Causal Inference In Observational Studies Research Articles

A doubly robust estimator for the Mann Whitney Wilcoxon rank sum test when applied for causal inference in observational studies

ABSTRACT The Mann–Whitney–Wilcoxon rank sum test (MWWRST) is a widely used method for comparing two treatment groups in randomized control trials, particularly when dealing with highly skewed data. However, when applied to observational study data, the MWWRST often yields invalid results for causal inference. To address this limitation, Wu et al. (Causal inference for Mann–Whitney–Wilcoxon rank sum and other nonparametric statistics, Stat. Med. 33 (2014), pp. 1261–1271) introduced an approach that incorporates inverse probability weighting (IPW) into this rank-based statistic to mitigate confounding effects. Subsequently, Mao (On causal estimation using U-statistics, Biometrika 105 (2018), pp. 215–220), Zhang et al. (Estimating Mann Whitney-type causal effects, J. Causal Inference 7 (2019), ARTICLE ID 20180010), and Ai et al. (A Mann–Whitney test of distributional effects in a multivalued treatment, J. Stat. Plan. Inference 209 (2020), pp. 85–100) extended this IPW estimator to develop doubly robust estimators. Nevertheless, each of these approaches has notable limitations. Mao's method imposes stringent assumptions that may not align with real-world study data. Zhang et al.'s (Estimating Mann Whitney-type causal effects, J. Causal Inference 7 (2019), ARTICLE ID 20180010) estimators rely on bootstrap inference, which suffers from computational inefficiency and lacks known asymptotic properties. Meanwhile, Ai et al. (A Mann–Whitney test of distributional effects in a multivalued treatment, J. Stat. Plan. Inference 209 (2020), pp. 85–100) primarily focus on testing the null hypothesis of equal distributions between two groups, which is a more stringent assumption that may not be well-suited to the primary practical application of MWWRST. In this paper, we aim to address these limitations by leveraging functional response models (FRM) to develop doubly robust estimators. We demonstrate the performance of our proposed approach using both simulated and real study data.

The application of target trials with longitudinal targeted maximum likelihood estimation to assess the effect of alcohol consumption in adolescence on depressive symptoms in adulthood

Abstract Time-varying confounding is a common challenge for causal inference in observational studies with time-varying treatments, long follow-up periods, and participant dropout. Confounder adjustment using traditional approaches can be limited by data sparsity, weight instability, and computational issues. The Nicotine Dependence in Teens Study is a prospective cohort study, and we used data from 21 data collection cycles carried out from 1999 to 2008 among 1294 students recruited from 10 high schools in Montreal, Quebec, Canada, including follow-up into adulthood. Our aim in this study was to estimate associations of timing of alcohol initiation and cumulative duration of alcohol use with depression symptoms in adulthood. Based on the target trials framework, we defined intention-to-treat and as-treated parameters in a marginal structural model with sex as a potential effect-modifier. We then used the observational data to emulate the trials. For estimation, we used pooled longitudinal target maximum likelihood estimation, a plug-in estimator with double-robust and local efficiency properties. We describe strategies for dealing with high-dimensional potential drinking patterns and practical positivity violations due to a long follow-up time, including modifying the effect of interest by removing sparsely observed drinking patterns from the loss function and applying longitudinal modified treatment policies to represent the effect of discouraging drinking.

Causal effects of later-eating rhythm on adiposity in children through the comparison of two cohorts in the UK and China: a cross-cohort study

