Sensitivity analysis for interactions under unmeasured confounding

Unveiling challenges in Mendelian randomization for gene-environment interaction

Gene-environment (GxE) interactions play a crucial role in understanding the complex etiology of various traits, but assessing them using observational data can be challenging due to unmeasured confounders for lifestyle and environmental risk factors. Mendelian randomization (MR) has emerged as a valuable method for assessing causal relationships based on observational data. This approach utilizes genetic variants as instrumental variables (IVs) with the aim of providing a valid statistical test and estimation of causal effects in the presence of unmeasured confounders. MR has gained substantial popularity in recent years largely due to the success of genome-wide association studies. Many methods have been developed for MR; however, limited work has been done on evaluating GxE interaction. In this paper, we focus on two primary IV approaches: the two-stage predictor substitution and the two-stage residual inclusion, and extend them to accommodate GxE interaction under both the linear and logistic regression models for continuous and binary outcomes, respectively. Comprehensive simulation study and analytical derivations reveal that resolving the linear regression model is relatively straightforward. In contrast, the logistic regression model presents a considerably more intricate challenge, which demands additional effort.

Intergenerational continuity of loneliness and potential mechanisms: Young Finns Multigenerational Study

Evidence on the intergenerational continuity of loneliness and on potential mechanisms that connect loneliness across successive generations is limited. We examined the association between loneliness of (G0) parents (859 mothers and 570 fathers, mean age 74 years) and their children (G1) (433 sons and 558 daughters, mean age 47 years) producing 991 parent-offspring pairs and tested whether these associations were mediated through subjective socioeconomic position, temperament characteristics, cognitive performance, and depressive symptoms. Mean loneliness across parents had an independent effect on their adult children's experienced loneliness (OR = 1.72, 95% CI 1.23-2.42). We also found a robust effect of mothers' (OR = 1.64, 95% CI 1.17-2.29), but not of fathers' loneliness (OR = 1.47, 95% CI 0.96-2.25) on offspring's experienced loneliness in adulthood. The associations were partly mediated by offspring depressive (41-54%) and anxiety (29-31%) symptoms. The current findings emphasize the high interdependence of loneliness within families mediated partly by offspring's mental health problems.

Misclassification Bias in the Assessment of Gene-by-Environment Interactions

BACKGROUND

Misclassification bias is a common concern in epidemiologic studies. Despite strong bias on main effects, gene-environment interactions have been shown to be biased towards the null under gene-environment independence. In the context of a recent article examining the interaction between nerve agent exposure and paraoxonase-1 gene on Gulf War Illness, we aimed to assess the impact of recall bias-a common misclassfication bias-on the identification of gene-environment interactions when the independence assumption is violated.

METHODS

We derive equations to quantify the bias of the interaction, and numerically illustrate these results by simulating a case-control study of 1000 cases and 1000 controls. Simulation input parameters included exposure prevalence, strength of gene-environment dependence, strength of the main effect, exposure specificity among cases, and strength of the gene-environment interaction.

RESULTS

We show that, even if gene-environment independence is violated, we can bound possible gene-environment interactions by knowing the strength and direction of the gene-environment dependence ( ) and the observed gene-environment interaction ( )-thus often still allowing for the identification of such interactions. Depending on whether is larger or smaller than the inverse of , is a lower (if ) or upper (if ) bound for the true interaction. In addition, the bias magnitude is somewhat predictable by examining other characteristics such as exposure prevalence, the strength of the exposure main effect, and directions of the recall bias and gene-environment dependence.

CONCLUSIONS

Even if gene-environment dependence exists, we may still be able to identify gene-environment interactions even when misclassification bias is present.

Evaluation of a Gene-Environment Interaction of PON1 and Low-Level Nerve Agent Exposure with Gulf War Illness: A Prevalence Case-Control Study Drawn from the U.S. Military Health Survey's National Population Sample

BACKGROUND

Consensus on the etiology of 1991 Gulf War illness (GWI) has been limited by lack of objective individual-level environmental exposure information and assumed recall bias.

OBJECTIVES

We investigated a prestated hypothesis of the association of GWI with a gene-environment (GxE) interaction of the paraoxonase-1 (PON1) Q192R polymorphism and low-level nerve agent exposure.

METHODS

A prevalence sample of 508 GWI cases and 508 nonpaired controls was drawn from the 8,020 participants in the U.S. Military Health Survey, a representative sample survey of military veterans who served during the Gulf War. The PON1 Q192R genotype was measured by real-time polymerase chain reaction (RT-PCR), and the serum Q and R isoenzyme activity levels were measured with PON1-specific substrates. Low-level nerve agent exposure was estimated by survey questions on having heard nerve agent alarms during deployment.

