Instrumental Variable Assumptions Research Articles

Sjögren’s Disease and Oral Health: A Genetic Instrumental Variable Analysis

Epidemiological studies have consistently shown that Sjögren’s disease (SjD) increases the risk of dental caries. Despite similar evidence indicating an elevated risk of periodontitis, SjD remains a disputed risk factor for this disease. The risk of bias in observational research is a major impediment to confirming this link. Within an instrumental variable framework, genetic variants associated with a risk factor can be used to proxy its effect on an outcome while avoiding common sources of observational study bias. In this study, we leveraged an instrumental variable approach to investigate whether SjD affects the risk of caries and periodontitis. A total of 57 genetic variants strongly associated with SjD were identified from a genome-wide association study of 2,247 European descent cases and 332,115 controls. We tested for associations of these genetic instruments with caries (measured as the number of decayed, missing, and filled surfaces in 26,792 individuals) and periodontitis (17,353 clinical periodontitis cases and 28,210 European controls). Several sensitivity analyses were used to further validate the primary inverse variance weighted (IVW) estimate. IVW analysis revealed an adverse effect of SjD on caries (β = 0.039, P = 6.3e-16) and periodontitis (odds ratio = 1.033, P = 2.3e-05). Sensitivity analyses, conducted to assess the robustness to potential violations of instrumental variable assumptions, further support these findings. Our results showed that SjD has a detrimental effect on caries and also suggest that SjD promotes periodontitis.

Addressing unmeasured confounders in cohort studies: Instrumental variable method for a time-fixed exposure on an outcome trajectory.

Instrumental variable methods, which handle unmeasured confounding by targeting the part of the exposure explained by an exogenous variable not subject to confounding, have gained much interest in observational studies. We consider the very frequent setting of estimating the unconfounded effect of an exposure measured at baseline on the subsequent trajectory of an outcome repeatedly measured over time. We didactically explain how to apply the instrumental variable method in such setting by adapting the two-stage classical methodology with (1) the prediction of the exposure according to the instrumental variable, (2) its inclusion into a mixed model to quantify the exposure association with the subsequent outcome trajectory, and (3) the computation of the estimated total variance. A simulation study illustrates the consequences of unmeasured confounding in classical analyses and the usefulness of the instrumental variable approach. The methodology is then applied to 6224 participants of the 3C cohort to estimate the association of type-2 diabetes with subsequent cognitive trajectory, using 42 genetic polymorphisms as instrumental variables. This contribution shows how to handle endogeneity when interested in repeated outcomes, along with a R implementation. However, it should still be used with caution as it relies on instrumental variable assumptions hardly testable inpractice.

Double Machine Learning for Sample Selection Models

This article considers the evaluation of discretely distributed treatments when outcomes are only observed for a subpopulation due to sample selection or outcome attrition. For identification, we combine a selection-on-observables assumption for treatment assignment with either selection-on-observables or instrumental variable assumptions concerning the outcome attrition/sample selection process. We also consider dynamic confounding, meaning that covariates that jointly affect sample selection and the outcome may (at least partly) be influenced by the treatment. To control in a data-driven way for a potentially high dimensional set of pre- and/or post-treatment covariates, we adapt the double machine learning framework for treatment evaluation to sample selection problems. We make use of (a) Neyman-orthogonal, doubly robust, and efficient score functions, which imply the robustness of treatment effect estimation to moderate regularization biases in the machine learning-based estimation of the outcome, treatment, or sample selection models and (b) sample splitting (or cross-fitting) to prevent overfitting bias. We demonstrate that the proposed estimators are asymptotically normal and root-n consistent and investigate their finite sample properties in a simulation study. We also apply our proposed methodology to the Job Corps data. The estimator is available in the causalweight package for the statistical software R.

Relaxing parametric assumptions for non-linear Mendelian randomization using a doubly-ranked stratification method.

