DOCSLIB.ORG

The Practice of Causal Inference in Cancer Epidemiology

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
The Practice of Causal Inference in Cancer Epidemiology Vol. 5. 303-31 1, April 1996 Cancer Epidemiology, Biomarkers & Prevention 303 Review The Practice of Causal Inference in Cancer Epidemiology Douglas L. Weedt and Lester S. Gorelic causes lung cancer (3) and to guide causal inference for occu- Preventive Oncology Branch ID. L. W.l and Comprehensive Minority pational and environmental diseases (4). From 1965 to 1995, Biomedical Program IL. S. 0.1. National Cancer Institute, Bethesda, Maryland many associations have been examined in terms of the central 20892 questions of causal inference. Causal inference is often practiced in review articles and editorials. There, epidemiologists (and others) summarize evi- Abstract dence and consider the issues of causality and public health Causal inference is an important link between the recommendations for specific exposure-cancer associations. practice of cancer epidemiology and effective cancer The purpose of this paper is to take a first step toward system- prevention. Although many papers and epidemiology atically reviewing the practice of causal inference in cancer textbooks have vigorously debated theoretical issues in epidemiology. Techniques used to assess causation and to make causal inference, almost no attention has been paid to the public health recommendations are summarized for two asso- issue of how causal inference is practiced. In this paper, ciations: alcohol and breast cancer, and vasectomy and prostate we review two series of review papers published between cancer. The alcohol and breast cancer association is timely, 1985 and 1994 to find answers to the following questions: controversial, and in the public eye (5). It involves a common which studies and prior review papers were cited, which exposure and a common cancer and has a large body of em- causal criteria were used, and what causal conclusions pirical evidence; over 50 studies and over a dozen reviews have and public health recommendations ensued. Fourteen appeared. The association between vasectomy and prostate published reviews on alcohol and breast cancer and 6 cancer also involves a common exposure and a common cancer. published reviews on vasectomy and prostate cancer were It is a relatively recent controversy with a smaller empirical examined. For both series of reviews, nearly all available literature and relatively few reviews. published studies were cited except for ecological studies The intent of this “review of reviews” is not to criticize and prior reviews. Sources of causal criteria were often individual reviewers nor to make judgments regarding associ- not provided. When they appeared, all citations were ations. The limitations of making summary statements from either the 1964 Surgeon General’s report or works of small selected samples are recognized. Recommendations for Austin Bradford Hill. Reviews often excluded and possible improvements to the practice of causal inference are sometimes altered criteria without giving reasons for offered. These are based upon the summary findings and upon these changes. The criteria of consistency and strength of our understanding of two areas of related research: the methods association were almost always used accompanied by and theory of causal inference and the scientific quality of dose-response and biological plausibility in a majority of review papers. reviews. The criterion of temporality, considered by many methodologists to be a necessary causal condition, was infrequently used. Confounding and bias were often Materials and Methods added considerations. Public health recommendations were not discussed in nearly one-half of the reviews. The MEDLINE and CANCERLIT data bases were searched from 1977 through June 1994, using the keywords “alcohol and breast cancer” and “vasectomy and prostate cancer.” Reference Introduction lists of primary research studies and reviews identified were What causes cancer? How much and what kinds of evidence perused for mention of relevant articles. Tables of contents lead to causal conclusions? When should public health recom- from major medical, cancer, public health, and epidemiology mendations be made? These are central questions of causal journals available at the libraries of the NIH were also inference in cancer epidemiology. Causal inference has been an examined. important methodological focus for at least half a century (1, 2). For both associations, three types of review papers were A classic point of reference is the mid- I 960s, when two papers, found: editorials in which a review of existing literature was one by a committee appointed by the Surgeon General (3) and included and either cause or public health recommendations the other by Austin Bradford Hill (4), listed what are usually were considered, reviews devoted to the specific association, called causal criteria. These criteria are shown in Table I . At and “mini-reviews” subsumed within reviews of more general the time, they were used to support the claim that smoking issues. Typically, these mini-reviews were relatively brief sec- tions in papers on alcohol and cancer, diet and cancer, breast (and prostate) cancer epidemiology, and male sterilization. Received 5/23/95; revised I 2/4/95; accepted I 2/5/95. Only editorials and reviews devoted solely to the specific The costs of publication of this article were defrayed in part by the payment of associations are included in this paper, with one exception, an page charges. This article must therefore be hereby marked advertisement in IARC monograph on alcohol and cancer with a review of the accordance with 18 U.S.C. Section 1734 solely to indicate this fact. alcohol and breast cancer literature (6). To whom requests for reprints should be addressed, at National Cancer Institute, EPS T-4I, 9000 Rockville Pike, Bethesda, MD 20892. Phone: (301) 496-8640; Editorials and reviews are subsequently referred to as Fax: (3))l ) 402-0863. “reviews.” For each, the following questions were addressed: Downloaded from cebp.aacrjournals.org on September 27, 2021. © 1996 American Association for Cancer Research. .304 Review: Causal Inference in Cancer Epidemiology Table I Causal criteria as described by A. B. Hill (4) and by the Surgeon (3); in both, the original five criteria were used. The remaining General’s committee (3) four reviews cited works of Austin Bradford Hill (4, 97). In only one instance was a list of nine criteria used in the assess- Consistency” ment (80), although “epidemiological sense” replaced the cri- Strength” Dose-response tenon ofcoherence ofHill. Chu (76) cited Hill (4) but used only Biological plausibility four of his nine criteria: consistency, strength, biological cred- Specificity” ibility (i.e., plausibility), and experimentation. Longnecker (63) Temporality” also cited Hill (4) and used a subset comprising consistency, Experimentation strength, dose-response, and biological plausibility. Lowenfels Coherence’ et a!. (75) cited Hill (97) and used five criteria. In the eight Analogy reviews in which no source for criteria is cited, the number of “ Criteria described by the Surgeon General’s committee (3). criteria used varied from one (consistency; Ref. 72) to six of the original nine of Hill (4, 77). Overall Use of Causal Criteria and Other Considerations. Because the 1964 Surgeon General’s list ofcausal criteria form K Which studies are included? (Asked as a check for a subset of the 1965 criteria of Hill, and because reviewers systematic exclusion.) citing no source nevertheless used many of the same criteria, ( Which causal criteria are used and what sources for the overall use of criteria can be examined. Table 3 shows that these criteria are cited? the criterion of consistency was always used and strength of ( Are other considerations (e.g. , study design, bias, and association was nearly always used in this series of reviews. confounding) featured? Dose-response and biological plausibility were also used rela- ( What causal conclusions and public health recommen- tively frequently. In only 4 of 14 reviews was the criterion of dations are made? temporality discussed. The need for a randomized trial and its K What is the relationship of a conclusion or recommen- ethical and practical feasibility was considered in 3 of the 14 dation to the results of original research studies published by reviews. Coherence and analogy were rarely used. Table 2 the same author? reveals that bias (7 of 14) and confounding (I I of 14) were the K Do reviewers cite the methods, conclusions, or recom- most common additional considerations in this series. mendations of earlier reviews? Causal Conclusions. The question of cause was addressed in 10 of 14 reviews. None claimed cause. Conclusions included Results reasonably clear rejections of the idea (78, 80, 81) and those Akohol and Breast Cancer that left the issue undecided (70, 73, 74, 77, 82). Two reviewers The consumption of alcoholic beverages has been linked with appeared to lean in the direction of concluding cause. Long- breast cancer since 1974 (7). In the ensuing two decades, ecolog- necker (63) for example, argued that the “evidence appears to ical studies (8-1 1), case-control studies (12-50), and cohort stud- be growing stronger. “ Hiatt (79) argued that the data were “not ies (5 1-60) have been published, representing a total ofjust over yet sufficient to conclude .. causation.” 50 studies. Metaanalyses have also appeared (61-63). Reports Public Health Recommendations. Public health recommen- from a workshop (64-67) and papers discussing aspects of pub- dations were addressed in 8 of 14 reviews.
