The potential impact fraction of population weight reduction scenarios on non-communicable diseases in Belgium: application of the g-computation approach

BACKGROUND

Overweight is a major risk factor for non-communicable diseases (NCDs) in Europe, affecting almost 60% of all adults. Tackling obesity is therefore a key long-term health challenge and is vital to reduce premature mortality from NCDs. Methodological challenges remain however, to provide actionable evidence on the potential health benefits of population weight reduction interventions. This study aims to use a g-computation approach to assess the impact of hypothetical weight reduction scenarios on NCDs in Belgium in a multi-exposure context.

METHODS

Belgian health interview survey data (2008/2013/2018, n = 27 536) were linked to environmental data at the residential address. A g-computation approach was used to evaluate the potential impact fraction (PIF) of population weight reduction scenarios on four NCDs: diabetes, hypertension, cardiovascular disease (CVD), and musculoskeletal (MSK) disease. Four scenarios were considered: 1) a distribution shift where, for each individual with overweight, a counterfactual weight was drawn from the distribution of individuals with a "normal" BMI 2) a one-unit reduction of the BMI of individuals with overweight, 3) a modification of the BMI of individuals with overweight based on a weight loss of 10%, 4) a reduction of the waist circumference (WC) to half of the height among all people with a WC:height ratio greater than 0.5. Regression models were adjusted for socio-demographic, lifestyle, and environmental factors.

RESULTS

The first scenario resulted in preventing a proportion of cases ranging from 32.3% for diabetes to 6% for MSK diseases. The second scenario prevented a proportion of cases ranging from 4.5% for diabetes to 0.8% for MSK diseases. The third scenario prevented a proportion of cases, ranging from 13.6% for diabetes to 2.4% for MSK diseases and the fourth scenario prevented a proportion of cases ranging from 36.4% for diabetes to 7.1% for MSK diseases.

CONCLUSION

Implementing weight reduction scenarios among individuals with excess weight could lead to a substantial and statistically significant decrease in the prevalence of diabetes, hypertension, cardiovascular disease (CVD), and musculoskeletal (MSK) diseases in Belgium. The g-computation approach to assess PIF of interventions represents a straightforward approach for drawing causal inferences from observational data while providing useful information for policy makers.

Investigation of the structure and magnitude of time-varying uncontrolled confounding in simulated cohort data analyzed using g-computation

BACKGROUND

When estimating the effect of time-varying exposures on longer-term outcomes, the assumption of conditional exchangeability or no uncontrolled confounding extends beyond baseline confounding to include time-varying confounding. We illustrate the structures and magnitude of uncontrolled time-varying confounding in exposure effect estimates obtained from g-computation when sequential conditional exchangeability is violated.

METHODS

We used directed acyclic graphs (DAGs) to depict time-varying uncontrolled confounding. We performed simulations and used g-computation to quantify the effects of each time-varying exposure for each DAG type. Models adjusting all time-varying confounders were considered the true (bias-adjusted) estimate. The exclusion of time-varying uncontrolled confounders represented the biased effect estimate and an unmet 'no uncontrolled confounding' assumption. True and biased estimates were compared across DAGs, with different magnitudes of uncontrolled confounding.

RESULTS

Time-varying uncontrolled confounding can present in several scenarios, including relationships into subsequently measured exposure(s), outcome, unmeasured confounder(s) and other measured confounder(s). In simulations, effect estimates obtained from g-computation were more biased in DAGs when the uncontrolled confounders were directly related to the outcome. Complex DAGs that included relationships between uncontrolled confounders and other variables and relationships where exposures caused uncontrolled confounders at the next time point resulted in the most biased effect estimates. In these complex DAGs, excluding uncontrolled confounders affected the multiple effect estimates.

CONCLUSIONS

Time-varying uncontrolled confounding has the potential to substantially impact observed effect estimates. Given the importance of longitudinal studies in advising public health, the impact of time-varying uncontrolled confounding warrants more recognition and evaluation using quantitative bias analysis.

Community-Level Risk Factors for Firearm Assault and Homicide: The Role of Local Firearm Dealers and Alcohol Outlets

BACKGROUND

Identifying community characteristics associated with firearm assault could facilitate prevention. We investigated the effect of community firearm dealer and alcohol outlet densities on individual risk of firearm assault injury.

METHODS

In this density-sampled case-control study of Californians, January 2005-September 2015, cases comprised all residents with a fatal or nonfatal firearm assault injury. For each month, we sampled controls from the state population in a 4:1 ratio with cases. Exposures were monthly densities of county-level pawn and nonpawn firearm dealers and ZIP code-level off-premises alcohol outlets and bars and pubs ("bars/pubs"). We used case-control-weighted G-computation to estimate risk differences (RD) statewide and among younger Black men, comparing observed exposure densities to hypothetical interventions setting these densities to low. We estimated additive interactions between firearm and alcohol retailer density. Secondary analyses examined interventions targeted to high exposure density or outcome burden areas.

