Many statistical methods have been developed that treat within-subject correlation that accompanies the clustering of subjects in longitudinal data settings as a nuisance parameter, with the focus of analytic interest being on mean outcome or profiles over time. However, there is evidence that in certain settings (Elliott 2007; Harlow et al. 2000; Sammel et al. 2001
Kikuya et al. 2008) underlying variability in subject measures may also be important in predicting future health outcomes of interest. Here we develop a method for combining information from mean profiles and residual variance to assess associations with categorical outcomes in a joint modeling framework. We consider an application to relating word recall measures obtained over time to dementia onset from the Health and Retirement Survey.
Differential measurement error; Markov Chain Monte Carlo; Total recall; Dementia; Health and Retirement Survey
Apoptosis and the subsequent clearance of these dying cells occur throughout development and adult life in many tissues. Failure to promptly clear apoptotic cells has been linked to many diseases1-3. ELMO1 is an evolutionarily conserved cytoplasmic engulfment protein that functions downstream of the phosphatidylserine receptor BAI1, and, along with Dock180 and Rac1, promotes internalization of the dying cells4-7. Here, we generated ELMO1-deficient mice, and unexpectedly found them to be viable and grossly normal. However, ELMO1-deficient mice had a striking testicular pathology, with disrupted seminiferous epithelium, multi-nucleated giant cells, uncleared apoptotic germ cells, and decreased sperm output. Subsequent in vitro and in vivo analyses revealed a crucial role for ELMO1 in the phagocytic clearance of apoptotic germ cells by Sertoli cells lining the seminiferous epithelium. The engulfment receptor BAI1 and the GTPase Rac (upstream and downstream of ELMO1, respectively) were also important for Sertoli cell-mediated engulfment. Collectively, these findings uncover a selective requirement for ELMO1 in Sertoli cell-mediated removal of apoptotic germ cells and make a compelling case for a relationship between engulfment and tissue homeostasis in vivo.
Neighborhood disadvantage has consistently been linked to increased rates of morbidity and mortality, but the mechanisms through which neighborhood environments may get “under the skin” remain largely unknown. Differential exposure to chronic environmental stressors has been identified as a potential pathway linking neighborhood disadvantage and poor health, particularly through the dysregulation of stress-related biological pathways such as cortisol secretion, but the majority of existing observational studies on stress and neuroendocrine functioning have focused exclusively on individual-level stressors and psychosocial characteristics. This paper aims to fill that gap by examining the association between features of the neighborhood environment and the diurnal cortisol patterns of 308 individuals from Chicago, Illinois, USA. We found that respondents in neighborhoods with high levels of perceived and observed stressors or low levels of social support experienced a flatter rate of cortisol decline throughout the day. In addition, overall mean cortisol levels were found to be lower in higher stress, lower support neighborhoods. This study adds to the growing evidence of hypocortisolism among chronically stressed adult populations and suggests hypocortisolism rather than hypercortisolism as a potential mechanism linking social disadvantage to poor health.
USA; cortisol; neighborhood effects; health inequalities; multi-level modeling; stress
Changes in women’s menstrual bleeding patterns precede the onset of menopause. In this paper, the authors identify population subgroups based on menstrual characteristics of the menopausal transition experience. Using the TREMIN data set (1943–1979), the authors apply a Bayesian change-point model with 8 parameters for each woman that summarize change in menstrual bleeding patterns during the menopausal transition. The authors then use estimates from this model to classify menstrual patterns into subgroups using a K-medoids algorithm. They identify 6 subgroups of women whose transition experience can be distinguished by age at onset, variability of the menstrual cycle, and duration of the early transition. These results suggest that for most women, mean and variance change points are well aligned with proposed bleeding markers of the menopausal transition, but for some women they are not clearly associated. Increasing understanding of population differences in the transition experience may lead to new insights into ovarian aging. Because of age inclusion criteria, most longitudinal studies of the menopausal transition probably include only a subset of the 6 subgroups of women identified in this paper, suggesting a potential bias in the understanding of both the menopausal transition and the linkage between the transition and chronic disease.