Load more

Recommended publications
  • Statistics and Causal Inference (With Discussion)
    Applied Statistics Lecture Notes Kosuke Imai Department of Politics Princeton University February 2, 2008 Making statistical inferences means to learn about what you do not observe, which is called parameters, from what you do observe, which is called data. We learn the basic principles of statistical inference from a perspective of causal inference, which is a popular goal of political science research. Namely, we study statistics by learning how to make causal inferences with statistical methods. 1 Statistical Framework of Causal Inference What do we exactly mean when we say “An event A causes another event B”? Whether explicitly or implicitly, this question is asked and answered all the time in political science research. The most commonly used statistical framework of causality is based on the notion of counterfactuals (see Holland, 1986). That is, we ask the question “What would have happened if an event A were absent (or existent)?” The following example illustrates the fact that some causal questions are more difficult to answer than others. Example 1 (Counterfactual and Causality) Interpret each of the following statements as a causal statement. 1. A politician voted for the education bill because she is a democrat. 2. A politician voted for the education bill because she is liberal. 3. A politician voted for the education bill because she is a woman. In this framework, therefore, the fundamental problem of causal inference is that the coun- terfactual outcomes cannot be observed, and yet any causal inference requires both factual and counterfactual outcomes. This idea is formalized below using the potential outcomes notation.
    [Show full text]
  • Bayesian Causal Inference
    Bayesian Causal Inference Maximilian Kurthen Master’s Thesis Max Planck Institute for Astrophysics Within the Elite Master Program Theoretical and Mathematical Physics Ludwig Maximilian University of Munich Technical University of Munich Supervisor: PD Dr. Torsten Enßlin Munich, September 12, 2018 Abstract In this thesis we address the problem of two-variable causal inference. This task refers to inferring an existing causal relation between two random variables (i.e. X → Y or Y → X ) from purely observational data. We begin by outlining a few basic definitions in the context of causal discovery, following the widely used do-Calculus [Pea00]. We continue by briefly reviewing a number of state-of-the-art methods, including very recent ones such as CGNN [Gou+17] and KCDC [MST18]. The main contribution is the introduction of a novel inference model where we assume a Bayesian hierarchical model, pursuing the strategy of Bayesian model selection. In our model the distribution of the cause variable is given by a Poisson lognormal distribution, which allows to explicitly regard discretization effects. We assume Fourier diagonal covariance operators, where the values on the diagonal are given by power spectra. In the most shallow model these power spectra and the noise variance are fixed hyperparameters. In a deeper inference model we replace the noise variance as a given prior by expanding the inference over the noise variance itself, assuming only a smooth spatial structure of the noise variance. Finally, we make a similar expansion for the power spectra, replacing fixed power spectra as hyperparameters by an inference over those, where again smoothness enforcing priors are assumed.
    [Show full text]
  • Articles Causal Inference in Civil Rights Litigation
    VOLUME 122 DECEMBER 2008 NUMBER 2 © 2008 by The Harvard Law Review Association ARTICLES CAUSAL INFERENCE IN CIVIL RIGHTS LITIGATION D. James Greiner TABLE OF CONTENTS I. REGRESSION’S DIFFICULTIES AND THE ABSENCE OF A CAUSAL INFERENCE FRAMEWORK ...................................................540 A. What Is Regression? .........................................................................................................540 B. Problems with Regression: Bias, Ill-Posed Questions, and Specification Issues .....543 1. Bias of the Analyst......................................................................................................544 2. Ill-Posed Questions......................................................................................................544 3. Nature of the Model ...................................................................................................545 4. One Regression, or Two?............................................................................................555 C. The Fundamental Problem: Absence of a Causal Framework ....................................556 II. POTENTIAL OUTCOMES: DEFINING CAUSAL EFFECTS AND BEYOND .................557 A. Primitive Concepts: Treatment, Units, and the Fundamental Problem.....................558 B. Additional Units and the Non-Interference Assumption.............................................560 C. Donating Values: Filling in the Missing Counterfactuals ...........................................562 D. Randomization: Balance in Background Variables ......................................................563
    [Show full text]
  • Causation and Causal Inference in Epidemiology
    PUBLIC HEALTH MAnERS Causation and Causal Inference in Epidemiology Kenneth J. Rothman. DrPH. Sander Greenland. MA, MS, DrPH. C Stat fixed. In other words, a cause of a disease Concepts of cause and causal inference are largely self-taught from early learn- ing experiences. A model of causation that describes causes in terms of suffi- event is an event, condition, or characteristic cient causes and their component causes illuminates important principles such that preceded the disease event and without as multicausality, the dependence of the strength of component causes on the which the disease event either would not prevalence of complementary component causes, and interaction between com- have occun-ed at all or would not have oc- ponent causes. curred until some later time. Under this defi- Philosophers agree that causal propositions cannot be proved, and find flaws or nition it may be that no specific event, condi- practical limitations in all philosophies of causal inference. Hence, the role of logic, tion, or characteristic is sufiicient by itself to belief, and observation in evaluating causal propositions is not settled. Causal produce disease. This is not a definition, then, inference in epidemiology is better viewed as an exercise in measurement of an of a complete causal mechanism, but only a effect rather than as a criterion-guided process for deciding whether an effect is pres- component of it. A "sufficient cause," which ent or not. (4m JPub//cHea/f/i. 2005;95:S144-S150. doi:10.2105/AJPH.2004.059204) means a complete causal mechanism, can be defined as a set of minimal conditions and What do we mean by causation? Even among eral.