RESULTS

There were 67,850 cases and 268,122 controls. Observed (vs. low) densities of pawn firearm dealers and off-premises alcohol outlets were individually associated with elevated monthly risk of firearm assault per 100,000 people (RD pawn dealers : 0.06, 95% CI: 0.05, 0.08; RD off-premises outlets : 0.01, 95% CI: 0.01, 0.03), but nonpawn firearm dealer and bar/pub density were not; models targeting only areas with the highest outcome burden were similar. Among younger Black men, estimates were larger. There was no interaction between firearm and alcohol retailer density.

CONCLUSIONS

Our results are consistent with the hypothesis that limiting pawn firearm dealers and off-premises alcohol outlet densities can reduce interpersonal firearm violence.

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

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

Polychlorinated biphenyls, polychlorinated dibenzo-p-dioxins, polychlorinated dibenzofurans, pesticides, and diabetes in the Anniston Community Health Survey follow-up (ACHS II): single exposure and mixture analysis approaches

Dioxins and dioxin-like compounds measurements were added to polychlorinated biphenyls (PCBs) and organochlorine pesticides to expand the exposure profile in a follow-up to the Anniston Community Health Survey (ACHS II, 2014) and to study diabetes associations. Participants of ACHS I (2005-2007) still living within the study area were eligible to participate in ACHS II. Diabetes status (type-2) was determined by a doctor's diagnosis, fasting glucose ≥125 mg/dL, or being on any glycemic control medication. Incident diabetes cases were identified in ACHS II among those who did not have diabetes in ACHS I, using the same criteria. Thirty-five ortho-substituted PCBs, 6 pesticides, 7 polychlorinated dibenzo-p-dioxins (PCDD), 10 furans (PCDF), and 3 non-ortho PCBs were measured in 338 ACHS II participants. Dioxin toxic equivalents (TEQs) were calculated for all dioxin-like compounds. Main analyses used logistic regression models to calculate odds ratios (OR) and 95 % confidence intervals (CI). In models adjusted for age, race, sex, BMI, total lipids, family history of diabetes, and taking lipid lowering medication, the highest ORs for diabetes were observed for PCDD TEQ: 3.61 (95 % CI: 1.04, 12.46), dichloro-diphenyl dichloroethylene (p,p'-DDE): 2.07 (95 % CI 1.08, 3.97), and trans-Nonachlor: 2.55 (95 % CI 0.93, 7.02). The OR for sum 35 PCBs was 1.22 (95 % CI: 0.58-2.57). To complement the main analyses, we used BKMR and g-computation models to evaluate 12 mixture components including 4 TEQs, 2 PCB subsets and 6 pesticides; suggestive positive associations for the joint effect of the mixture analyses resulted in ORs of 1.40 (95% CI: -1.13, 3.93) for BKMR and 1.32 (95% CI: -1.12, 3.76) for g-computation. The mixture analyses provide further support to previously observed associations of trans-Nonachlor, p,p'- DDE, PCDD TEQ and some PCB groups with diabetes.

Marginal structural models with monotonicity constraints: A case study in out-of-hospital cardiac arrest patients

This paper deals with estimating the probability of a binary counterfactual outcome as a function of a continuous covariate under monotonicity constraints. We are motivated by the study of out-of-hospital cardiac arrest patients which aims to estimate the counterfactual 30-day survival probability if either all patients had received, or if none of the patients had received bystander cardiopulmonary resuscitation (CPR), as a function of the ambulance response time. It is natural to assume that the counterfactual 30-day survival probability cannot increase with increasing ambulance response time. We model the monotone relationship with a marginal structural model and B-splines. We then derive an estimating equation for the parameters of interest which however further relies on an auxiliary regression model for the observed 30-day survival probabilities. The predictions of the observed 30-day survival probabilities are used as pseudo-values for the unobserved counterfactual 30-day survival status. The methods are illustrated and contrasted with an unconstrained modeling approach in large-scale Danish registry data.

A connection between survival multistate models and causal inference for external treatment interruptions

Recently, treatment interruptions such as a clinical hold in randomized clinical trials have been investigated by using a multistate model approach. The phase III clinical trial START (Stimulating Targeted Antigenic Response To non-small-cell cancer) with primary endpoint overall survival was temporarily placed on hold for enrollment and treatment by the US Food and Drug Administration (FDA). Multistate models provide a flexible framework to account for treatment interruptions induced by a time-dependent external covariate. Extending previous work, we propose a censoring and a filtering approach both aimed at estimating the initial treatment effect on overall survival in the hypothetical situation of no clinical hold. A special focus is on creating a link to causal inference. We show that calculating the matrix of transition probabilities in the multistate model after application of censoring (or filtering) yields the desired causal interpretation. Assumptions in support of the identification of a causal effect by censoring (or filtering) are discussed. Thus, we provide the basis to apply causal censoring (or filtering) in more general settings such as the COVID-19 pandemic. A simulation study demonstrates that both causal censoring and filtering perform favorably compared to a naïve method ignoring the external impact.