menopause; menstrual cycle; ovary; women’s health
This study compared the hypertension prevalence, awareness, treatment and control in Chicago, Illinois and Detroit, Michigan to that of the general United States population (aged ≥ 25 years) for the period 2001–2003. We examined whether and how much 1) urban populations have less favorable hypertension-related outcomes and 2) the rates of racial/ethnic minorities lag behind those of Whites in order to determine if the national data understate the magnitude of hypertension-related outcomes and racial/ethnic disparities in two large cities in the Midwestern region of the United States and perhaps others.
Unstandardized and standardized hypertension-related outcome rates were estimated.
The hypertension-related outcomes among Chicago and Detroit residents lag behind the United States by 8%–14% and 10%–18% points, respectively. Additionally, this study highlights the complexity of the racial/ethnic differences in hypertension-related outcomes, where within each population, Blacks were more likely to have hypertension and to be aware of their hypertension status than Whites, and no less likely to be treated. Conversely, Hispanics were less likely to have hypertension and also less likely to be aware of their status when they do have hypertension when compared to Whites.
At a time when efficacious treatment for hypertension has been available for more than 50 years, continued racial/ethnic differences in the prevalence, awareness, treatment and control of hypertension is among public health’s greatest challenges. To achieve the proposed national hypertension-related goals, future policies must consider the social context of hypertension within central cities of urban areas. (Ethn Dis. 2012;22:391–397)
Hypertension; Minority Health; Population; Urban Health
In sample surveys where units have unequal probabilities of inclusion, associations between the inclusion probability and the statistic of interest can induce bias in unweighted estimates. This is true even in regression models, where the estimates of the population slope may be biased if the underlying mean model is misspecified or the sampling is nonignorable. Weights equal to the inverse of the probability of inclusion are often used to counteract this bias. Highly disproportional sample designs have highly variable weights; weight trimming reduces large weights to a maximum value, reducing variability but introducing bias. Most standard approaches are ad hoc in that they do not use the data to optimize bias-variance trade-offs. This article uses Bayesian model averaging to create “data driven” weight trimming estimators. We extend previous results for linear regression models (Elliott 2008) to generalized linear regression models, developing robust models that approximate fully-weighted estimators when bias correction is of greatest importance, and approximate unweighted estimators when variance reduction is critical.
Sample survey; sampling weights; weight winsorization; Bayesian population inference; weight pooling; variable selection; fractional Bayes Factors
Rapid and efficient removal of apoptotic cells by phagocytes plays a key role during development, tissue homeostasis, and in controlling immune responses1–5. An important feature of efficient clearance is the capacity of a single phagocyte to ingest multiple apoptotic cells successively, and to process the increased load of corpse-derived cellular material6–9. However, factors that influence sustained phagocytic capacity or how they in turn influence continued clearance by phagocytes are not known. Here we identify that the ability of a phagocyte to control its mitochondrial membrane potential is a critical factor in the capacity of a phagocyte to engulf apoptotic cells. Changing the phagocyte mitochondrial membrane potential (genetically or pharmacologically) significantly affected phagocytosis, with lower potential enhancing engulfment and higher membrane potential inhibiting uptake. We then identified that Ucp2, a mitochondrial membrane protein that acts to lower the mitochondrial membrane potential10–12, is upregulated in phagocytes engulfing apoptotic cells (but not synthetic targets, bacteria, or yeast). Loss of Ucp2 limited the capacity of phagocytes to continually ingest apoptotic cells, while overexpression of Ucp2 increased the capacity for engulfment and the ability to engulf multiple apoptotic cells. Mutational and pharmacological inhibition of Ucp2 uncoupling activity reversed the positive effect of Ucp2 on engulfment capacity, suggesting a direct role for Ucp2-mediated mitochondrial function in phagocytosis. Macrophages from Ucp2-deficient mice13, 14 were impaired in their capacity to engulf apoptotic cells in vitro, and Ucp2-deficient mice displayed profound in vivo defects in clearing dying cells in the thymus and the testes. Collectively, these data suggest that phagocytes alter the mitochondrial membrane potential during engulfment to regulate uptake of sequential apoptotic cells, and that Ucp2 is a key molecular determinant of this step in vivo. Since Ucp2 function has also been linked to metabolic diseases and atherosclerosis14–16, these data identifying a new role for Ucp2 in regulating apoptotic cell clearance may provide additional insights toward understanding the complex etiology and pathogenesis of these diseases.