    [Show full text]
  • Experiments & Observational Studies: Causal Inference in Statistics
    Experiments & Observational Studies: Causal Inference in Statistics Paul R. Rosenbaum Department of Statistics University of Pennsylvania Philadelphia, PA 19104-6340 1 My Concept of a ‘Tutorial’ In the computer era, we often receive compressed • files, .zip. Minimize redundancy, minimize storage, at the expense of intelligibility. Sometimes scientific journals seemed to have been • compressed. Tutorial goal is: ‘uncompress’. Make it possible to • read a current article or use current software without going back to dozens of earlier articles. 2 A Causal Question At age 45, Ms. Smith is diagnosed with stage II • breast cancer. Her oncologist discusses with her two possible treat- • ments: (i) lumpectomy alone, or (ii) lumpectomy plus irradiation. They decide on (ii). Ten years later, Ms. Smith is alive and the tumor • has not recurred. Her surgeon, Steve, and her radiologist, Rachael de- • bate. Rachael says: “The irradiation prevented the recur- • rence – without it, the tumor would have recurred.” Steve says: “You can’t know that. It’s a fantasy – • you’re making it up. We’ll never know.” 3 Many Causal Questions Steve and Rachael have this debate all the time. • About Ms. Jones, who had lumpectomy alone. About Ms. Davis, whose tumor recurred after a year. Whenever a patient treated with irradiation remains • disease free, Rachael says: “It was the irradiation.” Steve says: “You can’t know that. It’s a fantasy. We’ll never know.” Rachael says: “Let’s keep score, add ’em up.” Steve • says: “You don’t know what would have happened to Ms. Smith, or Ms. Jones, or Ms Davis – you just made it all up, it’s all fantasy.
    [Show full text]
  • Elements of Causal Inference
    Elements of Causal Inference Foundations and Learning Algorithms Adaptive Computation and Machine Learning Francis Bach, Editor Christopher Bishop, David Heckerman, Michael Jordan, and Michael Kearns, As- sociate Editors A complete list of books published in The Adaptive Computation and Machine Learning series appears at the back of this book. Elements of Causal Inference Foundations and Learning Algorithms Jonas Peters, Dominik Janzing, and Bernhard Scholkopf¨ The MIT Press Cambridge, Massachusetts London, England c 2017 Massachusetts Institute of Technology This work is licensed to the public under a Creative Commons Attribution- Non- Commercial-NoDerivatives 4.0 license (international): http://creativecommons.org/licenses/by-nc-nd/4.0/ All rights reserved except as licensed pursuant to the Creative Commons license identified above. Any reproduction or other use not licensed as above, by any electronic or mechanical means (including but not limited to photocopying, public distribution, online display, and digital information storage and retrieval) requires permission in writing from the publisher. This book was set in LaTeX by the authors. Printed and bound in the United States of America. Library of Congress Cataloging-in-Publication Data Names: Peters, Jonas. j Janzing, Dominik. j Scholkopf,¨ Bernhard. Title: Elements of causal inference : foundations and learning algorithms / Jonas Peters, Dominik Janzing, and Bernhard Scholkopf.¨ Description: Cambridge, MA : MIT Press, 2017. j Series: Adaptive computation and machine learning series
    [Show full text]
  • A Logic for Causal Inference in Time Series with Discrete and Continuous Variables
    A Logic for Causal Inference in Time Series with Discrete and Continuous Variables Samantha Kleinberg Columbia University New York, NY [email protected] Abstract windows of time. Similarly, Granger causality [1969] allows for only discrete lags between cause and effect and assumes Many applications of causal inference, such as find- that the relationships are between individual variables. Fi- ing the relationship between stock prices and news nally, prior work on inferring complex causal relationships reports, involve both discrete and continuous vari- represented as logical formulas identified factors that substan- ables observed over time. Inference with these tially impact the probability of the effect, but required that all complex sets of temporal data, though, has re- variables be discretized [Kleinberg and Mishra, 2009]. mained difficult and required a number of simplifi- In this work we address the problem of inference when we cations. We show that recent approaches for infer- are interested primarily in the level of the effect. We will ring temporal relationships (represented as logical show that instead of a difference in probability, a cause’s sig- formulas) can be adapted for inference with contin- nificance for an effect can be assessed using an average dif- uous valued effects. Building on advances in logic, ference in conditional expectation. By extending the under- PCTLc (an extension of PCTL with numerical con- lying logic used to represent the relationships, this approach straints) is introduced here to allow representation allows for structured representation and automated inference and inference of relationships with a mixture of of complex temporal relationships with discrete and contin- discrete and continuous components.
