Fast Causal Inference Research Articles

Utilizing Causal Pathway Analysis To Predict Change In Cardiorespiratory Fitness In The STRRIDE Randomized Trials

Exploring high-throughput data in a causal pathway framework facilitates accurate and parsimonious multivariable predictive models for clinical outcomes. Developing these models elucidates mechanisms by which regular exercise elicits health benefits. PURPOSE: As a proof-of-concept, we tested the utility of causal pathway discovery for predicting changes in V̇O2peak across the STRRIDE trials. METHODS: A total of 532 adults from three STRRIDE studies were randomized to one of 7 exercise interventions, ranging from doses of 8-22 kcal/kg/week; intensities of 50-75% V̇O2peak; and durations of 6-9 months. Six groups included aerobic exercise, two included resistance training, and one included dietary intervention. Graded maximal exercise treadmill testing with expired gas analysis determined Absolute V̇O2peak. The Fast Causal Inference algorithm was applied to discover the mechanistic relationship among 231 clinical and transcriptomic variables with 200 bootstrap resamples to assess the stability of the discovered causal graph. To estimate effect sizes, V̇O2peak was regressed on variables in its local causal neighborhood. RESULTS: Forty-four variables were identified in the causal vicinity (within three edges away) of V̇O2peak following exercise intervention in more than 50% of bootstrap resampling runs. Exercise intensity, age, and pre-training V̇O2peak were identified as the direct causes of post-training V̇O2peak. Among the studied variables, none mediated the effect of exercise dose. The following exercise combinations were the most effective in changing V̇O2peak: high amount/vigorous intensity aerobic training (β=0.39 L/min); low amount/vigorous intensity aerobic plus resistance training (β=0.33 L/min); low amount/vigorous intensity aerobic training (β=0.27 L/min); high amount/moderate intensity aerobic training (β=0.26 L/min); low amount/moderate intensity aerobic training (β=0.19 L/min); and resistance training (β=0.18 L/min). CONCLUSIONS: Multivariable causal graph-based inference confirmed an exercise dose-response for V̇O2peak. As we develop and incorporate additional molecular data, our future research will use this approach to maximize predictive ability for other clinical phenotypes to help make personalized lifestyle medicine a functional reality.

T221. CAUSAL RELATIONSHIPS BETWEEN EMPATHY, MOTIVATION AND FUNCTIONAL OUTCOMES IN EARLY SCHIZOPHRENIA IDENTIFIED WITH CAUSAL DISCOVERY MODELING

BackgroundSchizophrenia is characterized by devastating impairments in occupational and social functioning. These impairments result in poor quality of life, have significant societal burden, and are resistant to current treatments. Although a number of predictors of poor functional outcomes have been identified, the causal relationships between malleable illness variables and functional status must be discovered to establish the most promising candidate treatment targets. The aim of this research is to employ a data-driven causal network approach in the Recovery After an Initial Schizophrenia Episode (RAISE) trial to identify variables that have causal effects on occupational and social functioning. Our second aim was to validate our results in an independent sample drawn from the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) study.MethodsMeasures of neurocognition, motivation, empathy, psychiatric symptoms, duration of untreated psychosis, and social and occupational functioning were obtained for 279 participants with early schizophrenia enrolled in RAISE. Data were analyzed using a machine learning causal discovery algorithm, Greedy Fast Causal Inference, which infers causal relationships among the variables while identifying potential influences of latent variables. Causal effect sizes (ES) were estimated by fitting a linear structural equation model to the causal graph. Graph stability was evaluated by running GFCI on 1000 jackknifed and 1000 bootstrapped datasets. Results were validated using a similar set of variables in 187 participants in the CATIE study who were within five years of initiating antipsychotic treatment.ResultsBaseline empathy caused baseline motivation (ES = 0.77), which in turn caused both baseline social (ES = 1.5) and occupational (ES = 0.96) functioning. Baseline social and occupational functioning then caused six-month social and occupational functioning, respectively. Additionally, at the six-month time point, social functioning caused motivation (ES = 0.21), which in turn caused occupational functioning (ES = .92). Causal relationships between motivation and social functioning were cyclical through time. The causal structure in the CATIE dataset had some less determinate edge orientations, but otherwise confirmed the findings from RAISE. The jackknife and bootstrap causal graphs showed high concordance with all graph features in the social and occupational functioning subgraphs.DiscussionMotivation and empathy at baseline have causal effects on both occupational and social functioning six months after entering treatment in early schizophrenia. In addition, an improvement in social functioning over six-months may drive improvements in occupational functioning. This indicates that future research should test the efficacy of interventions targeting motivation and empathy for promoting functional recovery in early schizophrenia as soon as individuals enter treatment.