Recent advances in defining the molecular signaling pathways that regulate the phagocytosis of apoptotic cells have improved our understanding of this complex and evolutionarily conserved process. Studies in mice and humans suggest that the prompt removal of dying cells is crucial for immune tolerance and tissue homeostasis. Failed or defective clearance has emerged as an important contributing factor to a range of disease processes. This review addresses how specific molecular alterations of engulfment pathways are linked to pathogenic states. A better understanding of the apoptotic cell clearance process in healthy and diseased states could offer new therapeutic strategies.
The aim of this study was to define the functional role of a recently identified RahU protein from Pseudomonas aeruginosa in macrophages and its role in bacterial defense. Recombinant (r)-RahU had no significant effect on cell apoptosis or cell viability in human monocytic THP-1 cells. Gene expression array of murine macrophage cells (RAW 264.7) stimulated with LPS showed modulation of common transcripts involved in inflammation. Functional cellular analysis showed RAW cells incubated with r-RahU at 1.0–10 µg/ml (0.06–0.6 µM) inhibited accumulation of nitric oxide (NO) in the presence of LPS by 10 to 50%. The IC50 of r-RahU (0.6 µM) was distinct from the known inhibitors of NO production: prednisone (50 µM) and L-NMMA (100 µM). rRahU also significantly inhibited chemotactic activity of THP-1 cells toward CCL2 or chemotactic supernatants from apoptotic T-cells. These reports show previously unknown pleiotropic properties of RahU in modulating both microbial physiology and host innate immunity.
Innate immunity; Nitric oxide; Chemotaxis; Inflammation; Inflammation resolution
When the true end points (T) are difficult or costly to measure, surrogate markers (S) are often collected in clinical trials to help predict the effect of the treatment (Z). There is great interest in understanding the relationship among S, T, and Z. A principal stratification (PS) framework has been proposed by Frangakis and Rubin (2002) to study their causal associations. In this paper, we extend the framework to a multiple trial setting and propose a Bayesian hierarchical PS model to assess surrogacy. We apply the method to data from a large collection of colon cancer trials in which S and T are binary. We obtain the trial-specific causal measures among S, T, and Z, as well as their overall population-level counterparts that are invariant across trials. The method allows for information sharing across trials and reduces the nonidentifiability problem. We examine the frequentist properties of our model estimates and the impact of the monotonicity assumption using simulations. We also illustrate the challenges in evaluating surrogacy in the counterfactual framework that result from nonidentifiability.
Bayesian estimation; Counterfactual model; Identifiability; Multiple trials; Principal stratification; Surrogate marker
Whereas thousands of new neurons are generated daily during adult life, only a fraction of them survive and become part of neural circuits; the rest die, and their corpses are presumably cleared by resident phagocytes. How the dying neurons are removed and how such clearance influences neurogenesis are not well understood. Here, we identify an unexpected phagocytic role for the doublecortin (DCX)-positive neuronal progenitor cells during adult neurogenesis. Our in vivo and ex vivo studies demonstrate that DCX+ cells comprise a significant phagocytic population within the neurogenic zones. Intracellular engulfment protein ELMO1, which promotes Rac activation downstream of phagocytic receptors, was required for phagocytosis by DCX+ cells. Disruption of engulfment in vivo genetically (in Elmo1-null mice) or pharmacologically (in wild-type mice) led to reduced uptake by DCX+ cells, accumulation of apoptotic nuclei in the neurogenic niches and impaired neurogenesis. Collectively, these findings indicate a paradigm wherein DCX+ neuronal precursors also serve as phagocytes, and that their phagocytic activity critically contributes to neurogenesis in the adult brain.