A cautionary note on the use of G-computation in population adjustment

In a recent issue of the Journal; Remiro-Azócar et al. introduce a new method to adjust for population difference between two trials; when the individual patient data (IPD) are only accessible for one study. The proposed method generates the covariate data for the trial without IPD; then using a G-computation approach to transport information about the treatment effect from the other study with IPD to this trial. The authors advocate the use of G-computation over matching-adjusted indirect comparison because (i) the former allows for "useful extrapolation" when there is poor case-mix overlap between populations; and (ii) nonparametric; data-adaptive methods can be used to reduce the risk of (outcome) model misspecification. In this commentary; we provide a different perspective from these arguments. Despite certain disagreements; we believe that the proposed data generation approaches can open new and interesting research directions for population adjustment methodology in the future.

Mixture Effects of Polychlorinated Biphenyls (PCBs) and Three Organochlorine Pesticides on Cognitive Function in Mohawk Adults at Akwesasne

The Mohawks at Akwesasne have been highly exposed to polychlorinated biphenyls (PCBs), via releases from three aluminum foundries located near the reserve. They are also exposed to organochlorine pesticides, namely hexachlorobenzene (HCB), dichlorodiphenyldichloroethylene (DDE), and mirex. Previous studies have demonstrated reduced cognition in relation to total PCBs, but the effects of the mixtures of different PCB congener groups, HCB, DDE, and mirex on cognitive function have not been studied. Therefore, cognitive performance for executive function, scored via the digit symbol substitution test (DSST), in Mohawk adults aged 17-79 years (n = 301), was assessed in relation to serum concentrations of low-chlorinated PCBs, high-chlorinated PCBs, total PCBs, HCB, DDE, and mirex. We used mixture models employing the quantile-based g-computation method. The mixture effects of low-chlorinated PCBs, high-chlorinated PCBs, HCB, DDE, and mirex were significantly associated with 4.01 DSST scores decrements in the oldest age group, 47-79 years old. There were important contributions to mixture effects from low-chlorinated PCBs, high-chlorinated PCBs, and total PCBs, with smaller contributions of HCB and DDE. Our findings indicate that exposures to both low- and high-chlorinated PCBs increase the risk of cognitive decline in older adults, while DDE and HCB have less effect.

A Simulation Study Comparing the Performance of Time-Varying Inverse Probability Weighting and G-Computation in Survival Analysis

Inverse probability weighting (IPW) and g-computation are commonly used in time-varying analyses. To inform decisions on which to use, we compared these methods using a plasmode simulation based on data from the Effects of Aspirin in Gestation and Reproduction (EAGeR) Trial (June 15, 2007-July 15, 2011). In our main analysis, we simulated a cohort study of 1,226 individuals followed for up to 10 weeks. The exposure was weekly exercise, and the outcome was time to pregnancy. We controlled for 6 confounding factors: 4 baseline confounders (race, ever smoking, age, and body mass index) and 2 time-varying confounders (compliance with assigned treatment and nausea). We sought to estimate the average causal risk difference by 10 weeks, using IPW and g-computation implemented using a Monte Carlo estimator and iterated conditional expectations (ICE). Across 500 simulations, we compared the bias, empirical standard error (ESE), average standard error, standard error ratio, and 95% confidence interval coverage of each approach. IPW (bias = 0.02; ESE = 0.04; coverage = 92.6%) and Monte Carlo g-computation (bias = -0.01; ESE = 0.03; coverage = 94.2%) performed similarly. ICE g-computation was the least biased but least precise estimator (bias = 0.01; ESE = 0.06; coverage = 93.4%). When choosing an estimator, one should consider factors like the research question, the prevalences of the exposure and outcome, and the number of time points being analyzed.

Causal inference in case of near-violation of positivity: comparison of methods

In causal studies, the near-violation of the positivity may occur by chance, because of sample-to-sample fluctuation despite the theoretical veracity of the positivity assumption in the population. It may mostly happen when the exposure prevalence is low or when the sample size is small. We aimed to compare the robustness of g-computation (GC), inverse probability weighting (IPW), truncated IPW, targeted maximum likelihood estimation (TMLE), and truncated TMLE in this situation, using simulations and one real application. We also tested different extrapolation situations for the sub-group with a positivity violation. The results illustrated that the near-violation of the positivity impacted all methods. We demonstrated the robustness of GC and TMLE-based methods. Truncation helped in limiting the bias in near-violation situations, but at the cost of bias in normal conditions. The application illustrated the variability of the results between the methods and the importance of choosing the most appropriate one. In conclusion, compared to propensity score-based methods, methods based on outcome regression should be preferred when suspecting near-violation of the positivity assumption.

Causal Effect of Chronic Pain on Mortality Through Opioid Prescriptions: Application of the Front-Door Formula

BACKGROUND

Chronic pain is the leading cause of disability worldwide and is strongly associated with the epidemic of opioid overdosing events. However, the causal links between chronic pain, opioid prescriptions, and mortality remain unclear.

METHODS

This study included 13,884 US adults aged ≥20 years who provided data on chronic pain in the National Health and Nutrition Examination Survey 1999-2004 with linkage to mortality databases through 2015. We employed the generalized form of the front-door formula within the structural causal model framework to investigate the causal effect of chronic pain on all-cause mortality mediated by opioid prescriptions.