Non-linear Mendelian randomization is an extension to standard Mendelian randomization to explore the shape of the causal relationship between an exposure and outcome using an instrumental variable. A stratification approach to non-linear Mendelian randomization divides the population into strata and calculates separate instrumental variable estimates in each stratum. However, the standard implementation of stratification, referred to as the residual method, relies on strong parametric assumptions of linearity and homogeneity between the instrument and the exposure to form the strata. If these stratification assumptions are violated, the instrumental variable assumptions may be violated in the strata even if they are satisfied in the population, resulting in misleading estimates. We propose a new stratification method, referred to as the doubly-ranked method, that does not require strict parametric assumptions to create strata with different average levels of the exposure such that the instrumental variable assumptions are satisfied within the strata. Our simulation study indicates that the doubly-ranked method can obtain unbiased stratum-specific estimates and appropriate coverage rates even when the effect of the instrument on the exposure is non-linear or heterogeneous. Moreover, it can also provide unbiased estimates when the exposure is coarsened (that is, rounded, binned into categories, or truncated), a scenario that is common in applied practice and leads to substantial bias in the residual method. We applied the proposed doubly-ranked method to investigate the effect of alcohol intake on systolic blood pressure, and found evidence of a positive effect of alcohol intake, particularly at higher levels of alcohol consumption.

Semiparametric efficient G-estimation with invalid instrumental variables

Summary The instrumental variable method is widely used in the health and social sciences for identification and estimation of causal effects in the presence of potential unmeasured confounding. To improve efficiency, multiple instruments are routinely used, raising concerns about bias due to possible violation of the instrumental variable assumptions. To address such concerns, we introduce a new class of G-estimators that are guaranteed to remain consistent and asymptotically normal for the causal effect of interest provided that a set of at least $\gamma$ out of $K$ candidate instruments are valid, for $\gamma \leqslant K$ set by the analyst ex ante without necessarily knowing the identities of the valid and invalid instruments. We provide formal semiparametric efficiency theory supporting our results. Simulation studies and applications to UK Biobank data demonstrate the superior empirical performance of the proposed estimators compared with competing methods.

Abstract 361: Bidirectional Mendelian Randomization Investigating The Causal Relationship Between Major Depressive Disorder And Peripheral Artery Disease

Background: Observational studies have demonstrated strong bi-directional associations between depression and peripheral artery disease (PAD), and patients suffering from PAD who have depressive symptoms are known to have worse outcomes. Whether these associations represent causal pathways remains unknown but has important implications for disease management. Methods: We performed causal inference experiments using bi-directional Mendelian randomization. We employed summary statistics from the 2019 Psychiatric Genomics Consortium meta-analysis of major depressive disorder (MDD) comprised of 807,553 participants (246,363 cases) of European ancestry and the VA Million Veteran Program GWAS of PAD comprised of 243,060 participants (31,307 cases) of multiple ancestries to test the association between genetic liability to MDD and genetic liability to PAD. Genetic instruments for each exposure were constructed using independent (r 2 < 0.1 by 1000 Genomes EUR LD reference panel) variants associated with the exposure at genome-wide significance (p < 5x10 -8 ). Inverse-variance weighted MR was used as the primary analysis method, with weighted-median MR, MR-PRESSO, and Steiger filtering based MR performed as sensitivity analyses. F-statistics were calculated to assess weak-instrument bias. The Egger bias-intercept test was used to detect evidence of horizontal pleiotropy. Results: Increased genetic liability to MDD was associated with increased genetic liability to PAD (OR 1.2, 95%CI 1.1 - 1.3, p=.0001). We could not detect an effect of genetic liability to PAD on the genetic liability to MDD (OR 1.03, 95%CI 0.99 - 1.08). The results of sensitivity analyses that make different instrumental variable assumptions were consistent with these findings. Conclusions: These results suggest a causal relationship in which MDD increases the risk of PAD, but do not provide evidence to support the causal role of PAD in MDD. Ongoing analyses are focused on identifying potential mediators of the association between MDD and PAD and performing causal inference for the effect of other mental health traits on PAD.