A surrogate marker (S) is a variable that can be measured earlier and often easier than the true endpoint (T) in a clinical trial. Most previous research has been devoted to developing surrogacy measures to quantify how well S can replace T or examining the use of S in predicting the effect of a treatment (Z). However, the research often requires one to fit models for the distribution of T given S and Z. It is well known that such models do not have causal interpretations because the models condition on a post-randomization variable S. In this paper, we directly model the relationship among T, S and Z using a potential outcomes framework introduced by Frangakis and Rubin (2002). We propose a Bayesian estimation method to evaluate the causal probabilities associated with the cross-classification of the potential outcomes of S and T when S and T are both binary. We use a log-linear model to directly model the association between the potential outcomes of S and T through the odds ratios. The quantities derived from this approach always have causal interpretations. However, this causal model is not identifiable from the data without additional assumptions. To reduce the non-identifiability problem and increase the precision of statistical inferences, we assume monotonicity and incorporate prior belief that is plausible in the surrogate context by using prior distributions. We also explore the relationship among the surrogacy measures based on traditional models and this counterfactual model. The method is applied to the data from a glaucoma treatment study.
Bayesian Estimation; Counterfactual Model; Randomized Trial; Surrogate Marker
In longitudinal studies of developmental and disease processes, participants are followed prospectively with intermediate milestones identified as they occur. Frequently, studies enroll participants over a range of ages including ages at which some participants’ milestones have already passed. Ages at milestones that occur prior to study entry are left censored if individuals are enrolled in the study or left truncated if they are not. The authors examined the bias incurred by ignoring these issues when estimating the distribution of age at milestones or the time between 2 milestones. Methods that account for left truncation and censoring are considered. Data on the menopausal transition are used to illustrate the problem. Simulations show that bias can be substantial and that standard errors can be severely underestimated in naïve analyses that ignore left truncation. Bias can be reduced when analyses account for left truncation, although the results are unstable when the fraction truncated is high. Simulations suggest that a better solution, when possible, is to modify the study design so that information on current status (i.e., whether or not a milestone has passed) is collected on all potential participants, analyzing those who are past the milestone at the time of recruitment as left censored rather than excluding such individuals from the analysis.
bias (epidemiology); censoring; epidemiologic methods; longitudinal studies; study design; truncation
Statistical and cost efficiency can be achieved in population-based samples through stratification and/or clustering. Strata typically combine subgroups of the population that are similar with respect to an outcome. Clusters are often taken from pre-existing units, but may be formed to minimize between-cluster variance, or to equalize exposure to a treatment or risk factor. Area probability sample design procedures for the National Children’s Study required contiguous strata and clusters that maximized within-stratum and within-cluster homogeneity while maintaining approximately equal size of the strata or clusters. However, there were few methods that allowed such strata or clusters to be constructed under these contiguity and equal size constraints.
A search algorithm generates equal-size cluster sets that approximately span the space of all possible clusters of equal size. An optimal cluster set is chosen based on analysis of variance and convexity criteria.
The proposed algorithm is used to construct 10 strata based on demographics and air pollution measures in Kent County, MI, following census tract boundaries. A brief simulation study is also conducted.
The proposed algorithm is effective at uncovering underlying clusters from noisy data. It can be used in multi-stage sampling where equal-size strata or clusters are desired.