RESULTS

The GxE interaction of the Q192R genotype and hearing alarms was strongly associated with GWI on both the multiplicative [prevalence odds ratio (POR) of the interaction=3.41; 95% confidence interval (CI): 1.20, 9.72] and additive (synergy index=4.71; 95% CI: 1.82, 12.19) scales, adjusted for measured confounders. The Q192R genotype and the alarms variable were independent (adjusted POR in the controls=1.18; 95% CI: 0.81, 1.73; p=0.35), and the associations of GWI with the number of R alleles and quartiles of Q isoenzyme were monotonic. The adjusted relative excess risk due to interaction (aRERI) was 7.69 (95% CI: 2.71, 19.13). Substituting Q isoenzyme activity for the genotype in the analyses corroborated the findings. Sensitivity analyses suggested that recall bias had forced the estimate of the GxE interaction toward the null and that unmeasured confounding is unlikely to account for the findings. We found a GxE interaction involving the Q-correlated PON1 diazoxonase activity and a weak possible GxE involving the Khamisiyah plume model, but none involving the PON1 R isoenzyme activity, arylesterase activity, paraoxonase activity, butyrylcholinesterase genotypes or enzyme activity, or pyridostigmine.

DISCUSSION

Given gene-environment independence and monotonicity, the unconfounded aRERI>0 supports a mechanistic interaction. Together with the direct evidence of exposure to fallout from bombing of chemical weapon storage facilities and the extensive toxicologic evidence of biochemical protection from organophosphates by the Q isoenzyme, the findings provide strong evidence for an etiologic role of low-level nerve agent in GWI. https://doi.org/10.1289/EHP9009.

E-values for effect heterogeneity and approximations for causal interaction

Background

Estimates of effect heterogeneity (i.e. the extent to which the causal effect of one exposure varies across strata of a second exposure) can be biased if the exposure–outcome relationship is subject to uncontrolled confounding whose severity differs across strata of the second exposure.

Methods

We propose methods, analogous to the E-value for total effects, that help to assess the sensitivity of effect heterogeneity estimates to possible uncontrolled confounding. These E-value analogues characterize the severity of uncontrolled confounding strengths that would be required, hypothetically, to ‘explain away’ an estimate of multiplicative or additive effect heterogeneity in the sense that appropriately controlling for those confounder(s) would have shifted the effect heterogeneity estimate to the null, or alternatively would have shifted its confidence interval to include the null. One can also consider shifting the estimate or confidence interval to an arbitrary non-null value. All of these E-values can be obtained using the R package EValue.

Results

We illustrate applying the proposed E-value analogues to studies on: (i) effect heterogeneity by sex of the effect of educational attainment on dementia incidence and (ii) effect heterogeneity by age on the effect of obesity on all-cause mortality.

Conclusion

Reporting these proposed E-values could help characterize the robustness of effect heterogeneity estimates to potential uncontrolled confounding.

The Joint Effects of Depression and Comorbid Psychiatric Disorders on Suicide Deaths: Competing Antagonism as an Explanation for Subadditivity

BACKGROUND

Previous studies of the effect of interaction between psychiatric disorders on suicide have reported mixed results. We investigated the joint effect of depression and various comorbid psychiatric disorders on suicide.

METHODS

We conducted a population-based case-cohort study with all suicide deaths occurring between 1 January 1995 and 31 December 2015 in Denmark (n = 14,103) and a comparison subcohort comprised of a 5% random sample of the source population at baseline (n = 265,183). We quantified the joint effect of pairwise combinations of depression and major psychiatric disorders (e.g., organic disorders, substance use disorders, schizophrenia, bipolar disorder, neurotic disorders, eating disorders, personality disorders, intellectual disabilities, developmental disorders, and behavioral disorders) on suicide using marginal structural models and calculated the relative excess risk due to interaction. We assessed for the presence of competing antagonism for negative relative excess risk due to interactions.

RESULTS

All combinations of depression and comorbid psychiatric disorders were associated with increased suicide risk. For example, the rate of suicide among men with depression and neurotic disorders was 20 times (95% CI = 15, 26) the rate in men with neither disorder. Most disorder combinations were associated with subadditive suicide risk, and there was evidence of competing antagonism in most of these cases.

CONCLUSIONS

Subadditivity may be explained by competing antagonism. When both depression and a comorbid psychiatric disorder are present, they may compete to cause the outcome such that having 2 disorders may be no worse than having a single disorder with respect to suicide risk.

Black Lives Matter: A Decomposition of Racial Inequalities in Oral Cancer Screening

Simple Summary

Black Lives Matter has highlighted the increased social discrepancies that exist not only in the context of social justice but also in public health. Oral cancer screening is not exempt from disparity, with Black Americans less likely to seek screening leading to higher incidence and worse outcomes of oropharyngeal cancers. We investigate interaction analysis and Blinder–Oaxaca decomposition as tools to guide policy to address this disparity. Using National Health and Nutrition Examination Survey (NHANES) data from 2011–2018 we find that being both in poverty and Black results in sub-additive interaction, which is further deconstructed into differences in higher education levels and poverty status.

