to substantial improvements upon numerous baselines — including meta-learner strategies — and gain insight into performance differences across various experimental settings.
Nonparametric Estimation of Heterogeneous Treatment Effects: From Theory to Learning Algorithms
A Curth, M van der Schaar
AISTATS 2021
We investigated the strengths and weaknesses of a number of so-called meta-learners (model-agnostic strategies) that have been proposed in recent years. Such learners decompose the treatment effect estimation problem into separate sub-problems, each solvable using standard supervised learning methods. Choosing between different meta-learners in a data-driven manner is difficult, as it requires access to counterfactual information. Therefore, with the ultimate goal of building better understanding of the conditions under which some learners can be expected to perform better than others a priori, we theoretically analyze four broad meta-learning strategies which rely on plug-in estimation and pseudo-outcome regression. We highlight how this theoretical reasoning can be used to guide principled algorithm design and translate our analyses into practice by considering a variety of neural network architectures as base-learners for the discussed meta-learner strategies. In a simulation study, we showcase the relative strengths of the learners under different data-generating processes.
Treatment Effect Estimation Using Time-Series Data
While work on static treatment effects can already provide valuable insights, real-world data often involves more complex, time-dependent scenarios, such as when drug efficacy changes over time or patients receive multiple treatments at different stages.
Estimating treatment effects in these dynamic contexts allows us to understand how diseases evolve under different treatment strategies and to identify optimal intervention timings. Traditional static methods for treatment effect estimation struggle to capture these complexities, highlighting the need for approaches designed for time-varying treatments, which are able to take advantage of time-series data.
Our lab is at the forefront of designing such ML methods, advancing treatment effect estimation for more dynamic, real-world decision-making.
ODE Discovery for Longitudinal Heterogeneous Treatment Effects Inference
K Kacprzyk, S Holt, J Berrevoets, Z Qian, M van der Schaar
ICLR 2024
The machine learning community has focused on finding unbiased ways to infer treatment effects. Most solutions rely on neural networks to correct assignment bias, even in complex scenarios like high-dimensional or longitudinal settings.
In our study, we introduce a different approach specifically for longitudinal settings: closed-form ordinary differential equations (ODEs). Unlike neural networks, ODEs offer advantages like interpretability, handling irregular sampling, and different identification assumptions. Our main contribution is offering a new type of solution, potentially inspiring fresh innovations in treatment effects research.
We present our approach as a framework that can turn any ODE discovery method into a treatment effects method, opening doors for diverse approaches in this field.
Accounting For Informative Sampling When Learning to Forecast Treatment Outcomes Over Time
T Vanderschueren, A Curth, W Verbeke, M van der Schaar
ICML 2023
Informative sampling means that when instances are observed irregularly over time, the timing of observations isn’t random. Instead, it depends on factors such as the patient’s characteristics, past outcomes, and treatments received. This introduces a problem known as covariate shift, which can hinder accurate estimation of treatment outcomes if not properly addressed.
To tackle this, we introduce a general framework that treats informative sampling as a covariate shift problem. We propose a new method called TESAR-CDE, which implements this framework using Neural CDEs. Through simulations based on a clinical scenario, we demonstrate the effectiveness of our approach in learning treatment outcomes under informative sampling conditions.
Continuous-Time Modeling of Counterfactual Outcomes Using Neural Controlled Differential Equations
N Seedat, F Imrie, A Bellot, Z Qian, M van der Schaar
ICML 2022
Estimating counterfactual outcomes over time has the potential to unlock personalised healthcare by assisting decision-makers to answer ”what-if” questions. Existing causal inference approaches typically consider regular, discrete-time intervals between observations and treatment decisions and hence are unable to naturally model irregularly sampled data, which is common in practice. To handle arbitrary observation patterns, we interpret the data as samples from an underlying continuous-time process and model its latent trajectory explicitly using neural controlled differential equations. This leads to a new approach, the Treatment Effect Neural Controlled Differential Equation (TE-CDE), that allows the potential outcomes to be evaluated at any time point.
SyncTwin: Treatment Effect Estimation with Longitudinal Outcomes
Z Qian, Y Zhang, I Bica, A Wood, M van der Schaar
NeurIPS 2021
SyncTwin learns a patient-specific time-constant representation from the pre-treatment observations. SyncTwin issues counterfactual prediction of a target patient by constructing a synthetic twin that closely matches the target in representation. The reliability of the estimated treatment effect can be assessed by comparing the observed and synthetic pre-treatment outcomes.