Abstract 10816: Causal Effect of Atrial Fibrillation on Cardiovascular Diseases - A Mendelian Randomization Analysis in the Million Veteran Program

Introduction: Observational studies have implicated atrial fibrillation (AF) as a risk factor for a number of cardiovascular diseases (CVD). Whether AF is causally related to these CVD is unknown. Methods: We included participants with both electronic health record-derived phenotypes and imputed genotype data in this study. A genetic instrument for AF consisting of 406 common variants was derived using a pruning and thresholding method (r2<0.3; p-value <5x10-8) on the 2018 AF Genetics Consortium GWAS summary statistics. 1- and 2-sample MR analyses were performed. In 2-sample MR, the inverse-variance weighted method was used as the primary analysis. Additional sensitivity analyses were performed to account for potential violations of the instrumental variable assumptions. Results: We included 615,635 participants in the study (440,690 European, 118,531 African, 48,908 Hispanic, 7,506 Asian ancestry). Genetic risk of AF was associated with heart failure in both 1-sample (OR/log odds of genetically predicted AF =1.18, 95% CI 1.15 - 1.20) and 2-sample MR (OR/log odds of genetically predicted AF =1.13, 95% CI 1.11 - 1.15). Additionally, genetic risk of AF was associated with heart failure subtypes and non-ischemic cardiomyopathy. There was no association between genetic risk of AF and pulmonary embolism (PE) or deep venous thrombosis (DVT), however, genetic risk of AF was associated with PE after excluding individuals with concurrent or prior DVT (OR1-sample MR 1.08, 95% CI 1.01 - 1.15; OR2-sample MR 1.06, 95% CI 1.03 - 1.08). Lastly, we confirmed an association between genetic risk of AF and ischemic stroke but not all-cause dementia. Our results were robust to a number of sensitivity analyses. Conclusions: In a large multi-ancestry study, we delineate the potential causal contribution of AF to a number of CVD and highlight support for a causal association between AF and isolated PE. Our findings inform how prevention and treatment of AF may affect CVD incidence and outcomes.

Semiparametric optimal estimation with nonignorable nonresponse data

When the response mechanism is believed to be not missing at random (NMAR), a valid analysis requires stronger assumptions on the response mechanism than standard statistical methods would otherwise require. Semiparametric estimators have been developed under the parametric model assumptions on the response mechanism. In this paper, a new statistical test is proposed to guarantee model identifiability without using instrumental variable assumption. Furthermore, we develop optimal semiparametric estimation for parameters such as the population mean. Specifically, we propose two semiparametric optimal estimators that do not require any model assumptions other than the response mechanism. Asymptotic properties of the proposed estimators are discussed. An extensive simulation study is presented to compare with some existing methods. We present an application of our method using Korean labor and income panel survey data.

586Effects of maternal circulating amino acids on offspring birthweight: a Mendelian randomisation analysis

Abstract Background It is suggested amino acids are critical for fetal growth, but analyses assessing causality are lacking. Mendelian randomisation (MR) can be used to examine causal effects under instrumental variable (IV) assumptions. Methods We conducted a two-sample MR study utilizing summary data from genome-wide association studies (GWAS) of amino acids (sample 1, n = 86,507) and of offspring birthweight (sample 2, combined UK Biobank and Early Growth Genetics Consortium, n = 406,063). Seventy-five independent single nucleotide polymorphisms (SNPs) robustly associated with 18 amino acids (p < 4.9 × 10-10) were used as genetic instruments. Wald ratio and inverse variance weighted methods were used in MR main analysis. Sensitivity analyses were performed to explore IV assumption violations. To explore whether there was consistency between SNP-amino acid associations in pregnancy and in the GWAS, the latter were compared to associations in the Born in Bradford cohort. Results There was evidence of positive causal effects of maternal alanine (51.9 g birthweight increase per SD increase in amino acid level, 95% CI: 24.2, 79.5), glutamine (51.3 g, 95% CI: 33.5, 69.0), glycine (10.4 g, 95% CI: 1.3, 19.6) and serine (27.1 g, 95% CI: 11.2, 43.0) on birthweight and inverse causal effects of maternal isoleucine (-109.7 g, 95% CI: -194.6, -24.9) and histidine (-41.1 g, 95% CI: -78.5, -3.7) on birthweight. Sensitivity analyses to explore reverse causality and bias due to horizontal pleiotropy supported our findings. Conclusions Some maternal circulating amino acids have causal effects on birthweight. Key messages MR can be extended to probe effects of maternal nutrition on offspring development.

