Dataset Information

Recursive partitioning for heterogeneous causal effects.

ABSTRACT: In this paper we propose methods for estimating heterogeneity in causal effects in experimental and observational studies and for conducting hypothesis tests about the magnitude of differences in treatment effects across subsets of the population. We provide a data-driven approach to partition the data into subpopulations that differ in the magnitude of their treatment effects. The approach enables the construction of valid confidence intervals for treatment effects, even with many covariates relative to the sample size, and without "sparsity" assumptions. We propose an "honest" approach to estimation, whereby one sample is used to construct the partition and another to estimate treatment effects for each subpopulation. Our approach builds on regression tree methods, modified to optimize for goodness of fit in treatment effects and to account for honest estimation. Our model selection criterion anticipates that bias will be eliminated by honest estimation and also accounts for the effect of making additional splits on the variance of treatment effect estimates within each subpopulation. We address the challenge that the "ground truth" for a causal effect is not observed for any individual unit, so that standard approaches to cross-validation must be modified. Through a simulation study, we show that for our preferred method honest estimation results in nominal coverage for 90% confidence intervals, whereas coverage ranges between 74% and 84% for nonhonest approaches. Honest estimation requires estimating the model with a smaller sample size; the cost in terms of mean squared error of treatment effects for our preferred method ranges between 7-22%.

SUBMITTER: Athey S

PROVIDER: S-EPMC4941430 | biostudies-other | 2016 Jul

REPOSITORIES: biostudies-other

ACCESS DATA

Publications

Recursive partitioning for heterogeneous causal effects.

Athey Susan S Imbens Guido G

Proceedings of the National Academy of Sciences of the United States of America 20160701 27

In this paper we propose methods for estimating heterogeneity in causal effects in experimental and observational studies and for conducting hypothesis tests about the magnitude of differences in treatment effects across subsets of the population. We provide a data-driven approach to partition the data into subpopulations that differ in the magnitude of their treatment effects. The approach enables the construction of valid confidence intervals for treatment effects, even with many covariates re ...[more]

PMID: 27382149

Similar Datasets

Project description:The HapMap (haplotype map) projects have produced valuable genetic resources in life science research communities, allowing researchers to investigate sequence variations and conduct genome-wide association study (GWAS) analyses. A typical HapMap project may require sequencing hundreds, even thousands, of individual lines or accessions within a species. Due to limitations in current sequencing technology, the genotype values for some accessions cannot be clearly called. Additionally, allelic heterozygosity can be very high in some lines, causing genetic and sometimes phenotypic segregation in their descendants. Genetic and phenotypic segregation degrades the original accession's specificity and makes it difficult to distinguish one accession from another. Therefore, it is vitally important to determine and validate HapMap accessions before one conducts a GWAS analysis. However, to the best of our knowledge, there are no prior methodologies or tools that can readily distinguish or validate multiple accessions in a HapMap population. We devised a bioinformatics approach to distinguish multiple HapMap accessions using only a minimum number of genetic markers. First, we assign each candidate marker with a distinguishing score (DS), which measures its capability in distinguishing accessions. The DS score prioritizes those markers with higher percentages of homozygous genotypes (allele combinations), as they can be stably passed on to offspring. Next, we apply the "set-partitioning" concept to select optimal markers by recursively partitioning accession sets. Subsequently, we build a hierarchical decision tree in which a specific path represents the selected markers and the homogenous genotypes that can be used to distinguish one accession from others in the HapMap population. Based on these algorithms, we developed a web tool named MAD-HiDTree (Multiple Accession Distinguishment-Hierarchical Decision Tree), designed to analyze a user-input genotype matrix and construct a hierarchical decision tree. Using genetic marker data extracted from the Medicago truncatula HapMap population, we successfully constructed hierarchical decision trees by which the original 262 M. truncatula accessions could be efficiently distinguished. PCR experiments verified our proposed method, confirming that MAD-HiDTree can be used for the identification of a specific accession. MAD-HiDTree was developed in C/C++ in Linux. Both the source code and test data are publicly available at https://bioinfo.noble.org/MAD-HiDTree/.

Project description:Objective: To identify narcolepsy related regional brain activity alterations compared with matched healthy controls. To determine whether these changes can be used to distinguish narcolepsy from healthy controls by recursive partitioning analysis (RPA) and receiver operating characteristic (ROC) curve analysis. Method: Fifty-one narcolepsy with cataplexy patients (26 adults and 25 juveniles) and sixty matched heathy controls (30 adults and 30 juveniles) were recruited. All subjects underwent a resting-state functional magnetic resonance imaging scan. Fractional low-frequency fluctuations (fALFF) was used to investigate narcolepsy induced regional brain activity alterations among adult and juveniles, respectively. Recursive partitioning analysis and Receiver operating curve analysis was used to seek the ability of fALFF values within brain regions in distinguishing narcolepsy from healthy controls. Results: Compared with healthy controls, both adult and juvenile narcolepsy had lower fALFF values in bilateral medial superior frontal gyrus, bilateral inferior parietal lobule and supra-marginal gyrus. Compared with healthy controls, both adult and juvenile narcolepsy had higher fALFF values in bilateral sensorimotor cortex and middle temporal gyrus. Also juvenile narcolepsy had higher fALFF in right putamen and right thalamus compared with healthy controls. Based on RPA and ROC curve analysis, in adult participants, fALFF differences in right medial superior frontal gyrus can discriminate narcolepsy from healthy controls with high degree of sensitivity (100%) and specificity (88.9%). In juvenile participants, fALFF differences in left superior frontal gyrus can discriminate narcolepsy from healthy controls with moderate degree of sensitivity (57.1%) and specificity (88.9%). Conclusion: Compared with healthy controls, both the adult and juvenile narcolepsy showed overlap brain regions in fALFF differences after case-control comparison. Furthermore, we propose that fALFF value can be a helpful imaging biomarker in distinguishing narcolepsy from healthy controls among both adults and juveniles.

Dataset Information

Recursive partitioning for heterogeneous causal effects.

Publications

Recursive partitioning for heterogeneous causal effects.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets