On this page, you will find a series of tutorials on individualized treatment effect inference. Each tutorial offers a mini-syllabus in its own right, and is composed of a selection of modules tailored for specific purposes, time commitments, and levels of familiarity with AI and machine learning.
We will continue to adapt this page over time, adding new videos to introduce and highlight the latest advancements and research. Please continue to check here for new content!
Why is individualized treatment effect inference important?
A major challenge in the domain of healthcare is ascertaining whether a given treatment influences or determines an outcome—for instance, whether there is a survival benefit to prescribing a certain medication, such as the ability of a statin to lower the risk of cardiovascular disease.
Machine learning is capable of enabling truly personalized healthcare; this is what our lab calls “bespoke medicine.”
Bespoke medicine entails far more than providing predictions for individual patients: we also need to understand the effect of specific treatments on specific patients at specific times. This is what we call individualized treatment effect inference. It is a substantially more complex undertaking than prediction, and every bit as important.
Our lab has built a position of leadership in this area. We have defined the research agenda by outlining and addressing key complexities and challenges, and by laying the theoretical groundwork for model development. In our development of algorithms, we have identified and targeted an extensive range of potential clinical applications using both clinical trials and observational data as inputs.
If you’d like to learn more before starting our tutorial series, the link below will take you to a page providing an introduction to individualized treatment effect inference, as well as an overview of some key projects that have driven the entire research area forward.
How to use this tutorial series
This series is comprised of 6 tutorials on individualized treatment effect inference, each of which takes a different approach to the topic.
It’s possible that you’ll want to view the entire series, but unlikely that you’ll need to do so, since the applicability of each tutorial will depend on your familiarity with/intended usage of machine learning and AI, as well as the amount of time you have available.
An overview of the tutorials available can be found below.
A. Fundamentals
A primer offering a basic overview of some of the core concepts of individualized treatment effects. Touches on applications using time-series data.
Course length: short // Level of familiarity: minimal // Intended usage: familiarization
Watch this tutorial
Introduction, key concepts, and tutorial series structure
Mihaela van der Schaar provides an introduction to individualized treatment effect inference, and describes a number of key underlying concepts, challenges, and formalisms.
Additionally, Mihaela explains the structure and purpose of this series of tutorials.
Meta-learners for CATE estimation
Alicia Curth explains how to estimate heterogeneous treatment effects using any supervised learning method, using meta-learners for CATE estimation.
This presentation is partly adapted from an AISTATS 2021 paper entitled “Nonparametric Estimation of Heterogeneous Treatment Effects: From Theory to Learning Algorithms.”
The full paper (authored by Alicia Curth and Mihaela van der Schaar) can be found here.
ITE with time series data
Ioana Bica shares approaches to individualized treatment effect inference when working with time series (longitudinal) data.
This presentation is partly adapted from an ICLR 2021 paper entitled “Clairvoyance: A Pipeline Toolkit for Medical Time Series.” The full paper (authored by Daniel Jarrett, Jinsung Yoon, Ioana Bica, Zhaozhi Qian, Ari Ercole, and Mihaela van der Schaar) can be found here.
This presentation also features work featured in an ICML 2020 paper entitled “Time Series Deconfounder: Estimating Treatment Effects over Time in the Presence of Hidden Confounders.” This paper (authored by Ioana Bica, Ahmed Alaa, and Mihaela van der Schaar) can be found here.
B. Theory
A high-level introduction to the theoretical concepts behind individualized treatment effect inferences. Includes some approaches leveraging automated machine learning.
Course length: short // Level of familiarity: minimal // Intended usage: familiarization
Watch this tutorial
Introduction, key concepts, and tutorial series structure
Mihaela van der Schaar provides an introduction to individualized treatment effect inference, and describes a number of key underlying concepts, challenges, and formalisms.
Additionally, Mihaela explains the structure and purpose of this series of tutorials.
Meta-learners for CATE estimation
Alicia Curth explains how to estimate heterogeneous treatment effects using any supervised learning method, using meta-learners for CATE estimation.
This presentation is partly adapted from an AISTATS 2021 paper entitled “Nonparametric Estimation of Heterogeneous Treatment Effects: From Theory to Learning Algorithms.”
The full paper (authored by Alicia Curth and Mihaela van der Schaar) can be found here.
AutoML for ITE model selection
In this short video, Ahmed Alaa explains how a plug-in estimation approach can enable accurate prediction of the comparative performances of state-of-the-art causal inference methods applied to a given observational study.