Abstract

(1) Background: The Black Lives Matter movement has highlighted the discrepancies in public health in regard to race. This study aims to investigate tools that can be used to analyze and investigate this discrepancy, which can be applied to policymaking. (2) Methods: National Health and Nutrition Examination Survey (NHANES) data from 2011–2018 was combined (N = 22,617) to investigate discrepancies of oral cancer screening in Black Americans. We give examples of counterfactual techniques that can be used to guide policy. Inverse probability treatment weighting (IPTW) was used to remove all measured confounding in an interaction analysis to assess the combined effect of socioeconomic status and race. Blinder–Oaxaca decomposition was then used to investigate the intervenable factors associated with differences in race. (3) Results: Sub-additive interaction was found on additive and multiplicative scales when all measured confounding was removed via IPTW (relative excess risk due to interaction (RERI)(OR) = −0.55 (−0.67–−0.42)). Decomposition analysis found that 32% of the discrepancy could be explained by characteristics of higher education and poverty status. (4) Conclusions: Black Americans in poverty are less likely to seek oral cancer screening than the additive likelihood would suggest. Blinder–Oaxaca decomposition is a strong tool to use for guiding policy as it quantifies clear breakdowns of what intervenable factors there are that would improve the discrepancy the most.

Moving from two- to multi-way interactions among binary risk factors on the additive scale

Many studies have focused on investigating deviations from additive interaction of two dichotomous risk factors on a binary outcome. There is, however, a gap in the literature with respect to interactions on the additive scale of >2 risk factors. In this paper, we present an approach for examining deviations from additive interaction among three or more binary exposures. The relative excess risk due to interaction (RERI) is used as measure of additive interaction. First, we concentrate on three risk factors – we propose to decompose the total RERI to: the RERI owned to the joint presence of all three risk factors and the RERI of any two risk factors, given that the third is absent. We then extend this approach, to >3 binary risk factors. For illustration, we use a sample from data from the Greek EPIC cohort and we investigate the association with overall mortality of Mediterranean diet, body mass index , and smoking. Our formulae enable better interpretability of any evidence for deviations from additivity owned to more than two risk factors and provide simple ways of communicating such results from a public health perspective by attributing any excess relative risk to specific combinations of these factors.

Abbreviations: BMI: Body Mass Index; ERR: excess relative risk; EPIC: European Prospective Investigation into Cancer and nutrition; MD: Mediterranean diet; RERI: relative excess risk due to interaction; RR: relative risk; TotRERI: total relative excess risk due to interaction

Estimating the Relative Excess Risk Due to Interaction in Clustered-Data Settings

The risk difference scale is often of primary interest when evaluating public health impacts of interventions on binary outcomes. However, few investigators report findings in terms of additive interaction, probably because the models typically used for binary outcomes implicitly measure interaction on the multiplicative scale. One measure with which to assess additive interaction from multiplicative models is the relative excess risk due to interaction (RERI). The RERI measure has been applied in many contexts, but one limitation of previous approaches is that clustering in data has rarely been considered. We evaluated the RERI metric for the setting of clustered data using both population-averaged and cluster-conditional models. In simulation studies, we found that estimation and inference for the RERI using population-averaged models was straightforward. However, frequentist implementations of cluster-conditional models including random intercepts often failed to converge or produced degenerate variance estimates. We developed a Bayesian implementation of log binomial random-intercept models, which represents an attractive alternative for estimating the RERI in cluster-conditional models. We applied the methods to an observational study of adverse birth outcomes in mothers with human immunodeficiency virus, in which mothers were clustered within clinical research sites.

Studying additive interaction in a healthcare database: Case study of NSAIDs, cardiovascular profiles, and acute myocardial infarction

Purpose

There are clinical trial data on risk of acute myocardial infarction (MI) with nonsteroidal anti-inflammatory drugs (NSAIDs) in patients at increased cardiovascular (CV) risk requiring chronic daily treatment. This study investigated whether risks of acute MI with real-world prescription NSAIDs, such as low-dose or intermittent use, vary according to an individual’s CV profile.

Methods

Nested case-control analyses were carried out on an administrative health cohort from Quebec, Canada by randomly selecting 10 controls per case matched on age ± 1 year, sex, and month and year of cohort entry. We measured the additive joint effects on acute MI of current NSAID use and presence of hypertension, coronary heart disease (CHD), history of previous MI, or concomitant use of cardioprotective aspirin. The endpoint was the relative excess risk due to interaction (RERI). To verify the robustness of interaction findings, we performed sensitivity analyses with varying specifications of NSAID exposure-related variables.

Results

The cohort consisted of 233 816 elderly individuals, including 21 256 acute MI cases. For hypertension, CHD, and previous MI, we identified additive interactions on MI risk with some but not all NSAIDs, which also depended on the definition of NSAID exposure. Hypertension was sub-additive with naproxen but not with the other NSAIDs. Celecoxib and CHD were sub-additive in the primary analysis only (modelling NSAID dose on index date or up to 7 days before–best-fitting base model) whereas celecoxib and rofecoxib were super-additive with a history of previous MI in the secondary analysis only (modelling NSAID use on index date). For cardioprotective aspirin we found no evidence for an additive interaction with any of the NSAIDs.

Conclusions

Alternative specifications of NSAID exposure concurred in finding that concomitant use of cardioprotective aspirin does not attenuate the risks of acute MI with NSAIDs. However we were unable to demonstrate consistent interactions between an individual’s cardiovascular comorbidities and NSAID-associated acute MI. Our study highlights challenges of studying additive interactions in a healthcare database and underscores the need for sensitivity analyses.

