2018
DOI: 10.1002/sim.7623
|View full text |Cite
|
Sign up to set email alerts
|

Some methods for heterogeneous treatment effect estimation in high dimensions

Abstract: When devising a course of treatment for a patient, doctors often have little quantitative evidence on which to base their decisions, beyond their medical education and published clinical trials. Stanford Health Care alone has millions of electronic medical records that are only just recently being leveraged to inform better treatment recommendations. These data present a unique challenge because they are high dimensional and observational. Our goal is to make personalized treatment recommendations based on the… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
5

Citation Types

1
187
0
2

Year Published

2018
2018
2024
2024

Publication Types

Select...
5

Relationship

1
4

Authors

Journals

citations
Cited by 151 publications
(190 citation statements)
references
References 36 publications
1
187
0
2
Order By: Relevance
“…The most common version of the former is inverse probability of treatment weighting, in which outcomes are weighted by the inverse probability of the observed treatment as follows (we drop the index i for clarity): Ytrue˜=ZYe(X)+(1Z)Y1e(X), where efalse(xfalse)double-struckPfalse[Z=1false|X=xfalse] is the famous propensity score that plays a central role in causal inference from observational data. It can be shown that the transformed outcome trueY˜ is an unbiased estimator for the treatment effect for that patient . This implies that we can use any off‐the‐shelf regression model, eg, random forests or neural networks, to estimate the mean of trueY˜ given X to arrive at an unbiased estimate for the treatment effect for a given patient.…”
Section: Introductionmentioning
confidence: 99%
See 4 more Smart Citations
“…The most common version of the former is inverse probability of treatment weighting, in which outcomes are weighted by the inverse probability of the observed treatment as follows (we drop the index i for clarity): Ytrue˜=ZYe(X)+(1Z)Y1e(X), where efalse(xfalse)double-struckPfalse[Z=1false|X=xfalse] is the famous propensity score that plays a central role in causal inference from observational data. It can be shown that the transformed outcome trueY˜ is an unbiased estimator for the treatment effect for that patient . This implies that we can use any off‐the‐shelf regression model, eg, random forests or neural networks, to estimate the mean of trueY˜ given X to arrive at an unbiased estimate for the treatment effect for a given patient.…”
Section: Introductionmentioning
confidence: 99%
“…This implies that we can use any off‐the‐shelf regression model, eg, random forests or neural networks, to estimate the mean of trueY˜ given X to arrive at an unbiased estimate for the treatment effect for a given patient. However, this approach has been shown to suffer from high variance in situations where the propensity score e ( x ) is close to 0 or 1, especially in the presence of large effect sizes . Much of the recent work on CATE estimation has thus focused on regression modeling of the conditional mean functions, μ 1 ( x ) and μ 0 ( x ), using flexible nonparametric models that are adapted from standard models from machine learning to specifically capture treatment effect heterogeneity.…”
Section: Introductionmentioning
confidence: 99%
See 3 more Smart Citations