Sample Design; Stratification; Clustering; epsem; National Children’s Study
There has been substantive interest in the assessment of surrogate endpoints in medical research. These are measures which could potentially replace “true” endpoints in clinical trials and lead to studies that require less follow-up. Recent research in the area has focused on assessments using causal inference frameworks. Beginning with a simple model for associating the surrogate and true endpoints in the population, we approach the problem as one of endogenous covariates. An instrumental variables estimator and general two-stage algorithm is proposed. Existing surrogacy frameworks are then evaluated in the context of the model. In addition, we define an extended relative effect estimator as well as a sensitivity analysis for assessing what we term the treatment instrumentality assumption. A numerical example is used to illustrate the methodology.
Clinical Trial; Counterfactual; Nonlinear response; Prentice Criterion; Structural equations model
Macrophages change their phenotype and biological functions depending on the microenvironment. In atherosclerosis, oxidative tissue damage accompanies chronic inflammation, however, macrophage phenotypic changes in response to oxidatively modified molecules are not known.
To examine macrophage phenotypic changes in response to oxidized phospholipids that are present in atherosclerotic lesions.
Methods and Results
We show that oxidized phospholipid-treated macrophages develop into a novel phenotype (Mox), strikingly different from the conventional M1 and M2 macrophage phenotypes. Compared to M1 and M2, Mox show a different gene expression pattern, as well as decreased phagocytotic and chemotactic capacity. Treatment with oxidized phospholipids induces both M1 and M2 macrophages to switch to the Mox phenotype. Whole genome expression array analysis and subsequent gene ontology clustering revealed that the Mox phenotype was characterized by abundant over-representation of Nrf2-mediated expression of redox-regulatory genes. In macrophages isolated from Nrf2−/− mice, oxidized phospholipid-induced gene expression and regulation of redox status were compromised. Moreover, we found that Mox macrophages comprise 30% of all macrophages cells in advanced atherosclerotic lesions of LDL receptor knockout mice.
Together, we identify Nrf2 as a key regulator in the formation of a novel macrophage phenotype (Mox) that develops in response to oxidative tissue damage. The unique biological properties of Mox suggest this phenotype may play an important role in atherosclerotic lesion development as well as in other settings of chronic inflammation.
Macrophages; Oxidized Phospholipids; Atherosclerosis; Nrf2
Apoptotic cells release ‘find-me’ signals at the earliest stages of death to recruit phagocytes1. The nucleotides ATP and UTP represent one class of find-me signals2, but their mechanism of release is not known. Here, we identify the plasma membrane channel pannexin 1 (PANX1) as a mediator of find-me signal/nucleotide release from apoptotic cells. Pharmacological inhibition and siRNA-mediated knockdown of PANX1 led to decreased nucleotide release and monocyte recruitment by apoptotic cells. Conversely, PANX1 over-expression enhanced nucleotide release from apoptotic cells and phagocyte recruitment. Patch-clamp recordings showed that PANX1 was basally inactive, and that induction of PANX1 currents occurred only during apoptosis. Mechanistically, PANX1 itself was a target of effector caspases (caspases 3 and 7), and a specific caspase-cleavage site within PANX1 was essential for PANX1 function during apoptosis. Expression of truncated PANX1 (at the putative caspase cleavage site) resulted in a constitutively open channel. PANX1 was also important for the ‘selective’ plasma membrane permeability of early apoptotic cells to specific dyes3. Collectively, these data identify PANX1 as a plasma membrane channel mediating the regulated release of find-me signals and selective plasma membrane permeability during apoptosis, and a new mechanism of PANX1 activation by caspases.
