Eric J. Tchetgen

Negative Controls: A Tool for Detecting Confounding and Bias in Observational Studies

Noncausal associations between exposures and outcomes are a threat to validity of causal inference in observational studies. Many techniques have been developed for study design and analysis to identify and eliminate such errors. Such problems are not expected to compromise experimental studies, where careful standardization of conditions (for laboratory work) and randomization (for population studies) should, if applied properly, eliminate most such noncausal associations. We argue, however, that a routine precaution taken in the design of biologic laboratory experiments—the use of “negative controls”—is designed to detect both suspected and unsuspected sources of spurious causal inference. In epidemiology, analogous negative controls help to identify and resolve confounding as well as other sources of error, including recall bias or analytic flaws. We distinguish 2 types of negative controls (exposure controls and outcome controls), describe examples of each type from the epidemiologic literature, and identify the conditions for the use of such negative controls to detect confounding. We conclude that negative controls should be more commonly employed in observational studies, and that additional work is needed to specify the conditions under which negative controls will be sensitive detectors of other sources of error in observational studies.

Weight loss and survival in HIV-positive patients in the era of highly active antiretroviral therapy.

Weight loss and wasting have long been established as strong predictors of mortality in HIV-infected patients. Today, despite the effectiveness of highly active antiretroviral therapy (HAART), there is evidence that HIV-related wasting is still an important comorbidity in many patients. We conducted a study to determine if wasting is still associated with decreased survival in patients receiving HAART and which parameter (weight, fat-free mass [FFM], body cell mass [BCM], or fat mass [FM]) is most strongly associated with mortality. The study population consisted of 678 HIV-positive participants enrolled in the Nutrition for Healthy Living study. Weight, FFM, BCM, and FM were assessed for all participants at 6-month intervals. At each follow-up visit, percent losses of each parameter were calculated from values at baseline and the previous visit. Cox proportional hazards models were used to estimate and compare the relative risks of death for each parameter, adjusting for potential confounders such as HAART use, body mass index, and CD4 cell counts. In analyses examining the parameters separately and together in the same model, weight loss emerged as the strongest independent predictor of mortality. Weight loss of >or=10% from baseline or the previous visit was significantly associated with a four- to sixfold increase in mortality compared with maintenance or gaining of weight. Even one episode of weight loss of >or=3% from baseline or >or=5% from the previous visit was predictive of mortality. In summary, despite the apparent benefits of HAART use on HIV-related survival, weight loss remains an independent predictor of mortality. In addition, FFM or BCM estimated using bioelectrical impedance analysis does not add further prognostic value over weight loss.

Credible Mendelian Randomization Studies: Approaches for Evaluating the Instrumental Variable Assumptions

As with other instrumental variable (IV) analyses, Mendelian randomization (MR) studies rest on strong assumptions. These assumptions are not routinely systematically evaluated in MR applications, although such evaluation could add to the credibility of MR analyses. In this article, the authors present several methods that are useful for evaluating the validity of an MR study. They apply these methods to a recent MR study that used fat mass and obesity-associated (FTO) genotype as an IV to estimate the effect of obesity on mental disorder. These approaches to evaluating assumptions for valid IV analyses are not fail-safe, in that there are situations where the approaches might either fail to identify a biased IV or inappropriately suggest that a valid IV is biased. Therefore, the authors describe the assumptions upon which the IV assessments rely. The methods they describe are relevant to any IV analysis, regardless of whether it is based on a genetic IV or other possible sources of exogenous variation. Methods that assess the IV assumptions are generally not conclusive, but routinely applying such methods is nonetheless likely to improve the scientific contributions of MR studies.

On causal inference in the presence of interference

Interference is said to be present when the exposure or treatment received by one individual may affect the outcomes of other individuals. Such interference can arise in settings in which the outcomes of the various individuals come about through social interactions. When interference is present, causal inference is rendered considerably more complex, and the literature on causal inference in the presence of interference has just recently begun to develop. In this article we summarise some of the concepts and results from the existing literature and extend that literature in considering new results for finite sample inference, new inverse probability weighting estimators in the presence of interference and new causal estimands of interest.

Methodological challenges in mendelian randomization.