RESULTS

We identified a total of 718 participants at 3 years of follow-up and 1260 participants at 5 years as having died from all causes. Opioid prescriptions increased the risk of all-cause mortality with an estimated odds ratio (OR) (95% confidence interval) = 1.5 (1.1, 1.9) at 3 years and 1.3 (1.1, 1.6) at 5 years. The front-door formula revealed that chronic pain increased the risk of all-cause mortality through opioid prescriptions; OR = 1.06 (1.01, 1.11) at 3 years and 1.03 (1.01, 1.06) at 5 years. Our bias analysis showed that our findings based on the front-door formula were likely robust to plausible sources of bias from uncontrolled exposure-mediator or mediator-outcome confounding.

CONCLUSIONS

Chronic pain increased the risk of all-cause mortality through opioid prescriptions. Our findings highlight the importance of careful guideline-based chronic pain management to prevent death from possibly inappropriate opioid prescriptions driven by chronic pain.

Structural Racism and Quantitative Causal Inference: A Life Course Mediation Framework for Decomposing Racial Health Disparities

Quantitative studies of racial health disparities often use static measures of self-reported race and conventional regression estimators, which critics argue is inconsistent with social-constructivist theories of race, racialization, and racism. We demonstrate an alternative counterfactual approach to explain how multiple racialized systems dynamically shape health over time, examining racial inequities in cardiometabolic risk in the National Longitudinal Study of Adolescent to Adult Health. This framework accounts for the dynamics of time-varying confounding and mediation that is required in operationalizing a "race" variable as part of a social process (racism) rather than a separable, individual characteristic. We decompose the observed disparity into three types of effects: a controlled direct effect ("unobserved racism"), proportions attributable to interaction ("racial discrimination"), and pure indirect effects ("emergent discrimination"). We discuss the limitations of counterfactual approaches while highlighting how they can be combined with critical theories to quantify how interlocking systems produce racial health inequities.

Using Metrics of a Mixture Effect and Nutrition from an Observational Study for Consideration towards Causal Inference

Environmental exposures to a myriad of chemicals are associated with adverse health effects in humans, while good nutrition is associated with improved health. Single chemical in vivo and in vitro studies demonstrate causal links between the chemicals and outcomes, but such studies do not represent human exposure to environmental mixtures. One way of summarizing the effect of the joint action of chemical mixtures is through an empirically weighted index using weighted quantile sum (WQS) regression. My Nutrition Index (MNI) is a metric of overall dietary nutrition based on guideline values, including for pregnant women. Our objective is to demonstrate the use of an index as a metric for more causally linking human exposure to health outcomes using observational data. We use both a WQS index of 26 endocrine-disrupting chemicals (EDCs) and MNI using data from the SELMA pregnancy cohort to conduct causal inference using g-computation with counterfactuals for assumed either reduced prenatal EDC exposures or improved prenatal nutrition. Reducing the EDC exposure using the WQS index as a metric or improving dietary nutrition using MNI as a metric, the counterfactuals in a causal inference with one SD change indicate significant improvement in cognitive function. Evaluation of such a strategy may support decision makers for risk management of EDCs and individual choices for improving dietary nutrition.

The Role of the Natural Course in Causal Analysis

The average causal effect compares counterfactual outcomes if everyone had been exposed versus if everyone had been unexposed, which can be an unrealistic contrast. Alternatively, we can target effects that compare counterfactual outcomes against the factual outcomes observed in the sample (i.e., we can compare against the natural course). Here, we demonstrate how the natural course can be estimated and used in causal analyses for model validation and effect estimation. Our example is an analysis assessing the impact of taking aspirin on pregnancy, 26 weeks after randomization, in the Effects of Aspirin in Gestation and Reproduction trial (United States, 2006-2012). To validate our models, we estimated the natural course using g-computation and then compared that against the observed incidence of pregnancy. We observed good agreement between the observed and model-based natural courses. We then estimated an effect that compared the natural course against the scenario in which participants assigned to aspirin always complied. If participants had always complied, there would have been 5.0 (95% confidence interval: 2.2, 7.8) more pregnancies per 100 women than was observed. It is good practice to estimate the natural course for model validation when using parametric models, but whether one should estimate a natural course contrast depends on the underlying research questions.

G-Computation and Agent-Based Modeling for Social Epidemiology: Can Population Interventions Prevent Posttraumatic Stress Disorder?

Agent-based modeling and g-computation can both be used to estimate impacts of intervening on complex systems. We explored each modeling approach within an applied example: interventions to reduce posttraumatic stress disorder (PTSD). We used data from a cohort of 2,282 adults representative of the adult population of the New York City metropolitan area from 2002-2006, of whom 16.3% developed PTSD over their lifetimes. We built 4 models: g-computation, an agent-based model (ABM) with no between-agent interactions, an ABM with violent-interaction dynamics, and an ABM with neighborhood dynamics. Three interventions were tested: 1) reducing violent victimization by 37.2% (real-world reduction); 2) reducing violent victimization by100%; and 3) supplementing the income of 20% of lower-income participants. The g-computation model estimated population-level PTSD risk reductions of 0.12% (95% confidence interval (CI): -0.16, 0.29), 0.28% (95% CI: -0.30, 0.70), and 1.55% (95% CI: 0.40, 2.12), respectively. The ABM with no interactions replicated the findings from g-computation. Introduction of interaction dynamics modestly decreased estimated intervention effects (income-supplement risk reduction dropped to 1.47%), whereas introduction of neighborhood dynamics modestly increased effectiveness (income-supplement risk reduction increased to 1.58%). Compared with g-computation, agent-based modeling permitted deeper exploration of complex systems dynamics at the cost of further assumptions.