BackgroundLater-eating rhythm (LER) refers to a later timing, greater energy intake, and higher meal frequency in the evening. The role of childhood LER in obesity development is emerging, but most evidence is cross-sectional. Cross-context comparison allows the improvement of causal inference in observational studies by comparing cohorts with different confounding structures. This method is applied to assess the causal effects of LER on adiposity, by exploring the likelihood of residual confounding due to socioeconomic status. MethodsIn this cross-cohort analysis, we used ongoing birth cohort data from the UK Avon Longitudinal Study of Parents and Children (ALSPAC) established in 1991, and the nationally representative China Health and Nutrition Survey (CHNS) collected in 1989–2011. Children with available data at age 7 years were eligible. We applied indices of inequality for assessing confounding structure by regressing LER/adiposity on the standardised score of socioeconomic status (SES) in each cohort. We used multivariable linear and binary logistic regressions to model cross-sectional and prospective associations between LER at 7 years of age and body-mass index (BMI) at ages 7 and 9 years in both cohorts. Analyses were adjusted by age, sex, ethnicity, residency, and socioeconomic status. We used a p value for the Cochrane Q-test obtained from meta-analysis to test for heterogeneity between cohorts. FindingsWe analysed data from 4019 children (2170 [54·0%] female; 1849 [46.0%] male) in ALSPAC and 1749 (788 [45·1%] female; 961 [54.9%] male) in CHNS. The associations between SES and LER or adiposity differed between ALSPAC and CHNS (SES and energy intake for evening main meal: b=1·81 [95% CI 0·81 to 2·81] vs –3·02 [–4·76 to –1·27]; SES and frequency of evening snacks: odds ratio [OR]=0·51 [95% CI 0·41–0·63] vs 5·71 [3·54–9·22]; SES and BMI: b=–0·42 [–0·65 to –0·18] vs 1·29 [0·75 to 1·84]). Positive associations between frequency of evening snacks and BMI were seen in both cross-sectional and longitudinal analyses in both cohorts (mean change of BMI with 1 day increase of consuming evening snacks b=0·09 [0·02 to 0·15]; 0·13 [0·03 to 0·22] kg/m2 per day in ALSPAC, and b=0·11 [–0·07 to 0·28]; 0·30 [0·07 to 0·52] kg/m2 per day in CHNS). No associations were found for energy intake. p values for heterogeneity ranged from 0·107 to 0·932. InterpretationBoth cohorts showed consistent results despite varied dietary cultures and SES patterning of LER or adiposity. Energy intake in the evening or night was not associated with adiposity, whereas evening snacking was. More recent, high-quality cohorts are warranted to enhance the strength of the conclusions. FundingNone.

The directed acyclic graph helped identify confounders in the association between coronary heart disease and pesticide exposure among greenhouse vegetable farmers.

To explore the causal pathways associated with coronary heart disease (CHD) and pesticide exposure using a directed acyclic graph (DAG) analysis and to investigate the potential benefits of DAG by comparing it with logistic regression. This cross-sectional study enrolled 1368 participants from April 2015 to May 2017. Trained research investigators interviewed farmers using a self-administered questionnaire. Logistic regression and DAG models were used to identify the associations between CHD and chronic pesticide exposure. A total of 150 (11.0%) of the 1368 participants are characterized as having CHD. High pesticide exposure (odds ratio = 2.852, 95% confidence intervals: 1.951-4.171) is associated with CHD when compare with low pesticide exposure by both DAG and logistic analyses. After adjusting for the additional potential influence of factors identified by the DAG analysis, there is no significant association, such as the results in logistic regression: ethnicity, education level, settlement time, and mixed pesticide status. Specifically, age, meal frequency, and consumption of fresh fruit, according to the DAG analysis, are independent factors for CHD. High pesticide exposure is a risk factor for CHD as indicated by both DAG and logistic regression analyses. DAG can be a preferable improvement over traditional regression methods to identify sources of bias and causal inference in observational studies, especially for complex research questions.

Regression-assisted Bayesian record linkage for causal inference in observational studies with covariates spread over two files

We consider causal inference for observational studies with data spread over two files. One file includes the treatment, outcome, and some covariates measured on a set of individuals, and the other file includes additional causally-relevant covariates measured on a partially overlapping set of individuals. By linking records in the two databases, the analyst can control for more covariates, thereby reducing the risk of bias compared to using only one file alone. When analysts do not have access to a unique identifier that enables perfect, error-free linkages, they typically rely on probabilistic record linkage to construct a single linked data set, and estimate causal effects using these linked data. This typical practice does not propagate uncertainty from imperfect linkages to the causal inferences. Further, it does not take advantage of relationships among the variables to improve the linkage quality. We address these shortcomings by fusing regression-assisted, Bayesian probabilistic record linkage with causal inference. The Markov chain Monte Carlo sampler generates multiple plausible linked data files as byproducts that analysts can use for multiple imputation inferences. Here, we show results for two causal estimators based on propensity score overlap weights. Using simulations and data from the Italy Survey on Household Income and Wealth, we show that our approach can improve the accuracy of estimated treatment effects.