The medical experts can interpret the estimate by examining the most important contributing individuals to the synthetic twin. In the real-data experiment, SyncTwin successfully reproduced the findings of a randomised controlled clinical trial using observational data, which demonstrates its usability in the complex real-world EHR.
Estimating Treatment Effects in More Complex Settings
Extending the applicability of treatment effect estimation methods requires addressing additional complexities that arise in real-world settings. As healthcare decisions become increasingly nuanced, it is essential to adapt these methods to account for various challenges that may impact their effectiveness. For instance, we focus on several critical areas, including time-to-event data, where outcomes unfold over time and may be influenced by competing events.
Our work also addresses policy analysis for evaluating healthcare interventions, the identification of patient subgroups in clinical trials that benefit most from treatments, and the development of generalised causal sensitivity analysis techniques to assess the robustness of estimates against assumptions and unmeasured confounding. These advancements broaden the scope of treatment effect estimation methods for diverse clinical and policy applications.
Quantifying Aleatoric Uncertainty of the Treatment Effect: A Novel Orthogonal Learner
V Melnychuk, S Feuerriegel, M van der Schaar
NeurIPS 2024
Estimating causal quantities from observational data is crucial for understanding the safety and effectiveness of medical treatments. However, to make reliable inferences, medical practitioners require not only estimating averaged causal quantities, such as the conditional average treatment effect, but also understanding the randomness of the treatment effect as a random variable. This randomness is referred to as aleatoric uncertainty and is necessary for understanding the probability of benefit from treatment or quantiles of the treatment effect. Yet, the aleatoric uncertainty of the treatment effect has received surprisingly little attention in the causal machine learning community. To fill this gap, we aim to quantify the aleatoric uncertainty of the treatment effect at the covariate-conditional level, namely, the conditional distribution of the treatment effect (CDTE). Unlike average causal quantities, the CDTE is not point identifiable without strong additional assumptions. As a remedy, we employ partial identification to obtain sharp bounds on the CDTE and thereby quantify the aleatoric uncertainty of the treatment effect. We then develop a novel, orthogonal learner for the bounds on the CDTE, which we call AU-learner. We further show that our AU-learner has several strengths in that it satisfies Neyman-orthogonality and is doubly robust. Finally, we propose a fully-parametric deep learning instantiation of our AU-learner.
Defining Expertise: Applications to Treatment Effect Estimation
A Hüyük, Q Wei, A Curth, M van der Schaar
ICLR 2024
Decision-makers, like doctors, use their expertise to guide actions, such as prescribing treatments based on predictions of outcomes. This expertise is valuable because treatments frequently prescribed by experts are likely more effective. However, in machine learning, we often overlook this expertise and don’t use it as a guiding principle, especially in treatment effect estimation research, where assumptions are typically limited to overlap.
In our study, we argue that decision-makers’ expertise can inform the design and selection of treatment effect estimation methods. We identify two types of expertise – predictive and prognostic – and show empirically that:
- The dominant type of expertise in a domain significantly impacts the performance of different estimation methods.
- We can predict the type of expertise in a dataset, providing a quantitative basis for selecting the most suitable models.
By recognising and leveraging decision-makers’ expertise, we can enhance the effectiveness of treatment effect estimation methods.
A Neural Framework for Generalized Causal Sensitivity Analysis
A Curth, A Hüyük, M van der Schaar
ICLR 2024
A core assumption in treatment effect estimation is that there are no unobserved confounders. However, often this assumption doesn’t hold in many real-world situations. To address this, we propose an innovative approach for estimating causal effects in the presence of unobserved confounding. Our neural framework, NeuralCSA, addresses previous limitations in causal sensitivity analysis by offering compatibility with various sensitivity models, treatment types, and causal queries. This generality is achieved by learning a latent distribution shift corresponding to a treatment intervention using two conditional normalising flows, and the approach is supported by theoretical guarantees. Advances such as NeuralCSA help extend the applicability of treatment effect estimation approaches to real-world scenarios.
Adaptive Identification of Populations with Treatment Benefit in Clinical Trials: Machine Learning Challenges and Solutions
A Curth, A Hüyük, M van der Schaar
ICML 2023
This work changes the outlook on adaptive clinical trials, focusing on the dynamic identification of patient subpopulations that respond positively to a specific treatment. While this problem has been extensively explored in biostatistics, existing methodologies have been somewhat constrained in their adaptivity. Our objective is to broaden the scope of these trial designs by incorporating insights from recent advancements in machine learning for adaptive and online experimentation.