Bayesian G-Computation for Estimating Impacts of Interventions on Exposure Mixtures: Demonstration With Metals From Coal-Fired Power Plants and Birth Weight

The importance of studying the health impacts of exposure mixtures is increasingly being recognized, but such research presents many methodological and interpretation difficulties. We used Bayesian g-computation to estimate effects of a simulated public health action on exposure mixtures and birth weights in Milwaukee, Wisconsin, in 2011-2013. We linked data from birth records with census-tract-level air toxics data from the Environmental Protection Agency's National Air Toxics Assessment model. We estimated the difference between observed and expected birth weights that theoretically would have followed a hypothetical intervention to reduce exposure to 6 airborne metals by decommissioning 3 coal-fired power plants in Milwaukee County prior to 2010. Using Bayesian g-computation, we estimated a 68-g (95% credible interval: 25, 135) increase in birth weight following this hypothetical intervention. This example demonstrates the utility of our approach for using observational data to evaluate and contrast possible public health actions. Additionally, Bayesian g-computation offers a flexible strategy for estimating the effects of highly correlated exposures, addressing statistical issues such as variance inflation, and addressing conceptual issues such as the lack of interpretability of independent effects.

Union Burying Ground: Mortality, Mortality Inequities, and Sinking Labor Union Membership in the United States

BACKGROUND

Over the last several decades in the United States, socioeconomic life-expectancy inequities have increased 1-2 years. Declining labor-union density has fueled growing income inequities across classes and exacerbated racial income inequities. Using Panel Study of Income Dynamics (PSID) data, we examined the longitudinal union-mortality relationship and estimated whether declining union density has also exacerbated mortality inequities.

METHODS

Our sample included respondents ages 25-66 to the 1979-2015 PSID with mortality follow-up through age 68 and year 2017. To address healthy-worker bias, we used the parametric g-formula. First, we estimated how a scenario setting all (versus none) of respondents' employed-person-years to union-member employed-person-years would have affected mortality incidence. Next, we examined gender, racial, and educational effect modification. Finally, we estimated how racial and educational mortality inequities would have changed if union-membership prevalence had remained at 1979 (vs. 2015) levels throughout follow-up.

RESULTS

In the full sample (respondents = 23,022, observations = 146,681), the union scenario was associated with lower mortality incidence than the nonunion scenario (RR = 0.90, 95% CI = 0.80, 0.99; RD per 1,000 = -19, 95% CI = -37, -1). This protective association generally held across subgroups, although it was stronger among the more-educated. However, we found little evidence mortality inequities would have lessened if union membership had remained at 1979 levels.

CONCLUSIONS

To our knowledge, this was the first individual-level US-based study with repeated union-membership measurements to analyze the union-mortality relationship. We estimated a protective union-mortality association, but found little evidence declining union density has exacerbated mortality inequities; importantly, we did not incorporate contextual-level effects. See video abstract at, http://links.lww.com/EDE/B839.

Population intervention effects in observational studies to emulate target trial results: reconciling the effects of improved sanitation on child growth

BACKGROUND

Improved sanitation has been associated with improved child growth in observational studies, but multiple randomized trials that delivered improved sanitation found no effect on child growth. We assessed to what extent differences in the effect estimated in the two study designs (the effect of treatment in observational studies and the effect of treatment assignment in trials) could explain the contradictory results.

METHODS

We used parametric g-computation in five prospective studies (n = 21 524) and 59 cross-sectional Demographic and Health Surveys (DHS; n = 158 439). We compared the average treatment effect (ATE) for improved sanitation on mean length-for-age z-score (LAZ) among children aged <2 years to population intervention effects (PIEs), which are the observational analogue of the effect estimated in trials in which some participants are already exposed.

RESULTS

The ATE was >0.15 z-scores, a clinically meaningful difference, in most prospective studies but in <20% of DHS surveys. The PIE was always smaller than the ATE, and the magnitude of difference depended on the baseline prevalence of the improved sanitation. Interventions with suboptimal coverage and interventions delivered in populations with higher mean LAZ had a smaller effect on population-level LAZ.

CONCLUSIONS

Estimates of PIEs corresponding to anticipated trial results were often smaller than clinically meaningful effects. Incongruence between observational associations and null trial results may in part be explained by expected differences between the effects estimated. Using observational ATEs to set expectations for trials may overestimate the impact that sanitation interventions can achieve. PIEs predict realistic effects and should be more routinely estimated.