    [Show full text]
  • Week 10: Causality with Measured Confounding
    Week 10: Causality with Measured Confounding Brandon Stewart1 Princeton November 28 and 30, 2016 1These slides are heavily influenced by Matt Blackwell, Jens Hainmueller, Erin Hartman, Kosuke Imai and Gary King. Stewart (Princeton) Week 10: Measured Confounding November 28 and 30, 2016 1 / 176 Where We've Been and Where We're Going... Last Week I regression diagnostics This Week I Monday: F experimental Ideal F identification with measured confounding I Wednesday: F regression estimation Next Week I identification with unmeasured confounding I instrumental variables Long Run I causality with measured confounding ! unmeasured confounding ! repeated data Questions? Stewart (Princeton) Week 10: Measured Confounding November 28 and 30, 2016 2 / 176 1 The Experimental Ideal 2 Assumption of No Unmeasured Confounding 3 Fun With Censorship 4 Regression Estimators 5 Agnostic Regression 6 Regression and Causality 7 Regression Under Heterogeneous Effects 8 Fun with Visualization, Replication and the NYT 9 Appendix Subclassification Identification under Random Assignment Estimation Under Random Assignment Blocking Stewart (Princeton) Week 10: Measured Confounding November 28 and 30, 2016 3 / 176 1 The Experimental Ideal 2 Assumption of No Unmeasured Confounding 3 Fun With Censorship 4 Regression Estimators 5 Agnostic Regression 6 Regression and Causality 7 Regression Under Heterogeneous Effects 8 Fun with Visualization, Replication and the NYT 9 Appendix Subclassification Identification under Random Assignment Estimation Under Random Assignment Blocking Stewart
    [Show full text]
  • Editor's Comments
    EDITOR’S COMMENTS Opportunities and Challenges for Different Types of Online Experiments By: Elena Karahanna University of Georgia [email protected] Izak Benbasat University of British Columbia [email protected] Ravi Bapna University of Minnesota [email protected] Arun Rai Editor-in-Chief, MIS Quarterly Georgia State University [email protected] Motivation for and Objectives of the Editorial Experimental research has been an important research method in the Information Systems (IS) discipline. Recently, we have seen an expansion in the types of experiments conducted beyond traditional laboratory (lab) and field experiments. These new types of experiments, which leverage the online environment, provide new opportunities as well as new challenges for IS researchers. This diversity also creates the need for authors and reviewers to understand the respective strengths and limitations of various types of experi- mental research, and not mechanically apply the lens of their favorite type of experiment. The purpose of this editorial is to highlight the reasons that have propelled new types of experiments, categorize these along a set of dimensions, discuss their strengths and weaknesses, and highlight some new issues that emerge with these new opportunities for research. Our objective is not to be exhaustive in terms of the various types of experiments but to highlight opportunities and challenges that emerge for online variants that are more prominently seen in IS research. We, therefore, constrain our focus to lab, field, and natural experiments and their online variants.1 Changing Landscape of Experiments in IS Research Experiments have been a major research method in IS research since the origins of the field.