Anchor Regression: Heterogeneous Data Meet Causality

AbstractWe consider the problem of predicting a response variable from a set of covariates on a data set that differs in distribution from the training data. Causal parameters are optimal in terms of predictive accuracy if in the new distribution either many variables are affected by interventions or only some variables are affected, but the perturbations are strong. If the training and test distributions differ by a shift, causal parameters might be too conservative to perform well on the above task. This motivates anchor regression, a method that makes use of exogenous variables to solve a relaxation of the ‘causal’ minimax problem by considering a modification of the least-squares loss. The procedure naturally provides an interpolation between the solutions of ordinary least squares (OLS) and two-stage least squares. We prove that the estimator satisfies predictive guarantees in terms of distributional robustness against shifts in a linear class; these guarantees are valid even if the instrumental variable assumptions are violated. If anchor regression and least squares provide the same answer (‘anchor stability’), we establish that OLS parameters are invariant under certain distributional changes. Anchor regression is shown empirically to improve replicability and protect against distributional shifts.

Blood Pressure and Risk of Cardiovascular Disease in UK Biobank: A Mendelian Randomization Study.

This study aims to evaluate the causal association of blood pressure (BP) with cardiovascular diseases (CVDs). Two-sample Mendelian randomization was performed using a large genome-wide association study (n=299 024) and the UK Biobank cohort (n=375 256). We identified 327 and 364 single-nucleotide polymorphisms strongly and independently associated with systolic BP and diastolic BP, respectively, as genetic instruments to assess the causal association of BP with total CVD, CVD mortality, and 14 cardiovascular conditions. Nonlinearity was examined with nonlinear instrumental variable assumptions. Genetically predicted BP was significantly positively associated with total CVD (systolic BP, per 10 mm Hg: odds ratio [OR], 1.32 [95% CI, 1.25-1.40]; diastolic BP, per 5 mm Hg: OR, 1.20 [95% CI, 1.15-1.26]). Similar positive causal associations were observed for 14 cardiovascular conditions including ischemic heart disease (systolic BP, per 10 mm Hg: OR, 1.33 [95% CI, 1.24-1.41]; diastolic BP, per 5 mm Hg: OR, 1.20 [95% CI, 1.14-1.27]) and stroke (systolic BP, per 10 mm Hg: OR, 1.35 [95% CI, 1.24-1.48]; diastolic BP, per 5 mm Hg: OR, 1.20 [95% CI, 1.12-1.28]). Nonlinearity Mendelian randomization test demonstrated linear causal association of BP with these outcomes. Consistent estimates were observed in sensitivity analyses, suggesting robustness of the associations and minimal horizontal pleiotropy. The linear positive causal association of BP and CVD was consistent with previous findings that lower BP is better, thus consolidating clinical knowledge on hypertension management in CVD risk reduction.

Direct and Indirect Effects under Sample Selection and Outcome Attrition

This paper extends the evaluation of direct and indirect treatment effects, i.e., mediation analysis, to the case that outcomes are only partially observed due to sample selection or outcome attrition. We assume sequential conditional independence of the treatment and the mediator, i.e., the variable through which the indirect effect operates. We also impose missing at random or instrumental variable assumptions on the outcome attrition process. Under these conditions, we derive identification results for the effects of interest that are based on inverse probability weighting by specific treatment, mediator, and/or selection propensity scores. We also provide a simulation study and an empirical application to the U.S. Project STAR data in which we assess the direct impact and indirect effect (via absenteeism) of smaller kindergarten classes on math test scores. The estimators considered are available in the ‘causalweight’ package for the statistical software ‘R’.

Is Being an only Child Harmful to Psychological Health?: Evidence from an Instrumental Variable Analysis of China's one-child Policy

SummaryThe paper evaluates the effects of being an only child in a family on psychological health, leveraging data on the one-child policy in China. We use an instrumental variable approach to address the potential unmeasured confounding between the fertility decision and psychological health, where the instrumental variable is an index of the intensity of the implementation of the policy. We establish an analytical link between the local instrumental variable approach and principal stratification to accommodate the continuous instrumental variable. Within the principal stratification framework, we postulate a Bayesian hierarchical model to infer various causal estimands of policy interest while adjusting for the clustering data structure. We apply the method to the data from the China Family Panel Studies and find small but statistically significant negative effects of being an only child on self-reported psychological health for some subpopulations. Our analysis reveals treatment effect heterogeneity with respect to both observed and unobserved characteristics. In particular, urban males suffer the most from being only children, and the negative effect has larger magnitude if the families were more resistant to the one-child policy. We also conduct a sensitivity analysis to assess the key instrumental variable assumption.