Does the Union Make Us Strong? Labor-Union Membership, Self-Rated Health, and Mental Illness: A Parametric G-Formula Approach

Union members enjoy better wages and benefits and greater power than nonmembers, which can improve health. However, the longitudinal union-health relationship remains uncertain, partially because of healthy-worker bias, which cannot be addressed without high-quality data and methods that account for exposure-confounder feedback and structural nonpositivity. Applying one such method, the parametric g-formula, to US-based Panel Study of Income Dynamics data, we analyzed the longitudinal relationships between union membership, poor/fair self-rated health (SRH), and moderate mental illness (Kessler 6-item score of ≥5). The SRH analyses included 16,719 respondents followed from 1985-2017, while the mental-illness analyses included 5,813 respondents followed from 2001-2017. Using the parametric g-formula, we contrasted cumulative incidence of the outcomes under 2 scenarios, one in which we set all employed-person-years to union-member employed-person-years (union scenario), and one in which we set no employed-person-years to union-member employed-person-years (nonunion scenario). We also examined whether the contrast varied by sex, sex and race, and sex and education. Overall, the union scenario was not associated with reduced incidence of poor/fair SRH (relative risk = 1.01, 95% confidence interval (CI): 0.95, 1.09; risk difference = 0.01, 95% CI: -0.03, 0.04) or moderate mental illness (relative risk = 1.02, 95% CI: 0.92, 1.12; risk difference = 0.01, 95% CI: -0.04, 0.06) relative to the nonunion scenario. These associations largely did not vary by subgroup.

A practical introduction to Bayesian estimation of causal effects: Parametric and nonparametric approaches

Substantial advances in Bayesian methods for causal inference have been made in recent years. We provide an introduction to Bayesian inference for causal effects for practicing statisticians who have some familiarity with Bayesian models and would like an overview of what it can add to causal estimation in practical settings. In the paper, we demonstrate how priors can induce shrinkage and sparsity in parametric models and be used to perform probabilistic sensitivity analyses around causal assumptions. We provide an overview of nonparametric Bayesian estimation and survey their applications in the causal inference literature. Inference in the point-treatment and time-varying treatment settings are considered. For the latter, we explore both static and dynamic treatment regimes. Throughout, we illustrate implementation using off-the-shelf open source software. We hope to leave the reader with implementation-level knowledge of Bayesian causal inference using both parametric and nonparametric models. All synthetic examples and code used in the paper are publicly available on a companion GitHub repository.

Two-stage g-computation: Evaluating Treatment and Intervention Impacts in Observational Cohorts When Exposure Information Is Partly Missing

Illustrations of the g-computation algorithm to evaluate population average treatment and intervention effects have been predominantly implemented in settings with complete exposure information. Thus, worked examples of approaches to handle missing data in this causal framework are needed to facilitate wider use of these estimators. We illustrate two-stage g-computation estimators that leverage partially observed information on the full study sample and complete exposure information on a subset to estimate causal effects. In a hypothetical cohort of 1,623 human immunodeficiency virus (HIV)-positive women with 30% complete opioid prescription information, we illustrate a two-stage extrapolation g-computation estimator for the average treatment effect of shorter or longer duration opioid prescriptions; we further illustrate two-stage inverse probability weighting and imputation g-computation estimators for the average intervention effect of shortening the duration of prescriptions relative to the status quo. Two-stage g-computation estimators approximated the true risk differences for the population average treatment and intervention effects while g-computation fit to the subset of complete cases was biased. In 10,000 Monte Carlo simulations, two-stage approaches considerably reduced bias and mean squared error and improved the coverage of 95% confidence limits. Although missing data threaten validity and precision, two-stage g-computation designs offer principled approaches to handling missing information.

Association between early antibiotic treatment and clinical outcomes in children hospitalized for asthma exacerbation

BACKGROUND

Professional society guidelines recommend against routine early antibiotic use in the treatment of asthma exacerbation without comorbid bacterial infection. However, high antibiotic prescribing rates have been reported in developed countries.

OBJECTIVE

We sought to assess the effectiveness of this strategy in the routine care of children.

METHODS

Using data on 48,743 children hospitalized for asthma exacerbation with no indication of bacterial infection during the period 2010 to 2018, we conducted a retrospective cohort study to compare clinical outcomes and resource utilization between children who received early antibiotic treatment and those who did not.

RESULTS

Overall, 19,866 children (41%) received early antibiotic treatment. According to the propensity score matching analysis, children with early antibiotic treatment had longer hospital stay (mean difference, 0.21 days; 95% CI, 0.18-0.28), higher hospitalization costs (mean difference, $83.5; 95% CI, 62.9-104.0), and higher risk of probiotic use (risk ratio, 2.01; 95% CI, 1.81-2.23) than children who did not receive early antibiotic therapy. Similar results were found from inverse probability of treatment weighting, g-computation, and instrumental variable methods and sensitivity analyses. The risks of mechanical ventilation and 30-day readmission were similar between the groups or slightly higher in the treated group, depending on the statistical models.

