Babette A. Brumback

Marginal structural models and causal inference in epidemiology

In observational studies with exposures or treatments that vary over time, standard approaches for adjustment of confounding are biased when there exist time-dependent confounders that are also affected by previous treatment. This paper introduces marginal structural models, a new class of causal models that allow for improved adjustment of confounding in those situations. The parameters of a marginal structural model can be consistently estimated using a new class of estimators, the inverse-probability-of-treatment weighted estimators.

Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men.

Standard methods for survival analysis, such as the time-dependent Cox model, may produce biased effect estimates when there exist time-dependent confounders that are themselves affected by previous treatment or exposure. Marginal structural models are a new class of causal models the parameters of which are estimated through inverse-probability-of-treatment weighting; these models allow for appropriate adjustment for confounding. We describe the marginal structural Cox proportional hazards model and use it to estimate the causal effect of zidovudine on the survival of human immunodeficiency virus-positive men participating in the Multicenter AIDS Cohort Study. In this study, CD4 lymphocyte count is both a time-dependent confounder of the causal effect of zidovudine on survival and is affected by past zidovudine treatment. The crude mortality rate ratio (95% confidence interval) for zidovudine was 3.6 (3.0-4.3), which reflects the presence of confounding. After controlling for baseline CD4 count and other baseline covariates using standard methods, the mortality rate ratio decreased to 2.3 (1.9-2.8). Using a marginal structural Cox model to control further for time-dependent confounding due to CD4 count and other time-dependent covariates, the mortality rate ratio was 0.7 (95% conservative confidence interval = 0.6-1.0). We compare marginal structural models with previously proposed causal methods.

Smoothing Spline Models for the Analysis of Nested and Crossed Samples of Curves

Abstract We introduce a class of models for an additive decomposition of groups of curves stratified by crossed and nested factors, generalizing smoothing splines to such samples by associating them with a corresponding mixed-effects model. The models are also useful for imputation of missing data and exploratory analysis of variance. We prove that the best linear unbiased predictors (BLUPs) from the extended mixed-effects model correspond to solutions of a generalized penalized regression where smoothing parameters are directly related to variance components, and we show that these solutions are natural cubic splines. The model parameters are estimated using a highly efficient implementation of the EM algorithm for restricted maximum likelihood (REML) estimation based on a preliminary eigenvector decomposition. Variability of computed estimates can be assessed with asymptotic techniques or with a novel hierarchical bootstrap resampling scheme for nested mixed-effects models. Our methods are applied to me...

Marginal Structural Models to Estimate the Joint Causal Effect of Nonrandomized Treatments

Even in the absence of unmeasured confounding factors or model misspecification, standard methods for estimating the causal effect of time-varying treatments on survival are biased when (a) there exists a time-dependent risk factor for survival that also predicts subsequent treatment, and (b) past treatment history predicts subsequent risk factor level. In contrast, methods based on marginal structural models (MSMs) can provide consistent estimates of causal effects when unmeasured confounding and model misspecification are absent. MSMs are a new class of causal models whose parameters are estimated using a new class of estimators—inverse-probability-of-treatment weighted estimators. We use a marginal structural Cox proportional hazards model to estimate the joint effect of zidovudine (AZT) and prophylaxis therapy for Pneumocystis carinii pneumonia on the survival of HIV-positive men in the Multicenter AIDS Cohort Study, an observational study of homosexual men. We obtained an estimated causal mortality rate (hazard) ratio of .67 (conservative 95% confidence interval .46-.98) for AZT and of 1.14 (.79, 1.64) for prophylaxis therapy. These estimates will be consistent for the true causal rate ratios when the functional forms chosen for our models are correct and data have been obtained on all time-independent and time-dependent covariates that predict both subsequent treatment and mortality.

Robust principal component analysis for functional data

A method for exploring the structure of populations of complex objects, such as images, is considered. The objects are summarized by feature vectors. The statistical backbone is Principal Component Analysis in the space of feature vectors. Visual insights come from representing the results in the original data space. In an ophthalmological example, endemic outliers motivate the development of a bounded influence approach to PCA.

Comparison of Parent-Only vs Family-Based Interventions for Overweight Children in Underserved Rural Settings: Outcomes From Project STORY

OBJECTIVE To assess the effectiveness of parent-only vs family-based interventions for pediatric weight management in underserved rural settings. DESIGN A 3-arm randomized controlled clinical trial. SETTING All sessions were conducted at Cooperative Extension Service offices in underserved rural counties. PARTICIPANTS Ninety-three overweight or obese children (8-14 years old) and their parent(s). INTERVENTION Families were randomized to (1) a behavioral family-based intervention, (2) a behavioral parent-only intervention, or (3) a wait-list control group. OUTCOME MEASURE The primary outcome measure was change in childrens standardized body mass index (BMI). RESULTS Seventy-one children completed posttreatment (month 4) and follow-up (month 10) assessments. At the month 4 assessment, children in the parent-only intervention demonstrated a greater decrease in BMI z score (mean difference [MD], 0.127; 95% confidence interval [CI], 0.027 to 0.226) than children in the control condition. No significant difference was found between the family-based intervention and the control condition (MD, 0.065; 95% CI, -0.027 to 0.158). At month 10 follow-up, children in the parent-only and family-based intervention groups demonstrated greater decreases in BMI z score from before treatment compared with those in the control group (MD, 0.115; 95% CI, 0.003 to 0.220; and MD, 0.136; 95% CI, 0.018 to 0.254, respectively). No difference was found in weight status change between the parent-only and family-based interventions at either assessment. CONCLUSIONS A parent-only intervention may be a viable and effective alternative to family-based treatment of childhood overweight. Cooperative Extension Service offices have the potential to serve as effective venues for the dissemination of obesity-related health promotion programs.