Learning genetic and environmental graphical models from family data.

Many challenging problems in biomedical research rely on understanding how variables are associated with each other and influenced by genetic and environmental factors. Probabilistic graphical models (PGMs) are widely acknowledged as a very natural and formal language to describe relationships among variables and have been extensively used for studying complex diseases and traits. In this work, we propose methods that leverage observational Gaussian family data for learning a decomposition of undirected and directed acyclic PGMs according to the influence of genetic and environmental factors. Many structure learning algorithms are strongly based on a conditional independence test. For independent measurements of normally distributed variables, conditional independence can be tested through standard tests for zero partial correlation. In family data, the assumption of independent measurements does not hold since related individuals are correlated due to mainly genetic factors. Based on univariate polygenic linear mixed models, we propose tests that account for the familial dependence structure and allow us to assess the significance of the partial correlation due to genetic (between-family) factors and due to other factors, denoted here as environmental (within-family) factors, separately. Then, we extend standard structure learning algorithms, including the IC/PC and the really fast causal inference (RFCI) algorithms, to Gaussian family data. The algorithms learn the most likely PGM and its decomposition into two components, one explained by genetic factors and the other by environmental factors. The proposed methods are evaluated by simulation studies and applied to the Genetic Analysis Workshop 13 simulated dataset, which captures significant features of the Framingham Heart Study.

Challenges and Opportunities with Causal Discovery Algorithms: Application to Alzheimer\u2019s Pathophysiology

Causal Structure Discovery (CSD) is the problem of identifying causal relationships from large quantities of data through computational methods. With the limited ability of traditional association-based computational methods to discover causal relationships, CSD methodologies are gaining popularity. The goal of the study was to systematically examine whether (i) CSD methods can discover the known causal relationships from observational clinical data and (ii) to offer guidance to accurately discover known causal relationships. We used Alzheimer’s disease (AD), a complex progressive disease, as a model because the well-established evidence provides a “gold-standard” causal graph for evaluation. We evaluated two CSD methods, Fast Causal Inference (FCI) and Fast Greedy Equivalence Search (FGES) in their ability to discover this structure from data collected by the Alzheimer’s Disease Neuroimaging Initiative (ADNI). We used structural equation models (which is not designed for CSD) as control. We applied these methods under three scenarios defined by increasing amounts of background knowledge provided to the methods. The methods were evaluated by comparing the resulting causal relationships with the “gold standard” graph that was constructed from literature. Dedicated CSD methods managed to discover graphs that nearly coincided with the gold standard. For best results, CSD algorithms should be used with longitudinal data providing as much prior knowledge as possible.

Assessing the collective utility of multiple analyses on clinical alcohol use disorder data.