CONCLUSIONS

Antibiotic therapy may be associated with prolonged hospital stay, elevated hospitalization costs, and high risk of probiotic use without improving treatment failure and readmission. Our findings highlight the need for reducing inappropriate antibiotic use among children hospitalized for asthma.

Joint and independent neurotoxic effects of early life exposures to a chemical mixture: A multi-pollutant approach combining ensemble learning and g-computation

BACKGROUND

Exposure to mercury (Hg) is associated with adverse developmental effects. However, Hg occurs with a multitude of chemicals. We assessed the associations of developmental exposure to multiple pollutants with children's neurodevelopment using a novel approach.

METHODS

Hg, polychlorinated biphenyls (PCBs), and perfluoroalkyl substances were measured in maternal and children's blood at 5-years (n=449 and 419). At 7-years, children were administered Boston Naming Test (BNT) and the Strengths and Difficulties Questionnaire (SDQ). We used the G-formula combined with SuperLearner to estimate independent and joint effects of chemicals at both ages. We constructed flexible exposure-response relationships and assessed interactions.

RESULTS

Most chemicals showed negative relationships with BNT scores. An inter-quartile range (IQR) increase in maternal Hg and perfluorooctanoic acid (PFOA) was associated with 0.15 standard deviation [SD] (95% Confidence Interval [CI]: -0.29,-0.03) and 0.14 SD (95%CI: -0.26,-0.05) lower scores in BNT, whereas a joint IQR increase in the mixture of chemicals was associated with 0.48 SD (95%CI: -0.69,-0.25) lower scores in BNT. An IQR increase in PFOA was associated with 0.11 SD (95%CI: 0.02,0.26) higher total SDQ difficulties scores. Maternal ∑PCBs concentrations were associated with lower SDQ scores (β=-0.09 SD; 95%CI: -0.19,0), whereas 5-years ∑PCBs showed a negative association (β=-0.09 SD; 95%CI: -0.21,0). Finally, a joint IQR increase in the mixture was associated with 0.22 SD (95%CI: 0.04,0.4) higher SDQ scores.

CONCLUSIONS

Using a novel statistical approach, we confirmed associations between prenatal mercury exposure and lower cognitive function. The potential developmental effects of PFASs need additional attention.

A Bayesian approach to the g-formula

Epidemiologists often wish to estimate quantities that are easy to communicate and correspond to the results of realistic public health interventions. Methods from causal inference can answer these questions. We adopt the language of potential outcomes under Rubin's original Bayesian framework and show that the parametric g-formula is easily amenable to a Bayesian approach. We show that the frequentist properties of the Bayesian g-formula suggest it improves the accuracy of estimates of causal effects in small samples or when data are sparse. We demonstrate an approach to estimate the effect of environmental tobacco smoke on body mass index among children aged 4-9 years who were enrolled in a longitudinal birth cohort in New York, USA. We provide an algorithm and supply SAS and Stan code that can be adopted to implement this computational approach more generally.

"Complete Streets" and Adult Bicyclist Fatalities: Applying G-Computation to Evaluate an Intervention That Affects the Size of a Population at Risk

"Complete streets" policies require transportation engineers to make provisions for pedestrians, bicyclists, and mass transit users. These policies may make bicycling safer for individual cyclists while increasing the overall number of bicycle fatalities if more people cycle due to improved infrastructure. We merged county-level records of complete streets policies with Fatality Analysis Reporting System counts of cyclist fatalities occurring between January 2000 and December 2015. Because comprehensive county-level estimates of numbers of cyclists were not available, we used bicycle commuter estimates from the American Community Survey and the US Census as a proxy for the cycling population and limited analysis to 183 counties (accounting for over half of the US population) for which cycle commuting estimates were consistently nonzero. We used G-computation to estimate the effect of complete streets policies on overall numbers of cyclist fatalities while also accounting for potential policy effects on the size of the cycling population. Over a period of 16 years, 5,254 cyclists died in these counties, representing 34 fatalities per 100,000 cyclist-years. We estimated that complete streets policies made cycling safer, averting 0.6 fatalities per 100,000 cyclist-years (95% confidence interval: -1.0, -0.3) by encouraging a 2.4% increase in cycling but producing only a 0.7% increase in cyclist fatalities. G-computation is a useful tool for understanding the impact of policy on risk and exposure.

A review of time scale fundamentals in the g-formula and insidious selection bias

PURPOSE OF REVIEW

We review recent examples of data analysis with the g-formula, a powerful tool for analyzing longitudinal data and survival analysis. Specifically, we focus on the common choices of time scale and review inferential issues that may arise.

RECENT FINDINGS

Researchers are increasingly engaged with questions that require time scales subject to left-truncation and right-censoring. The assumptions necessary for allowing right-censoring are well defined in the literature, whereas similar assumptions for left-truncation are not well defined. Policy and biologic considerations sometimes dictate that observational data must be analyzed on time scales that are subject to left-truncation, such as age.

SUMMARY

Further consideration of left-truncation is needed, especially when biologic or policy considerations dictate that age is the relevant time scale of interest. Methodologic development is needed to reduce potential for bias when left-truncation may occur.