The APPROACH trial: Assessing pain, patient-reported outcomes, and complementary and integrative health.

Electronic health record data can be used in multiple ways to facilitate real-world pragmatic studies. Electronic health record data can provide detailed information about utilization of treatment options to help identify appropriate comparison groups, access historical clinical characteristics of participants, and facilitate measuring longitudinal outcomes for the treatments being studied. An additional novel use of electronic health record data is to assess and understand referral pathways and other business practices that encourage or discourage patients from using different types of care. We describe an ongoing study utilizing access to real-time electronic health record data about changing patterns of complementary and integrative health services to demonstrate how electronic health record data can provide the foundation for a pragmatic study when randomization is not feasible. Conducting explanatory trials of the value of emerging therapies within a healthcare system poses ethical and pragmatic challenges, such as withholding access to specific services that are becoming widely available to patients. We describe how prospective examination of real-time electronic health record data can be used to construct and understand business practices as potential surrogates for direct randomization through an instrumental variables analytic approach. In this context, an example of a business practice is the internal hiring of acupuncturists who also provide yoga or Tai Chi classes and can offer these classes without additional cost compared to community acupuncturists. Here, the business practice of hiring internal acupuncturists is likely to encourage much higher rates of combined complementary and integrative health use compared to community referrals. We highlight the tradeoff in efficiency of this pragmatic approach and describe use of simulations to estimate the potential sample sizes needed for a variety of instrument strengths. While real-time monitoring of business practices from electronic health records provides insights into the validity of key independence assumptions associated with the instrumental variable approaches, we note that there may be some residual confounding by indication or selection bias and describe how alternative sources of electronic health record data can be used to assess the robustness of instrumental variable assumptions to address these challenges. Finally, we also highlight that while some clinical outcomes can be obtained directly from the electronic health record, such as longitudinal opioid utilization and pain intensity levels for the study of the value of complementary and integrative health, it is often critical to supplement clinical electronic health record-based measures with patient-reported outcomes. The experience of this example in evaluating complementary and integrative health demonstrates the use of electronic health record data in several novel ways that may be of use for designing future pragmatic trials.

Generalized instrumental inequalities: testing the instrumental variable independence assumption

Summary This paper proposes a new set of testable implications for the instrumental variable independence assumption for discrete treatment, but unrestricted outcome and instruments: generalized instrumental inequalities. When outcome and treatment are both binary, but instruments are unrestricted, we show that the generalized instrumental inequalities are necessary and sufficient to detect all observable violations of the instrumental variable independence assumption. To test the generalized instrumental inequalities, we propose an approach combining a sample splitting procedure and an inference method for intersection bounds. This idea allows one to easily implement the test using existing Stata packages. We apply our proposed strategy to assess the validity of the instrumental variable independence assumption for various instruments used in the returns to college literature.

Assessing causal treatment effect estimation when using large observational datasets

BackgroundRecently, there has been a heightened interest in developing and evaluating different methods for analysing observational data. This has been driven by the increased availability of large data resources such as Electronic Health Record (EHR) data alongside known limitations and changing characteristics of randomised controlled trials (RCTs). A wide range of methods are available for analysing observational data. However, various, sometimes strict, and often unverifiable assumptions must be made in order for the resulting effect estimates to have a causal interpretation. In this paper we will compare some common approaches to estimating treatment effects from observational data in order to highlight the importance of considering, and justifying, the relevant assumptions prior to conducting an observational analysis.MethodsA simulation study was conducted based upon a small cohort of patients with chronic obstructive pulmonary disease. Two-stage least squares instrumental variables, propensity score, and linear regression models were compared under a range of different scenarios including different strengths of instrumental variable and unmeasured confounding. The effects of violating the assumptions of the instrumental variables analysis were also assessed. Sample sizes of up to 200,000 patients were considered.ResultsTwo-stage least squares instrumental variable methods can yield unbiased treatment effect estimates in the presence of unmeasured confounding provided the sample size is sufficiently large. Adjusting for measured covariates in the analysis reduces the variability in the two-stage least squares estimates. In the simulation study, propensity score methods produced very similar results to linear regression for all scenarios. A weak instrument or strong unmeasured confounding led to an increase in uncertainty in the two-stage least squares instrumental variable effect estimates. A violation of the instrumental variable assumptions led to bias in the two-stage least squares effect estimates. Indeed, these were sometimes even more biased than those from a naïve linear regression model.ConclusionsInstrumental variable methods can perform better than naïve regression and propensity scores. However, the assumptions need to be carefully considered and justified prior to conducting an analysis or performance may be worse than if the problem of unmeasured confounding had been ignored altogether.