We give critical attention to the assumptions underlying Mendelian randomization analysis and their biological plausibility. Several scenarios violating the Mendelian randomization assumptions are described, including settings with inadequate phenotype definition, the setting of time-varying exposures, the presence of gene–environment interaction, the existence of measurement error, the possibility of reverse causation, and the presence of linkage disequilibrium. Data analysis examples are given, illustrating that the inappropriate use of instrumental variable techniques when the Mendelian randomization assumptions are violated can lead to biases of enormous magnitude. To help address some of the strong assumptions being made, three possible approaches are suggested. First, the original proposal of Katan (Lancet. 1986;1:507–508) for Mendelian randomization was not to use instrumental variable techniques to obtain estimates but merely to examine genotype–outcome associations to test for the presence of an effect of the exposure on the outcome. We show that this more modest goal and approach can circumvent many, though not all, the potential biases described. Second, we discuss the use of sensitivity analysis in evaluating the consequences of violations in the assumptions and in attempting to correct for those violations. Third, we suggest that a focus on negative, rather than positive, Mendelian randomization results may turn out to be more reliable.

The relationship between neighborhood poverty and alcohol use: estimation by marginal structural models

Background: Previous studies on the relationship of neighborhood disadvantage with alcohol use or misuse have often controlled for individual characteristics on the causal pathway, such as income—thus potentially underestimating the relationship between disadvantage and alcohol consumption. Methods: We used data from the Coronary Artery Risk Development in Young Adults study of 5115 adults aged 18–30 years at baseline and interviewed 7 times between 1985 and 2006. We estimated marginal structural models using inverse probability-of-treatment and censoring weights to assess the association between point-in-time/cumulative exposure to neighborhood poverty (proportion of census tract residents living in poverty) and alcohol use/binging, after accounting for time-dependent confounders including income, education, and occupation. Results: The log-normal model was used to estimate treatment weights while accounting for highly-skewed continuous neighborhood poverty data. In the weighted model, a one-unit increase in neighborhood poverty at the prior examination was associated with a 86% increase in the odds of binging (OR = 1.86 [95% confidence interval = 1.14–3.03]); the estimate from a standard generalized-estimating-equations model controlling for baseline and time-varying covariates was 1.47 (0.96–2.25). The inverse probability-of-treatment and censoring weighted estimate of the relative increase in the number of weekly drinks in the past year associated with cumulative neighborhood poverty was 1.53 (1.02–2.27); the estimate from a standard model was 1.16 (0.83–1.62). Conclusions: Cumulative and point-in-time measures of neighborhood poverty are important predictors of alcohol consumption. Estimators that more closely approximate a causal effect of neighborhood poverty on alcohol provided a stronger estimate than estimators from traditional regression models.

Semiparametric Theory for Causal Mediation Analysis: efficiency bounds, multiple robustness, and sensitivity analysis.

Whilst estimation of the marginal (total) causal effect of a point exposure on an outcome is arguably the most common objective of experimental and observational studies in the health and social sciences, in recent years, investigators have also become increasingly interested in mediation analysis. Specifically, upon evaluating the total effect of the exposure, investigators routinely wish to make inferences about the direct or indirect pathways of the effect of the exposure not through or through a mediator variable that occurs subsequently to the exposure and prior to the outcome. Although powerful semiparametric methodologies have been developed to analyze observational studies, that produce double robust and highly efficient estimates of the marginal total causal effect, similar methods for mediation analysis are currently lacking. Thus, this paper develops a general semiparametric framework for obtaining inferences about so-called marginal natural direct and indirect causal effects, while appropriately accounting for a large number of pre-exposure confounding factors for the exposure and the mediator variables. Our analytic framework is particularly appealing, because it gives new insights on issues of efficiency and robustness in the context of mediation analysis. In particular, we propose new multiply robust locally efficient estimators of the marginal natural indirect and direct causal effects, and develop a novel double robust sensitivity analysis framework for the assumption of ignorability of the mediator variable.

Genetic Variants on 15q25.1, Smoking, and Lung Cancer: An Assessment of Mediation and Interaction