One key distinction of the subpopulation selection problem is that we seek to identify groups that benefit from treatment, rather than just pinpointing the single subgroup with the largest effect, all within a limited budget. These nuances present unique challenges and criteria when crafting algorithmic solutions.
Drawing from these observations, we introduce AdaGGI and AdaGCPI, two meta-algorithms tailored for subpopulation construction. Through empirical analysis across various simulation scenarios, we evaluate their performance and gain valuable insights into their strengths and limitations across different settings. This exploration paves the way for more flexible and efficient adaptive clinical trials, ultimately enhancing patient care and treatment efficacy.
SurvITE: Learning Heterogeneous Treatment Effects from Time-to-Event Data
A Curth, C Lee, M van der Schaar
NeurIPS 2021
In this work, we study the problem of inferring heterogeneous treatment effects from time-to-event outcome data. While both the related problems of (i) estimating treatment effects for binary or continuous outcomes and (ii) predicting survival outcomes have been well studied in the recent machine learning literature, their combination – albeit of high practical relevance – has received considerably less attention. With the ultimate goal of reliably estimating the effects of treatments on instantaneous risk and survival probabilities, we focus on the problem of learning (discrete-time) treatment-specific conditional hazard functions. We find that unique challenges arise in this context due to a variety of covariate shift issues that go beyond a mere combination of well-studied confounding and censoring biases. We theoretically analyse their effects by adapting recent generalisation bounds from domain adaptation and treatment effect estimation to our setting and discuss implications for model design. We use the resulting insights to propose a novel deep learning method for treatment-specific hazard estimation based on balancing representations. We investigate performance across a range of experimental settings and empirically confirm that our method outperforms baselines by addressing covariate shifts from various sources
Policy Analysis using Synthetic Controls in Continuous-Time
A Bellot, M van der Schaar
PMLR 2021
Counterfactual estimation using synthetic controls is one of the most successful recent methodological developments in causal inference. Despite its popularity, the current description only considers time series aligned across units and synthetic controls expressed as linear combinations of observed control units. We propose a continuous-time alternative that models the latent counterfactual path explicitly using the formalism of controlled differential equations. This model is directly applicable to the general setting of irregularly-aligned multivariate time series and may be optimized in rich function spaces — thereby improving on some limitations of existing approaches.
Real-World Impact of Treatment Effect Estimation
The real-world applications of treatment effect estimation methods demonstrate their potential to significantly impact clinical practice and research. However, to fulfil their potential, these methods need to be well understood and adopted by the clinical community.
We aim to bridge the gap between research and practice, empowering healthcare professionals with tools that enhance their understanding of treatment effects and support evidence-based medicine. Through these applications, we contribute to shaping the future of medical practice and advancing personalised healthcare.
Causal machine learning for predicting treatment outcomes
S Feuerriegel, D Frauen, V Melnychuk, J Schweisthal, K Hess, A Curth, S Bauer, N Kilbertus, I S Kohane. M van der Schaar
Nature Medicine 2024
Causal machine learning (ML) provides adaptable, data-centric techniques for forecasting treatment outcomes like effectiveness and side effects, which aids in evaluating drug safety. A significant advantage of causal ML is its capability to estimate personalised treatment effects, enabling tailored clinical decision-making based on individual patient characteristics. This method can leverage both clinical trial data and real-world data from sources like clinical registries and electronic health records. However, it’s crucial to proceed with caution to avoid biased or inaccurate predictions.
In this Perspective, we highlight the advantages of causal ML over traditional statistical or ML methods and delineate its essential components and processes. Finally, we offer recommendations for ensuring the dependable application of causal ML and its successful integration into clinical practice.
Machine learning for clinical trials in the era of COVID-19
W R Zame, I Bica, C Shen, A Curth, H-S Lee, S Bailey, J Weatherall, D Wright, F Bretz. M van der Schaar
Statistics in Biopharmaceutical Research 2020
The COVID-19 pandemic has presented enormous challenge to clinical trials in particular, given the need for expedited development, approval, and distribution.