    [Show full text]
  • Natural Experiments and at the Same Time Distinguish Them from Randomized Controlled Experiments
    Natural Experiments∗ Roc´ıo Titiunik Professor of Politics Princeton University February 4, 2020 arXiv:2002.00202v1 [stat.ME] 1 Feb 2020 ∗Prepared for “Advances in Experimental Political Science”, edited by James Druckman and Donald Green, to be published by Cambridge University Press. I am grateful to Don Green and Jamie Druckman for their helpful feedback on multiple versions of this chapter, and to Marc Ratkovic and participants at the Advances in Experimental Political Science Conference held at Northwestern University in May 2019 for valuable comments and suggestions. I am also indebted to Alberto Abadie, Matias Cattaneo, Angus Deaton, and Guido Imbens for their insightful comments and criticisms, which not only improved the manuscript but also gave me much to think about for the future. Abstract The term natural experiment is used inconsistently. In one interpretation, it refers to an experiment where a treatment is randomly assigned by someone other than the researcher. In another interpretation, it refers to a study in which there is no controlled random assignment, but treatment is assigned by some external factor in a way that loosely resembles a randomized experiment—often described as an “as if random” as- signment. In yet another interpretation, it refers to any non-randomized study that compares a treatment to a control group, without any specific requirements on how the treatment is assigned. I introduce an alternative definition that seeks to clarify the integral features of natural experiments and at the same time distinguish them from randomized controlled experiments. I define a natural experiment as a research study where the treatment assignment mechanism (i) is neither designed nor implemented by the researcher, (ii) is unknown to the researcher, and (iii) is probabilistic by virtue of depending on an external factor.
    [Show full text]
  • Causal Inference in Statistics: an Overview∗†‡
    Statistics Surveys Vol. 3 (2009) 96–146 ISSN: 1935-7516 DOI: 10.1214/09-SS057 Causal inference in statistics: An overview∗†‡ Judea Pearl Computer Science Department University of California, Los Angeles, CA 90095 USA e-mail: [email protected] Abstract: This review presents empirical researcherswith recent advances in causal inference, and stresses the paradigmatic shifts that must be un- dertaken in moving from traditional statistical analysis to causal analysis of multivariate data. Special emphasis is placed on the assumptions that un- derly all causal inferences, the languages used in formulating those assump- tions, the conditional nature of all causal and counterfactual claims, and the methods that have been developed for the assessment of such claims. These advances are illustrated using a general theory of causation based on the Structural Causal Model (SCM) described in Pearl (2000a), which subsumes and unifies other approaches to causation, and provides a coher- ent mathematical foundation for the analysis of causes and counterfactuals. In particular, the paper surveys the development of mathematical tools for inferring (from a combination of data and assumptions) answers to three types of causal queries: (1) queries about the effects of potential interven- tions, (also called “causal effects” or “policy evaluation”) (2) queries about probabilities of counterfactuals, (including assessment of “regret,” “attri- bution” or “causes of effects”) and (3) queries about direct and indirect effects (also known as “mediation”). Finally, the paper defines the formal and conceptual relationships between the structural and potential-outcome frameworks and presents tools for a symbiotic analysis that uses the strong features of both. Keywords and phrases: Structuralequation models, confounding,graph- ical methods, counterfactuals, causal effects, potential-outcome, mediation, policy evaluation, causes of effects.
    [Show full text]
  • Multiple Causal Inference with Latent Confounding
    Multiple Causal Inference with Latent Confounding Rajesh Ranganath 1 Adler Perotte 2 Abstract from observational data. Causal inference from obser- Causal inference from observational data re- vational data requires assumptions. These assumptions quires assumptions. These assumptions range include measurement of all confounders (Rosenbaum and from measuring confounders to identifying in- Rubin, 1983), the presence of external randomness that struments. Traditionally, causal inference as- partially controls treatment (Angrist et al., 1996), and sumptions have focused on estimation of ef- structural assumptions on the randomness (Hoyer et al., fects for a single treatment. In this work, we 2009). construct techniques for estimation with mul- tiple treatments in the presence of unobserved Though causal inference from observational data has been confounding. We develop two assumptions used in many domains, the assumptions that underlie based on shared confounding between treat- these inferences focus on estimation of a causal effect ments and independence of treatments given with a single treatment. But many real-world applications the confounder. Together, these assumptions consist of multiple treatments. For example, the effects lead to a confounder estimator regularized by of genetic variation on various phenotypes (Consortium mutual information. For this estimator, we de- et al., 2007) or the effects of medications from order sets velop a tractable lower bound. To recover treat- in clinical medicine (O’connor et al., 2009) consist of ment effects, we use the residual information in causal problems with multiple treatments rather than a the treatments independent of the confounder. single treatment. Considering multiple treatments make We validate on simulations and an example from new kinds of assumptions possible.
    [Show full text]