This presentation is adapted from an ICML 2019 paper entitled “Validating Causal Inference Models via Influence Functions.”
The full paper (authored by Ahmed Alaa and Mihaela van der Schaar) can be found here.
C. Algorithms
An exploration of specific solutions to the complex challenges of developing algorithms for individualized treatment effect inference. Examples include accounting for hidden confounders, working with time-series data, inference for continuous treatments, and more.
Course length: medium // Level of familiarity: aware of common ML model types // Intended usage: algorithm development
Watch this tutorial
Introduction, key concepts, and tutorial series structure
Mihaela van der Schaar provides an introduction to individualized treatment effect inference, and describes a number of key underlying concepts, challenges, and formalisms.
Additionally, Mihaela explains the structure and purpose of this series of tutorials.
Meta-learners for CATE estimation
Alicia Curth explains how to estimate heterogeneous treatment effects using any supervised learning method, using meta-learners for CATE estimation.
This presentation is partly adapted from an AISTATS 2021 paper entitled “Nonparametric Estimation of Heterogeneous Treatment Effects: From Theory to Learning Algorithms.”
The full paper (authored by Alicia Curth and Mihaela van der Schaar) can be found here.
ITE with time series data
Ioana Bica shares approaches to individualized treatment effect inference when working with time series (longitudinal) data.
This presentation is partly adapted from an ICLR 2021 paper entitled “Clairvoyance: A Pipeline Toolkit for Medical Time Series.” The full paper (authored by Daniel Jarrett, Jinsung Yoon, Ioana Bica, Zhaozhi Qian, Ari Ercole, and Mihaela van der Schaar) can be found here.
This presentation also features work featured in an ICML 2020 paper entitled “Time Series Deconfounder: Estimating Treatment Effects over Time in the Presence of Hidden Confounders.” This paper (authored by Ioana Bica, Ahmed Alaa, and Mihaela van der Schaar) can be found here.
Learning overlapping representations for treatment effect estimation
Alexis Bellot introduces DKL-ITE, a counterfactual estimation model utilizing deep kernel regression and posterior regularization to learn individualized treatment effects using overlapping representation.
This presentation is partly adapted from an AISTATS 2020 paper entitled “Learning Overlapping Representations for the Estimation of Individualized Treatment Effects.”
The full paper (authored by Yao Zhang, Alexis Bellot, and Mihaela van der Schaar) can be found here.
Multi-cause hidden confounders over time
Ioana Bica discusses the challenge of individualized treatment effect estimation in the presence of multi-cause hidden confounders.
This presentation is adapted from work featured in an ICML 2020 paper entitled “Time Series Deconfounder: Estimating Treatment Effects over Time in the Presence of Hidden Confounders.”
The full paper (authored by Ioana Bica, Ahmed Alaa, and Mihaela van der Schaar) can be found here.
Synthetic control for longitudinal point treatments
Zhaozhi Qian explains and formalizes the problem of estimating treatment effects in the longitudinal setting. He then introduces SyncTwin, a method for producing interpretable and trustworthy results in this setting.
This presentation is adapted from a submitted paper entitled “SyncTwin: Transparent Treatment Effect Estimation under Temporal Confounding.”
Continuous treatments (dosage)
Ioana Bica introduces individualized treatment effect inference for continuous treatments (i.e. dosage).
This presentation is adapted from work featured in a NeurIPS 2020 paper entitled “Estimating the Effects of Continuous-valued Interventions using Generative Adversarial Networks.”
The full paper (authored by Ioana Bica, James Jordon, and Mihaela van der Schaar) can be found here.
Dynamic treatment regimes
Yao Zhang describes how individualized treatment effect inference methods can inform the creation of dynamic treatment regimes in healthcare.
This presentation is adapted from a NeurIPS 2020 paper entitled “Gradient regularized V-learning for dynamic treatment regimes.”
The full paper (authored by Yao Zhang and Mihaela van der Schaar) can be found here.
Organ transplantation
Jeroen Berrevoets describes how techniques for individualized treatment effect inference can be applied to the complex problem of allocating organs for transplantation.
This presentation is adapted from a NeurIPS 2020 paper entitled “OrganITE: Optimal transplant donor organ offering using an individual treatment effect.”
The full paper (authored by Jeroen Berrevoets, James Jordon, Ioana Bica, Alexander Gimson, and Mihaela van der Schaar) can be found here.
D. Applications
Showcase of potential real-world clinical applications of our ITE inference models. Examples include organ transplantation and dosage effect inference.
Course length: medium // Level of familiarity: minimal // Intended usage: real-world usage of models
Watch this tutorial
Introduction, key concepts, and tutorial series structure
Mihaela van der Schaar provides an introduction to individualized treatment effect inference, and describes a number of key underlying concepts, challenges, and formalisms.
Additionally, Mihaela explains the structure and purpose of this series of tutorials.
Organ transplantation
Jeroen Berrevoets describes how techniques for individualized treatment effect inference can be applied to the complex problem of allocating organs for transplantation.
This presentation is adapted from a NeurIPS 2020 paper entitled “OrganITE: Optimal transplant donor organ offering using an individual treatment effect.”
The full paper (authored by Jeroen Berrevoets, James Jordon, Ioana Bica, Alexander Gimson, and Mihaela van der Schaar) can be found here.
ITE with time series data
Ioana Bica shares approaches to individualized treatment effect inference when working with time series (longitudinal) data.
This presentation is partly adapted from an ICLR 2021 paper entitled “Clairvoyance: A Pipeline Toolkit for Medical Time Series.” The full paper (authored by Daniel Jarrett, Jinsung Yoon, Ioana Bica, Zhaozhi Qian, Ari Ercole, and Mihaela van der Schaar) can be found here.
This presentation also features work featured in an ICML 2020 paper entitled “Time Series Deconfounder: Estimating Treatment Effects over Time in the Presence of Hidden Confounders.” This paper (authored by Ioana Bica, Ahmed Alaa, and Mihaela van der Schaar) can be found here.
Synthetic control for longitudinal point treatments
Zhaozhi Qian explains and formalizes the problem of estimating treatment effects in the longitudinal setting. He then introduces SyncTwin, a method for producing interpretable and trustworthy results in this setting.
This presentation is adapted from a submitted paper entitled “SyncTwin: Transparent Treatment Effect Estimation under Temporal Confounding.”
Continuous treatments (dosage)
Ioana Bica introduces individualized treatment effect inference for continuous treatments (i.e. dosage).
This presentation is adapted from work featured in a NeurIPS 2020 paper entitled “Estimating the Effects of Continuous-valued Interventions using Generative Adversarial Networks.”
The full paper (authored by Ioana Bica, James Jordon, and Mihaela van der Schaar) can be found here.
E1. Advanced (overview)
Rapid runthrough of specific topics and problems in ITE inference.
Course length: short // Level of familiarity: familiar with a range of ML approaches and models // Intended usage: algorithm development
Watch this tutorial
AutoML for ITE model selection (short)
In this short video, Ahmed Alaa explains how a plug-in estimation approach can enable accurate prediction of the comparative performances of state-of-the-art causal inference methods applied to a given observational study.
A more detailed version of this presentation can be found in our “Advanced (full-length)” tutorial on ITE inference.
This presentation is adapted from an ICML 2019 paper entitled “Validating Causal Inference Models via Influence Functions.”
The full paper (authored by Ahmed Alaa and Mihaela van der Schaar) can be found here.
Synthetic control for longitudinal point treatments (short)
In this short video, Zhaozhi Qian explains and formalizes the problem of estimating treatment effects in the longitudinal setting. He then introduces SyncTwin, a method for producing interpretable and trustworthy results in this setting.
A more detailed version of this presentation can be found in our “Advanced (full-length)” tutorial on ITE inference.
This presentation is adapted from a submitted paper entitled “SyncTwin: Transparent Treatment Effect Estimation under Temporal Confounding.”
Organ transplantation (short)
In this short video, Jeroen Berrevoets describes how techniques for individualized treatment effect inference can be applied to the complex problem of allocating organs for transplantation.
A more detailed version of this presentation can be found in our “Advanced (full-length)” tutorial on ITE inference.
This presentation is adapted from a NeurIPS 2020 paper entitled “OrganITE: Optimal transplant donor organ offering using an individual treatment effect.”
The full paper (authored by Jeroen Berrevoets, James Jordon, Ioana Bica, Alexander Gimson, and Mihaela van der Schaar) can be found here.
E2. Advanced (in-depth)
In-depth examination of specific topics and problems in ITE inference.
Course length: medium-long // Level of familiarity: familiar with a range of ML approaches and models // Intended usage: algorithm development
Watch this tutorial
Introduction, key concepts, and tutorial series structure
Mihaela van der Schaar provides an introduction to individualized treatment effect inference, and describes a number of key underlying concepts, challenges, and formalisms.