Genome-wide association studies have identified variants on chromosome 15q25.1 that increase the risks of both lung cancer and nicotine dependence and associated smoking behavior. However, there remains debate as to whether the association with lung cancer is direct or is mediated by pathways related to smoking behavior. Here, the authors apply a novel method for mediation analysis, allowing for gene-environment interaction, to a lung cancer case-control study (1992-2004) conducted at Massachusetts General Hospital using 2 single nucleotide polymorphisms, rs8034191 and rs1051730, on 15q25.1. The results are validated using data from 3 other lung cancer studies. Tests for additive interaction (P = 2 × 10(-10) and P = 1 × 10(-9)) and multiplicative interaction (P = 0.01 and P = 0.01) were significant. Pooled analyses yielded a direct-effect odds ratio of 1.26 (95% confidence interval (CI): 1.19, 1.33; P = 2 × 10(-15)) for rs8034191 and an indirect-effect odds ratio of 1.01 (95% CI: 1.00, 1.01; P = 0.09); the proportion of increased risk mediated by smoking was 3.2%. For rs1051730, direct- and indirect-effect odds ratios were 1.26 (95% CI: 1.19, 1.33; P = 1 × 10(-15)) and 1.00 (95% CI: 0.99, 1.01; P = 0.22), respectively, with a proportion mediated of 2.3%. Adjustment for measurement error in smoking behavior allowing up to 75% measurement error increased the proportions mediated to 12.5% and 9.2%, respectively. These analyses indicate that the association of the variants with lung cancer operates primarily through other pathways.

Botswana's progress toward achieving the 2020 UNAIDS 90-90-90 antiretroviral therapy and virological suppression goals: a population-based survey

BACKGROUND HIV programmes face challenges achieving high rates of HIV testing and treatment needed to optimise health and to reduce transmission. We used data from the Botswana Combination Prevention Project study survey to assess Botswanas progress toward achieving UNAIDS targets for 2020: 90% of all people living with HIV knowing their status, 90% of these receiving sustained antiretroviral therapy (ART), and 90% of those having virological suppression (90-90-90). METHODS A population-based sample of individuals was recruited and interviewed in 30 rural and periurban communities from Oct 30, 2013, to Nov 24, 2015, as part of a large, ongoing community-randomised trial designed to assess the effect of a combination prevention package on HIV incidence. A random sample of about 20% of households in each community was selected. Consenting household residents aged 16-64 years who were Botswana citizens or spouses of citizens responded to a questionnaire and had blood drawn for HIV testing in the absence of documentation of positive HIV status. Viral load testing was done in all HIV-infected participants, irrespective of treatment status. We used modified Poisson generalised estimating equations to obtain prevalence ratios, corresponding Huber robust SEs, and 95% Wald CIs to examine associations between individual sociodemographic factors and a binary outcome indicating achievement of the three individual and combined overall 90-90-90 targets. The study is registered at ClinicalTrials.gov, number NCT01965470. FINDINGS 81% of enumerated eligible household members took part in the survey (10% refused and 9% were absent). Among 12 610 participants surveyed, 3596 (29%) were infected with HIV, and 2995 (83·3%, 95% CI 81·4-85·2) of these individuals already knew their HIV status. Among those who knew their HIV status, 2617 (87·4%, 95% CI 85·8-89·0) were receiving ART (95% of those eligible by national guidelines, and 73% of all infected people). Of the 2609 individuals receiving ART with a viral load measurement, 2517 (96·5%, 95% CI 96·0-97·0) had viral load of 400 copies per mL or less. Overall, 70·2% (95% CI 67·5-73·0) of HIV-infected people had virological suppression, close to the UNAIDS target of 73%. INTERPRETATION UNAIDS 90-90-90 targets are achievable even in resource-constrained settings with high HIV burden. FUNDING US Presidents Emergency Plan for AIDS Relief through the Centers for Disease Control and Prevention.

Instrumental variable estimation in a survival context

Bias due to unobserved confounding can seldom be ruled out with certainty when estimating the causal effect of a nonrandomized treatment. The instrumental variable (IV) design offers, under certain assumptions, the opportunity to tame confounding bias, without directly observing all confounders. The IV approach is very well developed in the context of linear regression and also for certain generalized linear models with a nonlinear link function. However, IV methods are not as well developed for regression analysis with a censored survival outcome. In this article, we develop the IV approach for regression analysis in a survival context, primarily under an additive hazards model, for which we describe 2 simple methods for estimating causal effects. The first method is a straightforward 2-stage regression approach analogous to 2-stage least squares commonly used for IV analysis in linear regression. In this approach, the fitted value from a first-stage regression of the exposure on the IV is entered in place of the exposure in the second-stage hazard model to recover a valid estimate of the treatment effect of interest. The second method is a so-called control function approach, which entails adding to the additive hazards outcome model, the residual from a first-stage regression of the exposure on the IV. Formal conditions are given justifying each strategy, and the methods are illustrated in a novel application to a Mendelian randomization study to evaluate the effect of diabetes on mortality using data from the Health and Retirement Study. We also establish that analogous strategies can also be used under a proportional hazards model specification, provided the outcome is rare over the entire follow-up.