Case-only approach to identifying markers predicting treatment effects on the relative risk scale

Retrospectively measuring markers on stored baseline samples from participants in a randomized controlled trial (RCT) may provide high quality evidence as to the value of the markers for treatment selection. Originally developed for approximating gene-environment interactions in the odds ratio scale, the case-only method has recently been advocated for assessing gene-treatment interactions on rare disease endpoints in randomized clinical trials. In this article, the case-only approach is shown to provide a consistent and efficient estimator of marker by treatment interactions and marker-specific treatment effects on the relative risk scale. The prohibitive rare-disease assumption is no longer needed, broadening the utility of the case-only approach. The case-only method is resource-efficient as markers only need to be measured in cases only. It eliminates the need to model the marker's main effect, and can be used with any parametric or nonparametric learning method. The utility of this approach is illustrated by an application to genetic data in the Women's Health Initiative (WHI) hormone therapy trial.

Decomposition of the Total Effect in the Presence of Multiple Mediators and Interactions

Mediation analysis allows decomposing a total effect into a direct effect of the exposure on the outcome and an indirect effect operating through a number of possible hypothesized pathways. Recent studies have provided formal definitions of direct and indirect effects when multiple mediators are of interest and have described parametric and semiparametric methods for their estimation. Investigating direct and indirect effects with multiple mediators, however, can be challenging in the presence of multiple exposure-mediator and mediator-mediator interactions. In this paper we derive a decomposition of the total effect that unifies mediation and interaction when multiple mediators are present. We illustrate the properties of the proposed framework in a secondary analysis of a pragmatic trial for the treatment of schizophrenia. The decomposition is employed to investigate the interplay of side effects and psychiatric symptoms in explaining the effect of antipsychotic medication on quality of life in schizophrenia patients. Our result offers a valuable tool to identify the proportions of total effect due to mediation and interaction when more than one mediator is present, providing the finest decomposition of the total effect that unifies multiple mediators and interactions.

Association of Maternal Prepregnancy Dyslipidemia With Adult Offspring Dyslipidemia in Excess of Anthropometric, Lifestyle, and Genetic Factors in the Framingham Heart Study

IMPORTANCE

Dyslipidemia in young adults in the United States during their childbearing years is common, and the consequences for the next generation are poorly understood. Further understanding of the harmful consequences of elevated low-density lipoprotein cholesterol (LDL-C) levels in young adults may help to inform population screening and management strategies.

OBJECTIVE

To examine whether adult levels of serum LDL-C are associated with maternal prepregnancy LDL-C levels beyond that attributable to inherited genetic sequence polymorphisms, diet, physical activity, and body mass index.

DESIGN, SETTING, AND PARTICIPANTS

The Framingham Heart Study is a multigenerational, population-based inception cohort initiated in 1948 in Framingham, Massachusetts. In this study of families, the analyses included 538 parent-offspring pairs with parental LDL-C levels measured in the study prior to the offspring's birth. Parental prebirth, parental concurrent, and adult offspring assessments occurred in 1971-1983, 1998-2001, and 2002-2005, respectively. Data analyses were conducted between March 1, 2013, and May 30, 2015.

EXPOSURES

Maternal prepregnancy LDL-C levels compared with paternal prepregnancy and parental concurrent LDL-C levels in association with adult offspring LDL-C levels.

MAIN OUTCOMES AND MEASURES

Adult offspring LDL-C levels were examined as both a continuous and dichotomous outcome (using a threshold of 130 mg/dL).

RESULTS

Among the 538 parent-offspring pairs, there were 241 mother-offspring and 297 father-offspring pairs with a mean (SD) offspring age of 26 (3) years. Adult offspring LDL-C levels were associated with maternal prepregnancy LDL-C levels after adjustment for family relatedness and offspring lifestyle, anthropometric factors, and inherited genetic variants (β = 0.32 [SE, 0.05] mg/dL; P < .001). After multivariable adjustment, adults who had been exposed to elevated maternal prepregnancy LDL-C levels were at a 3.8 (95% CI, 1.5-9.8) times higher odds of having elevated LDL-C levels (P = .005) and had an adjusted LDL-C level of 18 mg/dL (95% CI, 9-27 mg/dL) higher than did those without such exposure. Maternal prepregnancy LDL-C levels explained 13% of the variation in adult offspring LDL-C levels beyond common genetic variants and classic risk factors for elevated LDL-C levels.

CONCLUSIONS AND RELEVANCE

Adult offspring dyslipidemia is associated with maternal prepregnancy dyslipidemia in excess of measured lifestyle, anthropometric, and inherited genetic factors. The findings support the possibility of a maternal epigenetic contribution to cardiovascular disease risk in the general population. Further research is warranted to determine whether ongoing public health efforts to identify and reduce dyslipidemia in young adults prior to their childbearing years may have additional potential health benefits for the subsequent generation.

A critical appraisal of pharmacogenetic inference

In essence, pharmacogenetic research is aimed at discovering variants of importance to gene-treatment interaction. However, epidemiological studies are rarely set up with this goal in mind. It is therefore of great importance that researchers clearly communicate which assumptions they have had to make, and which inherent limitations apply to the interpretation of their results. This review discusses considerations of, and the underlying assumptions for, utilizing different response phenotypes and study designs popular in pharmacogenetic research to infer gene-treatment interaction effects, with a special focus on those dealing with of clinical effects of drug treatment.