The objective of this study was to assess the potential of combining graph learning methods with latent variable estimation methods for mining clinically useful information from observational clinical data sets. The data set contained self-reported measures of psychopathology symptoms from a clinical sample receiving treatment for alcohol use disorder. We used the traditional graph learning methods: Graphical Least Absolute Shrinkage and Selection Operator, and Friedman's hill climbing algorithm; traditional latent variable estimation method factor analysis; recently developed graph learning method Greedy Fast Causal Inference; and recently developed latent variable estimation method Find One Factor Clusters. Methods were assessed qualitatively by the content of their findings. Recently developed graphical methods identified potential latent variables (ie, not represented in the model) influencing particular scores. Recently developed latent effect estimation methods identified plausible cross-score loadings that were not found with factor analysis. A graphical analysis of individual items identified a mistake in wording on 1 questionnaire and provided further evidence that certain scores are not reflective of indirectly measured common causes. Our findings suggest that a combination of Greedy Fast Causal Inference and Find One Factor Clusters can enhance the evidence-based information yield from psychopathological constructs and questionnaires. Traditional methods provided some of the same information but missed other important findings. These conclusions point the way toward more informative interrogations of existing and future data sets than are commonly employed at present.

Causal Network Modeling of the Determinants of Drinking Behavior in Comorbid Alcohol Use and Anxiety Disorder.

Anxiety and depression disorders (internalizing psychopathology) occur in approximately 50% of patients with alcohol use disorder (AUD) and mark a 2-fold increase in the rate of relapse in the months following treatment. In a previous study using network modeling, we found that perceived stress and drinking to cope (DTC) with negative affect were central to maintaining network associations between internalizing psychopathology INTP and drinking in comorbid individuals. Here, we extend this approach to a causal framework. Measures of INTP, drinking urges/behavior, abstinence self-efficacy, and DTC were obtained from 362 adult AUD treatment patients who had a co-occurring anxiety disorder. Data were analyzed using a machine-learning algorithm ("Greedy Fast Causal Inference"[ GFCI]) that infers paths of causal influence while identifying potential influences associated with unmeasured ("latent") variables. DTC with negative affect served as a central hub for 2 distinct causal paths leading to drinking behavior, (i) a direct syndromic pathway originating with social anxiety and (ii) an indirect stress pathway originating with perceived stress. Findings expand the field's knowledge of the paths of influence that lead from internalizing disorder to drinking in AUD as shown by the first application in psychopathology of a powerful network analysis algorithm (GFCI) to model these causal relationships.

A constraint-based algorithm for causal discovery with cycles, latent variables and selection bias

Causal processes in nature may contain cycles, and real datasets may violate causal sufficiency as well as contain selection bias. No constraint-based causal discovery algorithm can currently handle cycles, latent variables and selection bias (CLS) simultaneously. I therefore introduce an algorithm called cyclic causal inference (CCI) that makes sound inferences with a conditional independence oracle under CLS, provided that we can represent the cyclic causal process as a non-recursive linear structural equation model with independent errors. Empirical results show that CCI outperforms the cyclic causal discovery algorithm in the cyclic case as well as rivals the fast causal inference and really fast causal inference algorithms in the acyclic case. An R implementation is available at https://github.com/ericstrobl/CCI .

Comparison of strategies for scalable causal discovery of latent variable models from mixed data

Modern technologies allow large, complex biomedical datasets to be collected from patient cohorts. These datasets are comprised of both continuous and categorical data (“Mixed Data”), and essential variables may be unobserved in this data due to the complex nature of biomedical phenomena. Causal inference algorithms can identify important relationships from biomedical data; however, handling the challenges of causal inference over mixed data with unmeasured confounders in a scalable way is still an open problem. Despite recent advances into causal discovery strategies that could potentially handle these challenges; individually, no study currently exists that comprehensively compares these approaches in this setting. In this paper, we present a comparative study that addresses this problem by comparing the accuracy and efficiency of different strategies in large, mixed datasets with latent confounders. We experiment with two extensions of the Fast Causal Inference algorithm: a maximum probability search procedure we recently developed to identify causal orientations more accurately, and a strategy which quickly eliminates unlikely adjacencies in order to achieve scalability to high-dimensional data. We demonstrate that these methods significantly outperform the state of the art in the field by achieving both accurate edge orientations and tractable running time in simulation experiments on datasets with up to 500 variables. Finally, we demonstrate the usability of the best performing approach on real data by applying it to a biomedical dataset of HIV-infected individuals.

Fast Causal Inference with Non-Random Missingness by Test-Wise Deletion.