Reconciling Algorithmic Fairness Criteria

Reconciling Algorithmic Fairness Criteria

Complex discontinuity designs using covariates: Impact of school grade retention on later life outcomes in Chile

Regression discontinuity designs are extensively used for causal inference in observational studies. However, they are usually confined to settings with simple treatment rules and determined by a single running variable with a single cutoff. Motivated by the problem of estimating the impact of grade retention on educational and juvenile crime outcomes in Chile, we propose a framework and methods for complex discontinuity designs that encompass multiple treatment rules. In this framework the observed covariates play a central role for identification, estimation, and generalization of causal effects. Identification is nonparametric and relies on a local strong ignorability assumption. Estimation proceeds, as in any observational study, under strong ignorability, yet in a neighborhood of the cutoffs of the running variables. We discuss estimation approaches based on matching and weighting, including complementary regression modeling adjustments. We present assumptions for generalization, that is, for identification and estimation of average treatment effects for target populations. We also describe two approaches to select the neighborhood for analysis. We find that grade retention in Chile has a negative impact on future grade retention but is not associated with dropping out of school or committing a juvenile crime.

Large Sample Properties of Matching for Balance

Matching methods are widely used for causal inference in observational studies. Among them, nearest neighbor matching is arguably the most popular. However, nearest neighbor matching does not generally yield an average treatment effect estimator that is $\sqrt{n}$-consistent (Abadie and Imbens, 2006). Are matching methods not $\sqrt{n}$-consistent in general? In this paper, we study a recent class of matching methods that use integer programming to directly target aggregate covariate balance as opposed to finding close neighbor matches. We show that under suitable conditions these methods can yield simple estimators that are $\sqrt{n}$-consistent and asymptotically optimal.

Lmtp: An R Package for Estimating the Causal Effects of Modified Treatment Policies

We present the lmtp R package for causal inference from longitudinal observational or randomized studies. This package implements the estimators of Díaz et al. (2021) for estimating general non-parametric causal effects based on modified treatment policies. Modified treatment policies generalize static and dynamic interventions, making lmtp and all-purpose package for non-parametric causal inference in observational studies. The methods provided can be applied to both point-treatment and longitudinal settings, and can account for time-varying exposure, covariates, and right censoring thereby providing a very general tool for causal inference. Additionally, two of the provided estimators are based on flexible machine learning regression algorithms, and avoid bias due to parametric model misspecification while maintaining valid statistical inference.

Improving the Validity of Causal Inferences in Observational Studies.

Editor's note: This is the 13th article in a series on clinical research by nurses. The series is designed to be used as a resource for nurses to understand the concepts and principles essential to research. Each column will present the concepts that underpin evidence-based practice-from research design to data interpretation. To see all the articles in the series, go to https://links.lww.com/AJN/A204.

Combining parametric and nonparametric models to estimate treatment effects in observational studies.

Performing causal inference in observational studies requires we assume confounding variables are correctly adjusted for. In settings with few discrete-valued confounders, standard models can be employed. However, as the number of confounders increases these models become less feasible as there are fewer observations available for each unique combination of confounding variables. In this paper, we propose a new model for estimating treatment effects in observational studies that incorporates both parametric and nonparametric outcome models. By conceptually splitting the data, we can combine these models while maintaining a conjugate framework, allowing us to avoid the use of Markov chain Monte Carlo (MCMC) methods. Approximations using the central limit theorem and random sampling allow our method to be scaled to high-dimensional confounders. Through simulation studies we show our method can be competitive with benchmark models while maintaining efficient computation, and illustrate the method on a large epidemiological health survey.

Overview on the generalized propensity scoring estimator for continuous treatment

Among kinds of methods for causal inference in observational studies, the propensity score (PS) method to control measured confounding is more widely used. PS method usually consists of two critical steps: first, estimating the propensity score, followed by calculating the causal parameters of interest by regression, weighting, matching, and stratification. Unlike the traditional dichotomous treatment, the generalized propensity scoring estimator used for continuous treatment has been proposed in recent years. Many methods have been developed to estimate the generalized propensity score or even estimate the balancing weight directly. This paper introduces the existing estimators from both the model-based and balance-based perspectives.