Transitional Regression Models, with Application to Environmental Time Series

Abstract Environmental epidemiologists often encounter time series data in the form of discrete or other nonnormal outcomes; for example, in modeling the relationship between air pollution and hospital admissions or mortality rates. We present a case study examining the association between pollen counts and meteorologic covariates. Although such time series data are inadequately described by standard methods for Gaussian time series, they are often autocorrelated, and warrant an analysis beyond those provided by ordinary generalized linear models (GLMs). Transitional regression models (TRMs), signifying nonlinear regression models expressed in terms of conditional means and variances given past observations, provide a unifying framework for two mainstream approaches to extending the GLM for autocorrelated data. The first approach models current outcomes with a GLM that incorporates past outcomes as covariates, whereas the second models individual outcomes with marginal GLMs and then couples the error terms with an autoregressive covariance matrix. Although the two approaches coincide for the Gaussian GLM, which serves as a helpful introductory example, in general they yield fundamentally different models. We analyze the pollen study using TRMs of both types and present parameter estimates together with asymptotic and bootstrap standard errors. In several cases we find evidence of residual autocorrelation; however, when we relax the TRM to allow for a nonparametric smooth trend, the autocorrelation disappears. This kind of trade-off between autocorrelation and flexibility is to be expected, and has a natural interpretation in terms of the covariance function for a nonparametric smoother. We provide an algorithm for fitting these flexible TRMs that is relatively easy to program with the generalized additive model software in S-PLUS.

Assessing the impact of a school‐based water treatment, hygiene and sanitation programme on pupil absence in Nyanza Province, Kenya: a cluster‐randomized trial

Objectives  There has been increased attention to access to water, sanitation and hygiene (WASH) at schools in developing countries, but a dearth of empirical studies on the impact. We conducted a cluster‐randomized trial of school‐based WASH on pupil absence in Nyanza Province, Kenya, from 2007 to 2008.

Methylation of CASP8, DCR2, and HIN-1 in Neuroblastoma Is Associated with Poor Outcome

Purpose: Epigenetic aberrations have been shown to play an important role in the pathogenesis of most cancers. To investigate the clinical significance of epigenetic changes in neuroblastoma, we evaluated the relationship between clinicopathologic variables and the pattern of gene methylation in neuroblastoma cell lines and tumors. Experimental Design: Methylation-specific PCR was used to evaluate the gene methylation status of 19 genes in 14 neuroblastoma cell lines and 8 genes in 70 primary neuroblastoma tumors. Associations between gene methylation, established prognostic factors, and outcome were evaluated. Log-rank tests were used to identify the number of methylated genes that was most predictive of overall survival. Results: Epigenetic changes were detected in the neuroblastoma cell lines and primary tumors, although the pattern of methylation varied. Eight of the 19 genes analyzed were methylated in >70% of the cell lines. Epigenetic changes of four genes were detected in only small numbers of cell lines. None of the cell lines had methylation of the other seven genes analyzed. In primary neuroblastoma tumors, high-risk disease and poor outcome were associated with methylation of DCR2, CASP8, and HIN-1 individually. Although methylation of the other five individual genes was not predictive of poor outcome, a trend toward decreased survival was seen in patients with a methylation phenotype, defined as ≥4 methylated genes (P = 0.055). Conclusion: Our study indicates that clinically aggressive neuroblastoma tumors have aberrant methylation of multiple genes and provides a rationale for exploring treatment strategies that include demethylating agents.

Human Leukocyte Antigen Class II and Cervical Cancer Risk: A Population-Based Study

The critical role of the human leukocyte antigen (HLA) system in presenting peptides to antigen-specific T cell receptors may explain why only some human papillomavirus (HPV)-infected women progress to cervical cancer. HLA class II DRB1 and DQB1 genes were examined in 315 women with invasive squamous cell cervical cancer (SCC) and 381 control subjects. Increased risks of SCC were associated with DRB1*1001, DRB1*1101, and DQB1*0301, and decreased risks were associated with DRB1*0301 and DRB1*13. Of squamous cell tumors, those containing HPV-16 were different from those not containing HPV-16 for 3 alleles: DRB1*0401, DRB1*07, and DQB1*06. Increased risks of SCC were associated with DRB1*0401-DQB1*0301 (odds ratio [OR], 1.7; 95% confidence interval [CI], 1.1-2.7) and DRB1*1101-DQB1*0301 (OR, 2.5; 95% CI, 1.4-4.5), and decreased risks were associated with DRB1*0301-DQB1*02 (OR, 0.7; 95% CI, 0.5-1.0) and DRB1*13-DQB1*06 (OR, 0.6; 95% CI, 0.4-0.9) haplotypes. These results add to the evidence that certain HLA class II alleles or allele combinations, or genes linked to them, make some women more susceptible to SCC.