Bias Analysis for Uncontrolled Confounding in the Health Sciences

Uncontrolled confounding due to unmeasured confounders biases causal inference in health science studies using observational and imperfect experimental designs. The adoption of methods for analysis of bias due to uncontrolled confounding has been slow, despite the increasing availability of such methods. Bias analysis for such uncontrolled confounding is most useful in big data studies and systematic reviews to gauge the extent to which extraneous preexposure variables that affect the exposure and the outcome can explain some or all of the reported exposure-outcome associations. We review methods that can be applied during or after data analysis to adjust for uncontrolled confounding for different outcomes, confounders, and study settings. We discuss relevant bias formulas and how to obtain the required information for applying them. Finally, we develop a new intuitive generalized bias analysis framework for simulating and adjusting for the amount of uncontrolled confounding due to not measuring and adjusting for one or more confounders.

Assessing moderated mediation in linear models requires fewer confounding assumptions than assessing mediation

It is well known from the mediation analysis literature that the identification of direct and indirect effects relies on strong no unmeasured confounding assumptions of no unmeasured confounding. Even in randomized studies the mediator may still be correlated with unobserved prognostic variables that affect the outcome, in which case the mediator's role in the causal process may not be inferred without bias. In the behavioural and social science literature very little attention has been given so far to the causal assumptions required for moderated mediation analysis. In this paper we focus on the index for moderated mediation, which measures by how much the mediated effect is larger or smaller for varying levels of the moderator. We show that in linear models this index can be estimated without bias in the presence of unmeasured common causes of the moderator, mediator and outcome under certain conditions. Importantly, one can thus use the test for moderated mediation to support evidence for mediation under less stringent confounding conditions. We illustrate our findings with data from a randomized experiment assessing the impact of being primed with social deception upon observer responses to others' pain, and from an observational study of individuals who ended a romantic relationship assessing the effect of attachment anxiety during the relationship on mental distress 2 years after the break-up.

Tests for Gene-Environment Interactions and Joint Effects With Exposure Misclassification

The number of methods for genome-wide testing of gene-environment (G-E) interactions continues to increase, with the aim of discovering new genetic risk factors and obtaining insight into the disease-gene-environment relationship. The relative performance of these methods, assessed on the basis of family-wise type I error rate and power, depends on underlying disease-gene-environment associations, estimates of which may be biased in the presence of exposure misclassification. This simulation study expands on a previously published simulation study of methods for detecting G-E interactions by evaluating the impact of exposure misclassification. We consider 7 single-step and modular screening methods for identifying G-E interaction at a genome-wide level and 7 joint tests for genetic association and G-E interaction, for which the goal is to discover new genetic susceptibility loci by leveraging G-E interaction when present. In terms of statistical power, modular methods that screen on the basis of the marginal disease-gene relationship are more robust to exposure misclassification. Joint tests that include main/marginal effects of a gene display a similar robustness, which confirms results from earlier studies. Our results offer an increased understanding of the strengths and limitations of methods for genome-wide searches for G-E interaction and joint tests in the presence of exposure misclassification.

Prenatal Antidepressant Exposure and Childhood Autism Spectrum Disorders: Cause for Concern?

There is empirical evidence for a role for serotonin in autism . In experimental animals, early life exposure to serotonergic antidepressants or maternal stress affects brain development, with subsequent changes in serotonin tone in adult animals. Recently, antidepressant exposure during pregnancy has been associated with autism in epidemiological studies. At least part of the association is potentially explained by maternal depression or factors associated with depression. Importantly, even if there is no causal relation between prenatal antidepressant exposure and autism, use of antidepressants during pregnancy is a marker of potential problems later in life across five independent study populations, and exposed children may need special attention regardless of the underlying mechanism. Future studies need to disentangle the effects of maternal depression and antidepressant use during pregnancy while adjusting for the postnatal environment. One promising strategy is to use results from basic science to guide the inclusion of potential biological intermediates in advanced epidemiological studies.

On the practice of ignoring center-patient interactions in evaluating hospital performance

We evaluate the performance of medical centers based on a continuous or binary patient outcome (e.g., 30‐day mortality). Common practice adjusts for differences in patient mix through outcome regression models, which include patient‐specific baseline covariates (e.g., age and disease stage) besides center effects. Because a large number of centers may need to be evaluated, the typical model postulates that the effect of a center on outcome is constant over patient characteristics. This may be violated, for example, when some centers are specialized in children or geriatric patients. Including interactions between certain patient characteristics and the many fixed center effects in the model increases the risk for overfitting, however, and could imply a loss of power for detecting centers with deviating mortality. Therefore, we assess how the common practice of ignoring such interactions impacts the bias and precision of directly and indirectly standardized risks. The reassuring conclusion is that the common practice of working with the main effects of a center has minor impact on hospital evaluation, unless some centers actually perform substantially better on a specific group of patients and there is strong confounding through the corresponding patient characteristic. The bias is then driven by an interplay of the relative center size, the overlap between covariate distributions, and the magnitude of the interaction effect. Interestingly, the bias on indirectly standardized risks is smaller than on directly standardized risks. We illustrate our findings by simulation and in an analysis of 30‐day mortality on Riksstroke. © 2015 The Authors. Statistics in Medicine published by John Wiley & Sons Ltd.