Estimating the Impact of Changes to Occupational Standards for Silica Exposure on Lung Cancer Mortality

BACKGROUND

Respiratory exposure to silica is associated with the risk of death owing to malignant and nonmalignant disease. 2.3 million US workers are exposed to silica. Occupational exposure limits for silica are derived from a number of lines of evidence, including observational studies. Observational studies may be subject to healthy worker survivor bias, which could result in underestimates of silica's impact on worker mortality and, in turn, bias risk estimates for occupational exposure limits.

METHODS

Using data on 65,999 workers pooled across multiple industries, we estimate the impacts of several hypothetical occupational exposure limits on silica exposure on lung cancer and all-cause mortality. We use the parametric g-formula, which can account for healthy worker survivor bias.

RESULTS

Assuming we could eliminate occupational exposure, we estimate that there would be 20.7 fewer deaths per 1,000 workers in our pooled study by age 80 (95% confidence interval = 14.5, 26.8), including 3.91 fewer deaths owing to lung cancer (95% CI = 1.53, 6.30). Less restrictive interventions demonstrated smaller but still substantial risk reductions.

CONCLUSIONS

Our results suggest that occupational exposure limits for silica can be further strengthened to reduce silica-associated mortality and illustrate how current risk analysis for occupational limits can be improved.

Analyzing medical costs with time-dependent treatment: The nested g-formula

As medical expenses continue to rise, methods to properly analyze cost outcomes are becoming of increasing relevance when seeking to compare average costs across treatments. Inverse probability weighted regression models have been developed to address the challenge of cost censoring in order to identify intent-to-treat effects (i.e., to compare mean costs between groups on the basis of their initial treatment assignment, irrespective of any subsequent changes to their treatment status). In this paper, we describe a nested g-computation procedure that can be used to compare mean costs between two or more time-varying treatment regimes. We highlight the relative advantages and limitations of this approach when compared with existing regression-based models. We illustrate the utility of this approach as a means to inform public policy by applying it to a simulated data example motivated by costs associated with cancer treatments. Simulations confirm that inference regarding intent-to-treat effects versus the joint causal effects estimated by the nested g-formula can lead to markedly different conclusions regarding differential costs. Therefore, it is essential to prespecify the desired target of inference when choosing between these two frameworks. The nested g-formula should be considered as a useful, complementary tool to existing methods when analyzing cost outcomes.

Living in Violent Neighbourhoods is Associated with Gestational Weight Gain Outside the Recommended Range

BACKGROUND

During pregnancy, most women do not meet gestational weight gain (GWG) guidelines, potentially resulting in adverse maternal and infant health consequences. Social environment determinants of GWG have been identified, but evidence on the relationship between neighbourhood violence and GWG is scant. Our study aims to examine the relationship between neighbourhood violence and GWG outside the recommended range.

METHODS

We used statewide vital statistics and health care utilization data from California for 2006-12 (n = 2 364 793) to examine the relationship of neighbourhood violence (quarters of zip-code rates of homicide and assault) in the first 37 weeks of pregnancy with GWG (categorized using the Institute of Medicine's pregnancy weight gain guidelines). We estimated risk ratios (RR) and marginal risk differences, and analyses were stratified by maternal race/ethnicity and prepregnancy body mass index.

RESULTS

Residence in neighbourhoods with the highest quartile of violence was associated with more excessive GWG (adjusted RR 1.04, 95% confidence interval CI 1.03, 1.05), compared to the lowest quartile of violence; violence was not associated with inadequate GWG. On the difference scale, this association translates to 2.3% more women gaining weight excessively rather than adequately if all women were exposed to high violence compared to if all women were exposed to low violence. Additionally, associations between neighbourhood violence and excessive GWG were larger in non-white women than in white women.

CONCLUSIONS

These findings support the hypothesis that violence can affect weight gain during pregnancy, emphasizing the importance of neighbourhood violence as a public health issue.

Direct estimation for adaptive treatment length policies: Methods and application to evaluating the effect of delayed PEG insertion

Dysphagia is a primary cause of death among patients diagnosed with amyotrophic lateral sclerosis (ALS), and percutaneous endoscopic gastrostomy (PEG) is a procedure to insert a tube into the stomach to assist or replace oral feeding. It is believed that PEG is beneficial and, generally, earlier insertion is preferable to later. However, gathering clinical evidence to support these beliefs on the use and timing of PEG is challenging because controlled clinical trials are not feasible and clinical endpoints are confounded with PEG in observational data. Moreover, the confounders are time-varying and time to PEG insertion may be only partially observed. We show how one can view this problem as an adaptive treatment length policy and propose a new estimator via g-computation. We show that our estimator is consistent and asymptotically normal for the causal estimand and explore its finite sample properties in simulation studies. Finally, using more than 10 years of data from Emory ALS clinic registry, we found no evidence to suggest that earlier PEG reduced 4-year mortality; thus, our results do not support the hypothesis and belief that initiating palliative care earlier extends life, on average. At the same, we cannot be certain that all important confounding variables are collected and observed to ensure our modeling assumptions are correct, so more work is needed to address these important end-of-life questions for ALS patients.