Most investigations in the social and health sciences aim to understand the directional or causal relationship between a treatment or risk factor and outcome. Given the multitude of pathways through which the treatment or risk factor may affect the outcome, there is also an interest in decomposing the effect of a treatment of risk factor into “direct” and “mediated” effects. For example, child's socioeconomic status (risk factor) may have a direct effect on the risk of death (outcome) and an effect that may be mediated through the adulthood socioeconomic status (mediator). Building on the potential outcome framework for causal inference, we develop a Bayesian approach for estimating direct and mediated effects in the context of a dichotomous mediator and dichotomous outcome, which is challenging as many parameters cannot be fully identified. We first define principal strata corresponding to the joint distribution of the observed and counterfactual values of the mediator, and define associate, dissociative, and mediated effects as functions of the differences in the mean outcome under differing treatment assignments within the principal strata. We then develop the likelihood properties and calculate nonparametric bounds of these causal effects assuming randomized treatment assignment. Because likelihood theory is not well developed for nonidentifiable parameters, we consider a Bayesian approach that allows the direct and mediated effects to be expressed in terms of the posterior distribution of the population parameters of interest. This range can be reduced by making further assumptions about the parameters that can be encoded in prior distribution assumptions. We perform sensitivity analyses by using several prior distributions that make weaker assumptions than monotonicity or the exclusion restriction. We consider an application that explores the mediating effects of adult poverty on the relationship between childhood poverty and risk of death.
Direct effect; Mediated effect; Monotonicity; Mortality; Poverty
Engulfment and prompt removal of apoptotic cells occurs from embryogenesis throughout the lifespan of multi-cellular organisms. A new player, Pallbearer, has recently been identified in Drosophila as being important for efficient engulfment by macrophages. Pallbearer is a component of the SCF E3 ubiquitin ligase complex involved in ubiquitylation of proteins targeted for proteasomal degradation. This work provides the first link between the cellular processes of ubiquitylation/proteasomal degradation and the ability to efficiently clear apoptotic cells.
During embryonic development large numbers of apoptotic cells are generated and subsequently removed rapidly and efficiently by phagocytes. In this issue of Cell, Kurant et al. describe a novel Drosophila transmembrane protein, Six Microns Under (SIMU), expressed on the surface of glial cells and macrophages that recognizes apoptotic neurons and promotes phagocytosis upstream of another phagocytic receptor Draper in the developing CNS.
We conducted a population-based case–control study to better delineate the relationship between individual alcohol consumption, alcohol outlets in the surrounding environment, and being assaulted with a gun.
An incidence density sampled case–control study was conducted in the entire city of Philadelphia from 2003 to 2006. We enrolled 677 cases that had been shot in an assault and 684 population-based controls. The relationships between 2 independent variables of interest, alcohol consumption and alcohol outlet availability, and the outcome of being assaulted with a gun were analyzed. Conditional logistic regression was used to adjust for numerous confounding variables.
After adjustment, heavy drinkers were 2.67 times as likely to be shot in an assault when compared with nondrinkers (p < 0.10) while light drinkers were not at significantly greater risk of being shot in an assault when compared with nondrinkers. Regression-adjusted analyses also demonstrated that being in an area of high off-premise alcohol outlet availability significantly increased the risk of being shot in an assault by 2.00 times (p < 0.05). Being in an area of high on-premise alcohol outlet availability did not significantly change this risk. Heavy drinkers in areas of high off-premise alcohol outlet availability were 9.34 times (p < 0.05) as likely to be shot in an assault.
This study finds that the gun assault risk to individuals who are near off-premise alcohol outlets is about the same as or statistically greater than the risk they incur from heavy drinking. The combination of heavy drinking and being near off-premise outlets resulted in greater risk than either factor alone. By comparison, light drinking and being near on-premise alcohol outlets were not associated with increased risks for gun assault. Cities should consider addressing alcohol-related factors, especially off-premise outlets, as highly modifiable and politically feasible approaches to reducing gun violence.