Molecular association of pathogenetic contributors to pre-eclampsia (pre-eclampsia associome)

Background

Pre-eclampsia is the most common complication occurring during pregnancy. In the majority of cases, it is concurrent with other pathologies in a comorbid manner (frequent co-occurrences in patients), such as diabetes mellitus, gestational diabetes and obesity. Providing bronchial asthma, pulmonary tuberculosis, certain neurodegenerative diseases and cancers as examples, we have shown previously that pairs of inversely comorbid pathologies (rare co-occurrences in patients) are more closely related to each other at the molecular genetic level compared with randomly generated pairs of diseases. Data in the literature concerning the causes of pre-eclampsia are abundant. However, the key mechanisms triggering this disease that are initiated by other pathological processes are thus far unknown. The aim of this work was to analyse the characteristic features of genetic networks that describe interactions between comorbid diseases, using pre-eclampsia as a case in point.

Results

The use of ANDSystem, Pathway Studio and STRING computer tools based on text-mining and database-mining approaches allowed us to reconstruct associative networks, representing molecular genetic interactions between genes, associated concurrently with comorbid disease pairs, including pre-eclampsia, diabetes mellitus, gestational diabetes and obesity. It was found that these associative networks statistically differed in the number of genes and interactions between them from those built for randomly chosen pairs of diseases. The associative network connecting all four diseases was composed of 16 genes (PLAT, ADIPOQ, ADRB3, LEPR, HP, TGFB1, TNFA, INS, CRP, CSRP1, IGFBP1, MBL2, ACE, ESR1, SHBG, ADA). Such an analysis allowed us to reveal differential gene risk factors for these diseases, and to propose certain, most probable, theoretical mechanisms of pre-eclampsia development in pregnant women. The mechanisms may include the following pathways: [TGFB1 or TNFA]-[IL1B]-[pre-eclampsia]; [TNFA or INS]-[NOS3]-[pre-eclampsia]; [INS]-[HSPA4 or CLU]-[pre-eclampsia]; [ACE]-[MTHFR]-[pre-eclampsia].

Conclusions

For pre-eclampsia, diabetes mellitus, gestational diabetes and obesity, we showed that the size and connectivity of the associative molecular genetic networks, which describe interactions between comorbid diseases, statistically exceeded the size and connectivity of those built for randomly chosen pairs of diseases. Recently, we have shown a similar result for inversely comorbid diseases. This suggests that comorbid and inversely comorbid diseases have common features concerning structural organization of associative molecular genetic networks.

Nonparametric Bounds and Sensitivity Analysis of Treatment Effects

This paper considers conducting inference about the effect of a treatment (or exposure) on an outcome of interest. In the ideal setting where treatment is assigned randomly, under certain assumptions the treatment effect is identifiable from the observable data and inference is straightforward. However, in other settings such as observational studies or randomized trials with noncompliance, the treatment effect is no longer identifiable without relying on untestable assumptions. Nonetheless, the observable data often do provide some information about the effect of treatment, that is, the parameter of interest is partially identifiable. Two approaches are often employed in this setting: (i) bounds are derived for the treatment effect under minimal assumptions, or (ii) additional untestable assumptions are invoked that render the treatment effect identifiable and then sensitivity analysis is conducted to assess how inference about the treatment effect changes as the untestable assumptions are varied. Approaches (i) and (ii) are considered in various settings, including assessing principal strata effects, direct and indirect effects and effects of time-varying exposures. Methods for drawing formal inference about partially identified parameters are also discussed.

Modification of the association between lead exposure and amyotrophic lateral sclerosis by iron and oxidative stress related gene polymorphisms

Our objective was to examine whether functional polymorphisms in hemochromatosis (HFE; H63D and C282Y), transferrin (TfC2), and glutathione-s-transferase Pi1 (GSTP1; Ile105Val) genes modify any lead-ALS association. We measured blood lead using atomic absorption spectroscopy and bone lead - a biomarker of cumulative lead exposure - using K-shell-X-ray fluorescence in 100 neurologist-confirmed ALS cases and 194 controls, the latter recruited as part of two separate studies; all subjects lived in New England. Participants were considered variant carriers or wild-type for each polymorphism. To assess effect modification, we included cross-product terms between lead biomarkers and each polymorphism in separate adjusted polytomous logistic regression models. Compared with wild-type, the odds ratio (OR) per 15.6 μg/g patella lead (interquartile range; IQR) was 8.24 (95% CI 0.94-72.19) times greater among C282Y variant carriers, and 0.34 (95% CI 0.15-0.78) times smaller among H63D variant carriers. Results were weaker for tibia lead. Compared with wild-type the OR per 2 μg/dl blood lead (IQR) was 0.36 (95% CI 0.19-0.68) times smaller among H63D variant carriers, and 1.96 (95% CI 0.98-3.92) times greater among GSTP1 variant carriers. In conclusion, we found that HFE and GSTP1 genotypes modified the association between lead biomarkers and ALS. Contrasting modification by the HFE polymorphisms H63D and C282Y may suggest that the modification is not simply the result of increased iron.