In a 2020 paper co-authored with some of our collaborators, published in Statistics in Biopharmaceutical Research, we identified ways in which machine learning can respond to the challenges inherent in clinical trials of COVID-19 treatments and vaccines.
We identified three key areas for support: ongoing clinical trials for non-COVID-19 drugs; clinical trials for repurposing drugs to treat COVID-19, and clinical trials for new drugs to treat COVID-19. Many of the research projects outlined above feature in the paper.
Machine Learning for Pharmacology
There are close links between our work on treatment effect estimation and pharmacology. In addition, our belief at the van der Schaar lab is that the integration of machine learning with pharmacology will unlock new frontiers in personalised medicine and clinical trials.
Click here to learn more.
Synthetic Model Combination: A new machine-learning method for pharmacometric model ensembling
A Chan, R Peck, M Gibbs, M van der Schaar
CPT: Pharmacometrics & Systems Pharmacology 2023
Machine-learning models perform notoriously poorly on data outside their training domain, however, due to a problem known as covariate shift, and so we argue that when ensembling models the weightings for individual instances must reflect their respective domains—in other words, models that are more likely to have seen information on that instance should have more attention paid to them. We introduce a method for such an instance-wise ensembling of models called Synthetic Model Combination (SMC), including a novel representation learning step for handling sparse high-dimensional domains.
We demonstrate the use of SMC on an example with dosing predictions for vancomycin, although emphasize the applicability of the method to any scenario involving the use of multiple models.
The potential and pitfalls of artificial intelligence in clinical pharmacology
M Johnson, M Patel, A Phipps, M van der Schaar, D Boulton, M Gibbs
CPT: Pharmacometrics & Systems Pharmacology 2023
Artificial intelligence (AI) involves using data and algorithms to perform activities normally achieved through human intelligence. AI and its key component machine learning contextualize data and enhance decision making to transform how we operate, discover, and develop drugs. Transforming clinical pharmacology (CP) as AI‐augmented CP (AI/CP) requires an ecosystem including digitized data collection, standardized processes, complementary technologies, and an ethical framework. This commentary aims to highlight the future perspectives of AI/CP in drug development.
Integrating Expert ODEs into Neural ODEs: Pharmacology and Disease Progression
Z Qian, W R Zame, L M Fleuren, P Elbers, M van der Schaar
NeurIPS 2022
Pharmacological models describe the dynamics of carefully-chosen medically meaningful variables in terms of systems of Ordinary Differential Equations (ODEs). However, these models only describe a limited collection of variables, and these variables are often not observable in clinical environments. To close this gap, we propose the latent hybridisation model (LHM) that integrates a system of expert-designed ODEs with machine-learned Neural ODEs to fully describe the dynamics of the system and to link the expert and latent variables to observable quantities.
We evaluated LHM on synthetic data as well as real-world intensive care data of COVID-19 patients. LHM consistently outperforms previous works, especially when few training samples are available such as at the beginning of the pandemic.
Estimating the Effects of Continuous-valued Interventions using Generative Adversarial Networks
I Bica, J Jordon, M van der Schaar
NeurIPS 2020
We tackle the problem of estimating the effect of discrete interventions from observational data by building on a modification of the generative adversarial networks (GANs) framework. Our model, SCIGAN, is flexible and capable of simultaneously estimating counterfactual outcomes for several different continuous interventions. The key idea is to use a significantly modified GAN model to learn to generate counterfactual outcomes, which can then be used to learn an inference model, using standard supervised methods, capable of estimating these counterfactuals for a new sample. To address the challenges presented by shifting to continuous interventions, we propose a novel architecture for our discriminator – we build a hierarchical discriminator that leverages the structure of the continuous intervention setting.
From Real-World Patient Data to Individualized Treatment Effects Using Machine Learning: Current and Future Methods to Address Underlying Challenges
I Bica, A M Alaa, C Lambert, M van der Schaar
Clinical Pharmacology & Therapeutics 2020
We describe the modeling choices of the state-of-the-art machine learning methods for causal inference, developed for estimating treatment effects both in the cross-section and longitudinal settings. Additionally, we highlight future research directions that could lead to achieving the full potential of leveraging EHRs and machine learning for making individualized treatment recommendations. We also discuss how experimental data from RCTs and Pharmacometric and Quantitative Systems Pharmacology approaches can be used to not only improve machine learning methods, but also provide ways for validating them.
Our Software for Clinical Impact
Clairvoyance
CRN
Te-CDE
CATENets