Additionally, Mihaela explains the structure and purpose of this series of tutorials.
Learning overlapping representations for treatment effect estimation
Alexis Bellot introduces DKL-ITE, a counterfactual estimation model utilizing deep kernel regression and posterior regularization to learn individualized treatment effects using overlapping representation.
This presentation is partly adapted from an AISTATS 2020 paper entitled “Learning Overlapping Representations for the Estimation of Individualized Treatment Effects.”
The full paper (authored by Yao Zhang, Alexis Bellot, and Mihaela van der Schaar) can be found here.
AutoML for ITE model selection
In this short video, Ahmed Alaa explains how a plug-in estimation approach can enable accurate prediction of the comparative performances of state-of-the-art causal inference methods applied to a given observational study.
This presentation is adapted from an ICML 2019 paper entitled “Validating Causal Inference Models via Influence Functions.”
The full paper (authored by Ahmed Alaa and Mihaela van der Schaar) can be found here.
Multi-cause hidden confounders
Ioana Bica discusses the challenge of individualized treatment effect estimation in the presence of multi-cause hidden confounders.
This presentation is adapted from work featured in an ICML 2020 paper entitled “Time Series Deconfounder: Estimating Treatment Effects over Time in the Presence of Hidden Confounders.”
The full paper (authored by Ioana Bica, Ahmed Alaa, and Mihaela van der Schaar) can be found here.
Synthetic control for longitudinal point treatments
Zhaozhi Qian explains and formalizes the problem of estimating treatment effects in the longitudinal setting. He then introduces SyncTwin, a method for producing interpretable and trustworthy results in this setting.
This presentation is adapted from a submitted paper entitled “SyncTwin: Transparent Treatment Effect Estimation under Temporal Confounding.”
Continuous treatments (dosage)
Ioana Bica introduces individualized treatment effect inference for continuous treatments (i.e. dosage).
This presentation is adapted from work featured in a NeurIPS 2020 paper entitled “Estimating the Effects of Continuous-valued Interventions using Generative Adversarial Networks.”
The full paper (authored by Ioana Bica, James Jordon, and Mihaela van der Schaar) can be found here.
Dynamic treatment regimes
Yao Zhang describes how individualized treatment effect inference methods can inform the creation of dynamic treatment regimes in healthcare.
This presentation is adapted from a NeurIPS 2020 paper entitled “Gradient regularized V-learning for dynamic treatment regimes.”
The full paper (authored by Yao Zhang and Mihaela van der Schaar) can be found here.
Uncertainty quantification
Yao Zhang explains how to quantify uncertainties in black-box model predictions for individualized treatment effect inference.
This presentation is adapted from a 2020 paper entitled “AutoCP: Automated Pipelines for Accurate Prediction Intervals.”
The full paper (authored by Yao Zhang, William Zame, and Mihaela van der Schaar) can be found here.
Post-hoc analysis of clinical trials
Yao Zhang introduces an individualized treatment effect inference method that can produce interpretable identification of patient subgroups with similar covariates and treatment responses. This can be used to understand heterogeneous treatment effects in post-hoc analysis of clinical trials.
This presentation is adapted from a NeurIPS 2020 paper entitled “Robust recursive partitioning for heterogeneous treatment effects with uncertainty quantification.”
The full paper (Hyun-Suk Lee, Yao Zhang, William Zame, Cong Shen, Jang-Won Lee, and Mihaela van der Schaar) can be found here.
Organ transplantation
Jeroen Berrevoets describes how techniques for individualized treatment effect inference can be applied to the complex problem of allocating organs for transplantation.
This presentation is adapted from a NeurIPS 2020 paper entitled “OrganITE: Optimal transplant donor organ offering using an individual treatment effect.”
The full paper (authored by Jeroen Berrevoets, James Jordon, Ioana Bica, Alexander Gimson, and Mihaela van der Schaar) can be found here.
Our publications and engagement sessions
A full list of our papers on causal inference, individualized treatment effect inference, and related topics, can be found here.
We would also encourage you to stay abreast of ongoing developments in this and other areas of machine learning for healthcare by signing up to take part in one of our two streams of online engagement sessions.
If you are a practicing clinician, please sign up for Revolutionizing Healthcare, which is a forum for members of the clinical community to share ideas and discuss topics that will define the future of machine learning in healthcare (no machine learning experience required).
If you are a machine learning student, you can join our Inspiration Exchange engagement sessions, in which we introduce and discuss new ideas and development of new methods, approaches, and techniques in machine learning for healthcare.