Violence; Injury; Alcohol; Alcohol Outlets; Geography; Epidemiology
Phagocytic removal of apoptotic cells occurs efficiently in vivo such that even in tissues with significant apoptosis, very few apoptotic cells are detectable1. This is thought to be due to the release of find-me signals by apoptotic cells that recruit motile phagocytes such as monocytes, macrophages, and dendritic cells, leading to the prompt clearance of the dying cells2. However, the identity and in vivo relevance of such find-me signals are not well understood. Here, through several lines of evidence, we identify extracellular nucleotides as a critical apoptotic cell find-me signal. We demonstrate the caspase-dependent release of ATP and UTP (in equimolar quantities) during the early stages of apoptosis by primary thymocytes and cell lines. Purified nucleotides at these concentrations were sufficient to induce monocyte recruitment comparable to apoptotic cell supernatants. Enzymatic removal of ATP and UTP (by apyrase or ectopic CD39 expression) abrogated the ability of apoptotic cell supernatants to recruit monocytes in vitro and in vivo. We then identified the ATP/UTP receptor P2Y2 as a critical sensor of nucleotides released by apoptotic cells using RNAi depletion studies in monocytes, and macrophages from P2Y2-null mice3. The in vivo relevance of nucleotides in apoptotic cell clearance was revealed by two approaches. First, in a murine air-pouch model, apoptotic cell supernatants induced a three-fold greater recruitment of monocytes and macrophages compared to supernatants from healthy cells; this recruitment was abolished by depletion of nucleotides and significantly decreased in P2Y2−/− mice. Second, clearance of apoptotic thymocytes was significantly impaired by either depletion of nucleotides or interference with P2Y receptor function (by pharmacological inhibition, or in P2Y2−/− mice). These results identify nucleotides as a critical find-me cue released by apoptotic cells to promote P2Y2-dependent phagocyte recruitment, and provide strong evidence for a clear relationship between a find-me signal and efficient corpse clearance in vivo.
Positive and negative affect data are often collected over time in psychiatric care settings, yet no generally accepted means are available to relate these data to useful diagnoses or treatments. Latent class analysis attempts data reduction by classifying subjects into one of K unobserved classes based on observed data. Latent class models have recently been extended to accommodate longitudinally observed data. We extend these approaches in a Bayesian framework to accommodate trajectories of both continuous and discrete data. We consider whether latent class models might be used to distinguish patients on the basis of trajectories of observed affect scores, reported events, and presence or absence of clinical depression.
Cardiovascular disease; Depression; DIC; General growth mixture modeling; Gibbs sampling; Label switching; Model choice
Lin et al. (http://www.biostatsresearch.com/upennbiostat/papers/, 2006) proposed a nested Markov compliance class model in the Imbens and Rubin compliance class model framework to account for time-varying subject noncompliance in longitudinal randomized intervention studies. We use superclasses, or latent compliance class principal strata, to describe longitudinal compliance patterns, and time-varying compliance classes are assumed to depend on the history of compliance. In this paper, we search for good subject-level baseline predictors of these superclasses and also examine the relationship between these superclasses and all-cause mortality. Since the superclasses are completely latent in all subjects, we utilize multiple imputation techniques to draw inferences. We apply this approach to a randomized intervention study for elderly primary care patients with depression.
longitudinal compliance class model; noncompliance; principal stratification; latent class model; multiple imputation; geriatric depression
In sample surveys where sampled units have unequal probabilities of inclusion, associations between the inclusion probabilities and the statistic of interest can induce bias. Weights equal to the inverse of the probability of inclusion are often used to counteract this bias. Highly disproportional sample designs have highly variable weights, which can introduce undesirable variability in statistics such as the population mean or linear regression estimates. Weight trimming reduces large weights to a fixed maximum value, reducing variability but introducing bias. Most standard approaches are ad-hoc in that they do not use the data to optimize bias-variance tradeoffs. This manuscript develops variable selection models, termed “weight pooling” models, that extend weight trimming procedures in a Bayesian model averaging framework to produce “data driven” weight trimming estimators. We develop robust yet efficient models that approximate fully-weighted estimators when bias correction is of greatest importance, and approximate unweighted estimators when variance reduction is critical.
Sample survey; sampling weights; Bayesian population inference; weight pooling; variable selection; fractional Bayes Factors