The impact of unmeasured confounding on causal inference in observational studies: A plasmode simulation study of targeted maximum likelihood estimation

The impact of unmeasured confounding on causal inference in observational studies: A plasmode simulation study of targeted maximum likelihood estimation

Application of Standardization for Causal Inference in Observational Studies: A Step-by-step Tutorial for Analysis Using R Software.

Epidemiological studies typically examine the causal effect of exposure on a health outcome. Standardization is one of the most straightforward methods for estimating causal estimands. However, compared to inverse probability weighting, there is a lack of user-centric explanations for implementing standardization to estimate causal estimands. This paper explains the standardization method using basic R functions only and how it is linked to the R package stdReg, which can be used to implement the same procedure. We provide a step-by-step tutorial for estimating causal risk differences, causal risk ratios, and causal odds ratios based on standardization. We also discuss how to carry out subgroup analysis in detail.

Air Pollution and Autism Spectrum Disorder in Israel: A Negative Control Analysis.

Residual confounding is a major concern for causal inference in observational studies on air pollution-autism spectrum disorder (ASD) associations. This study is aimed at assessing confounding in these associations using negative control exposures. This nested case-control study included all children diagnosed with ASD (detected through 31 December 2016) born during 2007-2012 in Israel and residing in the study area (N = 3,843), and matched controls of the same age (N = 38,430). We assigned individual house-level exposure estimates for each child. We estimated associations using logistic regression models, mutually adjusted for all relevant exposure periods (prepregnancy, pregnancy, and postnatal). We assessed residual confounding using postoutcome negative control exposure at age 28-36 months. In mutually adjusted models, we observed positive associations with ASD for postnatal exposures to NOx (odds ratio per interquartile range, 95% confidence interval: 1.19, 1.02-1.38) and NO2 (1.20, 1.00-1.43), and gestational exposure to PM2.5-10 (1.08, 1.01-1.15). The result for the negative control period was 1.04, 0.99-1.10 for PM2.5, suggesting some residual confounding, but no associations for PM2.5-10 (0.98, 0.81-1.18), NOx (1.02, 0.84-1.25), or NO2 (0.98, 0.81-1.18), suggesting no residual confounding. Our results further support a hypothesized causal link with ASD that is specific to postnatal exposures to traffic-related pollution.

Commentary: Using potential outcomes causal methods to assess whether reductions in PM2.5 result in decreased mortality

Causal inference regarding exposures to ambient fine particulate matter (PM2.5) and mortality estimated from observational studies is limited by confounding, among other factors. In light of a variety of causal inference frameworks and methods that have been developed over the past century to specifically quantify causal effects, three research teams were selected in 2016 to evaluate the causality of PM2.5-mortality association among Medicare beneficiaries, using their own selections of causal inference methods and study designs but the same data sources. With a particular focus on controlling for unmeasured confounding, two research teams adopted an instrumental variables approach under a quasi-experiment or natural experiment study design, whereas one team adopted a structural nested mean model under the traditional cohort study design. All three research teams reported results supporting an estimated counterfactual causal relationship between ambient PM2.5 and all-cause mortality, and their estimated causal relationships are largely of similar magnitudes to recent epidemiological studies based on regression analyses with omitted potential confounders. The causal methods used by all three research teams were built upon the potential outcomes framework. This framework has marked conceptual advantages over regression-based methods in addressing confounding and yielding unbiased estimates of average treatment effect in observational epidemiological studies. However, potential violations of the unverifiable assumptions underlying each causal method leave the results from all three studies subject to biases. We also note that the studies are not immune to some other common sources of bias, including exposure measurement errors, ecological study design, model uncertainty and specification errors, and irrelevant exposure windows, that can undermine the validity of causal inferences in observational studies. As a result, despite some apparent consistency of study results from the three research teams with the wider epidemiological literature on PM2.5-mortality statistical associations, caution seems warranted in drawing causal conclusions from the results. A possible way forward is to improve study design and reduce dependence of conclusions on untested assumptions by complementing potential outcomes methods with structural causal modeling and information-theoretic methods that emphasize empirically tested and validated relationships.

Identifiability of causal effects with multiple causes and a binary outcome.