Many real datasets contain values missing not at random (MNAR). In this scenario, investigators often perform list-wise deletion, or delete samples with any missing values, before applying causal discovery algorithms. List-wise deletion is a sound and general strategy when paired with algorithms such as FCI and RFCI, but the deletion procedure also eliminates otherwise good samples that contain only a few missing values. In this report, we show that we can more efficiently utilize the observed values with test-wise deletion while still maintaining algorithmic soundness. Here, test-wise deletion refers to the process of list-wise deleting samples only among the variables required for each conditional independence (CI) test used in constraint-based searches. Test-wise deletion therefore often saves more samples than list-wise deletion for each CI test, especially when we have a sparse underlying graph. Our theoretical results show that test-wise deletion is sound under the justifiable assumption that none of the missingness mechanisms causally affect each other in the underlying causal graph. We also find that FCI and RFCI with test-wise deletion outperform their list-wise deletion and imputation counterparts on average when MNAR holds in both synthetic and real data.

An algorithm for direct causal learning of influences on patient outcomes.

This study aims at developing and introducing a new algorithm, called direct causal learner (DCL), for learning the direct causal influences of a single target. We applied it to both simulated and real clinical and genome wide association study (GWAS) datasets and compared its performance to classic causal learning algorithms. The DCL algorithm learns the causes of a single target from passive data using Bayesian-scoring, instead of using independence checks, and a novel deletion algorithm. We generate 14,400 simulated datasets and measure the number of datasets for which DCL correctly and partially predicts the direct causes. We then compare its performance with the constraint-based path consistency (PC) and conservative PC (CPC) algorithms, the Bayesian-score based fast greedy search (FGS) algorithm, and the partial ancestral graphs algorithm fast causal inference (FCI). In addition, we extend our comparison of all five algorithms to both a real GWAS dataset and real breast cancer datasets over various time-points in order to observe how effective they are at predicting the causal influences of Alzheimer's disease and breast cancer survival. DCL consistently outperforms FGS, PC, CPC, and FCI in discovering the parents of the target for the datasets simulated using a simple network. Overall, DCL predicts significantly more datasets correctly (McNemar's test significance: p<<0.0001) than any of the other algorithms for these network types. For example, when assessing overall performance (simple and complex network results combined), DCL correctly predicts approximately 1400 more datasets than the top FGS method, 1600 more datasets than the top CPC method, 4500 more datasets than the top PC method, and 5600 more datasets than the top FCI method. Although FGS did correctly predict more datasets than DCL for the complex networks, and DCL correctly predicted only a few more datasets than CPC for these networks, there is no significant difference in performance between these three algorithms for this network type. However, when we use a more continuous measure of accuracy, we find that all the DCL methods are able to better partially predict more direct causes than FGS and CPC for the complex networks. In addition, DCL consistently had faster runtimes than the other algorithms. In the application to the real datasets, DCL identified rs6784615, located on the NISCH gene, and rs10824310, located on the PRKG1 gene, as direct causes of late onset Alzheimer's disease (LOAD) development. In addition, DCL identified ER category as a direct predictor of breast cancer mortality within 5 years, and HER2 status as a direct predictor of 10-year breast cancer mortality. These predictors have been identified in previous studies to have a direct causal relationship with their respective phenotypes, supporting the predictive power of DCL. When the other algorithms discovered predictors from the real datasets, these predictors were either also found by DCL or could not be supported by previous studies. Our results show that DCL outperforms FGS, PC, CPC, and FCI in almost every case, demonstrating its potential to advance causal learning. Furthermore, our DCL algorithm effectively identifies direct causes in the LOAD and Metabric GWAS datasets, which indicates its potential for clinical applications.