A unification of mediation and interaction: a 4-way decomposition

The overall effect of an exposure on an outcome, in the presence of a mediator with which the exposure may interact, can be decomposed into 4 components: (1) the effect of the exposure in the absence of the mediator, (2) the interactive effect when the mediator is left to what it would be in the absence of exposure, (3) a mediated interaction, and (4) a pure mediated effect. These 4 components, respectively, correspond to the portion of the effect that is due to neither mediation nor interaction, to just interaction (but not mediation), to both mediation and interaction, and to just mediation (but not interaction). This 4-way decomposition unites methods that attribute effects to interactions and methods that assess mediation. Certain combinations of these 4 components correspond to measures for mediation, whereas other combinations correspond to measures of interaction previously proposed in the literature. Prior decompositions in the literature are in essence special cases of this 4-way decomposition. The 4-way decomposition can be carried out using standard statistical models, and software is provided to estimate each of the 4 components. The 4-way decomposition provides maximum insight into how much of an effect is mediated, how much is due to interaction, how much is due to both mediation and interaction together, and how much is due to neither.

The role of environmental heterogeneity in meta-analysis of gene-environment interactions with quantitative traits

With challenges in data harmonization and environmental heterogeneity across various data sources, meta-analysis of gene-environment interaction studies can often involve subtle statistical issues. In this paper, we study the effect of environmental covariate heterogeneity (within and between cohorts) on two approaches for fixed-effect meta-analysis: the standard inverse-variance weighted meta-analysis and a meta-regression approach. Akin to the results in Simmonds and Higgins (), we obtain analytic efficiency results for both methods under certain assumptions. The relative efficiency of the two methods depends on the ratio of within versus between cohort variability of the environmental covariate. We propose to use an adaptively weighted estimator (AWE), between meta-analysis and meta-regression, for the interaction parameter. The AWE retains full efficiency of the joint analysis using individual level data under certain natural assumptions. Lin and Zeng (2010a, b) showed that a multivariate inverse-variance weighted estimator retains full efficiency as joint analysis using individual level data, if the estimates with full covariance matrices for all the common parameters are pooled across all studies. We show consistency of our work with Lin and Zeng (2010a, b). Without sacrificing much efficiency, the AWE uses only univariate summary statistics from each study, and bypasses issues with sharing individual level data or full covariance matrices across studies. We compare the performance of the methods both analytically and numerically. The methods are illustrated through meta-analysis of interaction between Single Nucleotide Polymorphisms in FTO gene and body mass index on high-density lipoprotein cholesterol data from a set of eight studies of type 2 diabetes.

A Tutorial on Interaction

In this tutorial, we provide a broad introduction to the topic of interaction between the effects of exposures. We discuss interaction on both additive and multiplicative scales using risks, and we discuss their relation to statistical models (e.g. linear, log-linear, and logistic models). We discuss and evaluate arguments that have been made for using additive or multiplicative scales to assess interaction. We further discuss approaches to presenting interaction analyses, different mechanistic forms of interaction, when interaction is robust to unmeasured confounding, interaction for continuous outcomes, qualitative or “crossover” interactions, methods for attributing effects to interactions, case-only estimators of interaction, and power and sample size calculations for additive and multiplicative interaction.

A review of covariate selection for non-experimental comparative effectiveness research

This paper addresses strategies for selecting variables for adjustment in non-experimental comparative effectiveness research and uses causal graphs to illustrate the causal network that relates treatment to outcome. Variables in the causal network take on multiple structural forms. Adjustment for a common cause pathway between treatment and outcome can remove confounding, whereas adjustment for other structural types may increase bias. For this reason, variable selection would ideally be based on an understanding of the causal network; however, the true causal network is rarely known. Therefore, we describe more practical variable selection approaches based on background knowledge when the causal structure is only partially known. These approaches include adjustment for all observed pretreatment variables thought to have some connection to the outcome, all known risk factors for the outcome, and all direct causes of the treatment or the outcome. Empirical approaches, such as forward and backward selection and automatic high-dimensional proxy adjustment, are also discussed. As there is a continuum between knowing and not knowing the causal, structural relations of variables, we recommend addressing variable selection in a practical way that involves a combination of background knowledge and empirical selection and that uses high-dimensional approaches. This empirical approach can be used to select from a set of a priori variables based on the researcher's knowledge to be included in the final analysis or to identify additional variables for consideration. This more limited use of empirically derived variables may reduce confounding while simultaneously reducing the risk of including variables that may increase bias.

Test for interactions between a genetic marker set and environment in generalized linear models

We consider in this paper testing for interactions between a genetic marker set and an environmental variable. A common practice in studying gene-environment (GE) interactions is to analyze one single-nucleotide polymorphism (SNP) at a time. It is of significant interest to analyze SNPs in a biologically defined set simultaneously, e.g. gene or pathway. In this paper, we first show that if the main effects of multiple SNPs in a set are associated with a disease/trait, the classical single SNP-GE interaction analysis can be biased. We derive the asymptotic bias and study the conditions under which the classical single SNP-GE interaction analysis is unbiased. We further show that, the simple minimum p-value-based SNP-set GE analysis, can be biased and have an inflated Type 1 error rate. To overcome these difficulties, we propose a computationally efficient and powerful gene-environment set association test (GESAT) in generalized linear models. Our method tests for SNP-set by environment interactions using a variance component test, and estimates the main SNP effects under the null hypothesis using ridge regression. We evaluate the performance of GESAT using simulation studies, and apply GESAT to data from the Harvard lung cancer genetic study to investigate GE interactions between the SNPs in the 15q24-25.1 region and smoking on lung cancer risk.