Unobserved confounding presents a major threat to causal inference in observational studies. Recently, several authors have suggested that this problem could be overcome in a shared confounding setting where multiple treatments are independent given a common latent confounder. It has been shown that under a linear Gaussian model for the treatments,the causal effect is not identifiable without parametric assumptions on the outcome model. In this note, we show that the causal effect is indeed identifiable if we assume a general binary choice model for the outcome with a non-probit link. Our identification approach is based on the incongruence between Gaussianity of the treatments and latent confounder and non-Gaussianity of a latent outcome variable. We further develop a two-step likelihood-based estimation procedure.

Mendelian randomization as a tool for causal inference in human nutrition and metabolism.

The current review describes the fundamentals of the Mendelian randomization framework and its current application for causal inference in human nutrition and metabolism. In the Mendelian randomization framework, genetic variants that are strongly associated with the potential risk factor are used as instrumental variables to determine whether the risk factor is a cause of the disease. Mendelian randomization studies are less susceptible to confounding and reverse causality compared with traditional observational studies. The Mendelian randomization study design has been increasingly used in recent years to appraise the causal associations of various nutritional factors, such as milk and alcohol intake, circulating levels of micronutrients and metabolites, and obesity with risk of different health outcomes. Mendelian randomization studies have confirmed some but challenged other nutrition-disease associations recognized by traditional observational studies. Yet, the causal role of many nutritional factors and intermediate metabolic changes for health and disease remains unresolved. Mendelian randomization can be used as a tool to improve causal inference in observational studies assessing the role of nutritional factors and metabolites in health and disease. There is a need for more large-scale genome-wide association studies to identify more genetic variants for nutritional factors that can be utilized for Mendelian randomization analyses.

Propensity score-based methods for causal inference in observational studies with non-binary treatments.

Propensity score methods are a part of the standard toolkit for applied researchers who wish to ascertain causal effects from observational data. While they were originally developed for binary treatments, several researchers have proposed generalizations of the propensity score methodology for non-binary treatment regimes. Such extensions have widened the applicability of propensity score methods and are indeed becoming increasingly popular themselves. In this article, we closely examine two methods that generalize propensity scores in this direction, namely, the propensity function (PF), and the generalized propensity score (GPS), along with two extensions of the GPS that aim to improve its robustness. We compare the assumptions, theoretical properties, and empirical performance of these methods. On a theoretical level, the GPS and its extensions are advantageous in that they are designed to estimate the full dose response function rather than the average treatment effect that is estimated with the PF. We compare GPS with a new PF method, both of which estimate the dose response function. We illustrate our findings and proposals through simulation studies, including one based on an empirical study about the effect of smoking on healthcare costs. While our proposed PF-based estimator preforms well, we generally advise caution in that all available methods can be biased by model misspecification and extrapolation.

Methods for Evaluating Causality in Observational Studies.

In clinical medical research, causality is demonstrated b controlled trials (RCTs). Often, however, an RCT cannot be conducted for ethical reasons, and sometimes for practical reasons as well. In such cases, knowledge can be derived from an observational study instead. In this article, we present two methods that have not been widely used in medical research to date. The methods of assessing causal inferences in observational studies are described on the basis of publications retrieved by a selective literature search. Two relatively new approaches-regression-discontinuity methods and interrupted time series-can be used to demonstrate a causal relationship under certain circumstances. The regression-discontinuity design is a quasi-experimental approach that can be applied if a continuous assignment variable is used with a threshold value. Patients are assigned to different treatment schemes on the basis of the threshold value. For assignment variables that are subject to random measurement error, it is assumed that, in a small interval around a threshold value, e.g., cholesterol values of 160 mg/dL, subjects are assigned essentially at random to one of two treatment groups. If patients with a value above the threshold are given a certain treatment, those with values below the threshold can serve as control group. Interrupted time series are a special type of regression-discontinuity design in which time is the assignment variable, and the threshold is a cutoff point. This is often an external event, such as the imposition of a smoking ban. A before-and-after comparison can be used to determine the effect of the intervention (e.g., the smoking ban) on health parameters such as the frequency of cardiovascular disease. The approaches described here can be used to derive causal inferences ies. They should only be applied after the prerequisites for their use have been carefully checked.