How Mega Is the Mega? Measuring the Spillover Effects of WeChat by Machine Learning and Econometrics

WeChat, an instant messaging app, is considered a mega app due to its dominance in terms of usage among Chinese smartphone users. Nevertheless, little is known about its externality in regard to the broader app market. Our work estimates the spillover effects of WeChat on the other top-50 most frequently used apps in China through data on users’ weekly app usage. Given the challenge of determining causal inference from observational data, we apply a graphical model and econometrics to estimate the spillover effects through two steps: (1) we determine the causal structure by estimating a partially ancestral diagram, using a Fast Causal Inference (FCI) algorithm; (2) given the causal structure, we find a valid adjustment set and estimate the causal effects by an econometric model with the adjustment set as controlling non-causal effects. Our findings show that the spillover effects of WeChat are limited; in fact, only two other apps, Tencent News and Taobao, receive positive spillover effects from WeChat. In addition, we show that, if researchers fail to account for the causal structure that we determined from the graphical model, it is easy to fall into the trap of confounding bias and selection bias when estimating causal effects.

CauseMap: fast inference of causality from complex time series

Background. Establishing health-related causal relationships is a central pursuit in biomedical research. Yet, the interdependent non-linearity of biological systems renders causal dynamics laborious and at times impractical to disentangle. This pursuit is further impeded by the dearth of time series that are sufficiently long to observe and understand recurrent patterns of flux. However, as data generation costs plummet and technologies like wearable devices democratize data collection, we anticipate a coming surge in the availability of biomedically-relevant time series data. Given the life-saving potential of these burgeoning resources, it is critical to invest in the development of open source software tools that are capable of drawing meaningful insight from vast amounts of time series data.Results. Here we present CauseMap, the first open source implementation of convergent cross mapping (CCM), a method for establishing causality from long time series data (≳25 observations). Compared to existing time series methods, CCM has the advantage of being model-free and robust to unmeasured confounding that could otherwise induce spurious associations. CCM builds on Takens’ Theorem, a well-established result from dynamical systems theory that requires only mild assumptions. This theorem allows us to reconstruct high dimensional system dynamics using a time series of only a single variable. These reconstructions can be thought of as shadows of the true causal system. If reconstructed shadows can predict points from opposing time series, we can infer that the corresponding variables are providing views of the same causal system, and so are causally related. Unlike traditional metrics, this test can establish the directionality of causation, even in the presence of feedback loops. Furthermore, since CCM can extract causal relationships from times series of, e.g., a single individual, it may be a valuable tool to personalized medicine. We implement CCM in Julia, a high-performance programming language designed for facile technical computing. Our software package, CauseMap, is platform-independent and freely available as an official Julia package.Conclusions. CauseMap is an efficient implementation of a state-of-the-art algorithm for detecting causality from time series data. We believe this tool will be a valuable resource for biomedical research and personalized medicine.

Learning high-dimensional directed acyclic graphs with latent and selection variables

We consider the problem of learning causal information between random variables in directed acyclic graphs (DAGs) when allowing arbitrarily many latent and selection variables. The FCI (Fast Causal Inference) algorithm has been explicitly designed to infer conditional independence and causal information in such settings. However, FCI is computationally infeasible for large graphs. We therefore propose the new RFCI algorithm, which is much faster than FCI. In some situations the output of RFCI is slightly less informative, in particular with respect to conditional independence information. However, we prove that any causal information in the output of RFCI is correct in the asymptotic limit. We also define a class of graphs on which the outputs of FCI and RFCI are identical. We prove consistency of FCI and RFCI in sparse high-dimensional settings, and demonstrate in simulations that the estimation performances of the algorithms are very similar. All software is implemented in the R-package pcalg.

Constraint-based inference algorithms for structural models with latent confounders— empirical application and simulations

Graphical methods for the discovery of structural models from observational data provide interesting tools for applied researchers. A problem often faced in empirical studies is the presence of latent confounders which produce associations between the observed variables. Although causal inference algorithms exist which can cope with latent confounders, empirical applications assessing the performance of such algorithms are largely lacking. In this study, we apply the constraint based Fast Causal Inference algorithm implemented in the software program TETRAD on a data set containing strategy and performance information about 608 business units. In contrast to the informative and reasonable results for the impirical data, simulation findings reveal problems in recovering some of the structural relations.