Robust methods in Mendelian randomization via penalization of heterogeneous causal estimates.

Methods have been developed for Mendelian randomization that can obtain consistent causal estimates under weaker assumptions than the standard instrumental variable assumptions. The median-based estimator and MR-Egger are examples of such methods. However, these methods can be sensitive to genetic variants with heterogeneous causal estimates. Such heterogeneity may arise from over-dispersion in the causal estimates, or specific variants with outlying causal estimates. In this paper, we develop three extensions to robust methods for Mendelian randomization with summarized data: 1) robust regression (MM-estimation); 2) penalized weights; and 3) Lasso penalization. Methods using these approaches are considered in two applied examples: one where there is evidence of over-dispersion in the causal estimates (the causal effect of body mass index on schizophrenia risk), and the other containing outliers (the causal effect of low-density lipoprotein cholesterol on Alzheimer’s disease risk). Through an extensive simulation study, we demonstrate that robust regression applied to the inverse-variance weighted method with penalized weights is a worthwhile additional sensitivity analysis for Mendelian randomization to provide robustness to variants with outlying causal estimates. The results from the applied examples and simulation study highlight the importance of using methods that make different assumptions to assess the robustness of findings from Mendelian randomization investigations with multiple genetic variants.

Epidemiology, genetic epidemiology and Mendelian randomisation: more need than ever to attend to detail

In the current era, with increasing availability of results from genetic association studies, finding genetic instruments for inferring causality in observational epidemiology has become apparently simple. Mendelian randomisation (MR) analyses are hence growing in popularity and, in particular, methods that can incorporate multiple instruments are being rapidly developed for these applications. Such analyses have enormous potential, but they all rely on strong, different, and inherently untestable assumptions. These have to be clearly stated and carefully justified for every application in order to avoid conclusions that cannot be replicated. In this article, we review the instrumental variable assumptions and discuss the popular linear additive structural model. We advocate the use of tests for the null hypothesis of ‘no causal effect’ and calculation of the bounds for a causal effect, whenever possible, as these do not rely on parametric modelling assumptions. We clarify the difference between a randomised trial and an MR study and we comment on the importance of validating instruments, especially when considering them for joint use in an analysis. We urge researchers to stand by their convictions, if satisfied that the relevant assumptions hold, and to interpret their results causally since that is the only reason for performing an MR analysis in the first place.

Statistics in Brief: Instrumental Variable Analysis: An Underutilized Method in Orthopaedic Research.

Statistics in Brief: Instrumental Variable Analysis: An Underutilized Method in Orthopaedic Research.

Meta-analysis and Mendelian randomization: A review.

Mendelian randomization (MR) uses genetic variants as instrumental variables to infer whether a risk factor causally affects a health outcome. Meta‐analysis has been used historically in MR to combine results from separate epidemiological studies, with each study using a small but select group of genetic variants. In recent years, it has been used to combine genome‐wide association study (GWAS) summary data for large numbers of genetic variants. Heterogeneity among the causal estimates obtained from multiple genetic variants points to a possible violation of the necessary instrumental variable assumptions. In this article, we provide a basic introduction to MR and the instrumental variable theory that it relies upon. We then describe how random effects models, meta‐regression, and robust regression are being used to test and adjust for heterogeneity in order to improve the rigor of the MR approach.