Environmental confounding in gene-environment interaction studies

We show that, in the presence of uncontrolled environmental confounding, joint tests for the presence of a main genetic effect and gene-environment interaction will be biased if the genetic and environmental factors are correlated, even if there is no effect of either the genetic factor or the environmental factor on the disease. When environmental confounding is ignored, such tests will in fact reject the joint null of no genetic effect with a probability that tends to 1 as the sample size increases. This problem with the joint test vanishes under gene-environment independence, but it still persists if estimating the gene-environment interaction parameter itself is of interest. Uncontrolled environmental confounding will bias estimates of gene-environment interaction parameters even under gene-environment independence, but it will not do so if the unmeasured confounding variable itself does not interact with the genetic factor. Under gene-environment independence, if the interaction parameter without controlling for the environmental confounder is nonzero, then there is gene-environment interaction either between the genetic factor and the environmental factor of interest or between the genetic factor and the unmeasured environmental confounder. We evaluate several recently proposed joint tests in a simulation study and discuss the implications of these results for the conduct of gene-environment interaction studies.

A framework for the analysis of heterogeneity of treatment effect in patient-centered outcomes research

Individuals vary in their response to a treatment. Understanding this heterogeneity of treatment effect is critical for evaluating how well a treatment can be expected to work for an individual or a subgroup of individuals. An overemphasis on hypothesis testing has resulted in a dichotomy of all heterogeneity of treatment effect analyses into confirmatory (hypothesis testing) and exploratory (hypothesis finding) analyses. This limited view of heterogeneity of treatment effect is inadequate for creating evidence that is useful for informing patient-centered decisions. An expanded framework for heterogeneity of treatment effect assessment is proposed. It recognizes four distinct goals of heterogeneity of treatment effect analyses: hypothesis testing, hypothesis finding, reporting subgroup effects for meta-analysis, and individual-level prediction. Accordingly, two new types of heterogeneity of treatment effect analyses are proposed: descriptive and predictive. Descriptive heterogeneity of treatment effect analyses report treatment effects for prespecified subgroups in accordance with prospectively specified analytic strategy. They need not be powered to detect heterogeneity of treatment effect. They emphasize estimation and reporting of subgroup effects rather than hypothesis testing. Sampling properties (e.g., standard error) of descriptive analysis can be characterized, thus facilitating meta-analysis of subgroup effects. Predictive heterogeneity of treatment effect analyses estimate probabilities of beneficial and adverse responses of individuals to treatments and facilitates optimal treatment decisions for different types of individuals. Procedures are also suggested to improve reliability of heterogeneity of treatment effect assessment from observational studies. Heterogeneity of treatment effect analysis should be identified as confirmatory, descriptive, exploratory, or predictive analysis. Evidence should be interpreted in a manner consistent with the analytic goal.

A three-way decomposition of a total effect into direct, indirect, and interactive effects

Recent theory in causal inference has provided concepts for mediation analysis and effect decomposition that allow one to decompose a total effect into a direct and an indirect effect. Here, it is shown that what is often taken as an indirect effect can in fact be further decomposed into a "pure" indirect effect and a mediated interactive effect, thus yielding a three-way decomposition of a total effect (direct, indirect, and interactive). This three-way decomposition applies to difference scales and also to additive ratio scales and additive hazard scales. Assumptions needed for the identification of each of these three effects are discussed and simple formulae are given for each when regression models allowing for interaction are used. The three-way decomposition is illustrated by examples from genetic and perinatal epidemiology, and discussion is given to what is gained over the traditional two-way decomposition into a direct and an indirect effect.

Invited commentary: assessing mechanistic interaction between coinfecting pathogens for diarrheal disease

The interaction estimates from Bhavnani et al. (Am J Epidemiol. 2012;176(5):387-395) are used to evaluate evidence for mechanistic interaction between coinfecting pathogens for diarrheal disease. Mechanistic interaction is said to be present if there are individuals for whom the outcome would occur if both of 2 exposures are present but would not occur if 1 or the other of the exposures is absent. In the epidemiologic literature, mechanistic interaction is often conceived of as synergism within Rothman's sufficient-cause framework. Tests for additive interaction are sometimes used to assess such synergism or mechanistic interaction, but testing for positive additive interaction only allows for the conclusion of mechanistic interaction under fairly strong "monotonicity" assumptions. Alternative tests for mechanistic interaction, which do not require monotonicity assumptions, have been developed more recently but require more substantial additive interaction to draw the conclusion of the presence of mechanistic interaction. The additive interaction reported by Bhavnani et al. is of sufficient magnitude to provide strong evidence of mechanistic interaction between rotavirus and Giardia and between rotavirus and Escherichia. coli/Shigellae, even without any assumptions about monotonicity.