Biostatistics - recent issues

Probability of detecting disease-associated single nucleotide polymorphisms in case-control genome-wide association studies

Gail, M. H., Pfeiffer, R. M., Wheeler, W., Pee, D. — 2008-03-17

Some case–control genome-wide association studies (CCGWASs) select promising single nucleotide polymorphisms (SNPs) by ranking corresponding p-values, rather than by applying the same p-value threshold to each SNP. For such a study, we define the detection probability (DP) for a specific disease-associated SNP as the probability that the SNP will be "T-selected," namely have one of the top T largest chi-square values (or smallest p-values) for trend tests of association. The corresponding proportion positive (PP) is the fraction of selected SNPs that are true disease-associated SNPs. We study DP and PP analytically and via simulations, both for fixed and for random effects models of genetic risk, that allow for heterogeneity in genetic risk. DP increases with genetic effect size and case–control sample size and decreases with the number of nondisease-associated SNPs, mainly through the ratio of T to N, the total number of SNPs. We show that DP increases very slowly with T, and the increment in DP per unit increase in T declines rapidly with T. DP is also diminished if the number of true disease SNPs exceeds T. For a genetic odds ratio per minor disease allele of 1.2 or less, even a CCGWAS with 1000 cases and 1000 controls requires T to be impractically large to achieve an acceptable DP, leading to PP values so low as to make the study futile and misleading. We further calculate the sample size of the initial CCGWAS that is required to minimize the total cost of a research program that also includes follow-up studies to examine the T-selected SNPs. A large initial CCGWAS is desirable if genetic effects are small or if the cost of a follow-up study is large.

Robust combination of multiple diagnostic tests for classifying censored event times

Cai, T., Cheng, S. — 2008-03-17

Recent advancement in technology promises to yield a multitude of tests for disease diagnosis and prognosis. When there are multiple sources of information available, it is often of interest to construct a composite score that can provide better classification accuracy than any individual measurement. In this paper, we consider robust procedures for optimally combining tests when test results are measured prior to disease onset and disease status evolves over time. To account for censoring of disease onset time, the most commonly used approach to combining tests to detect subsequent disease status is to fit a proportional hazards model (Cox, 1972) and use the estimated risk score. However, simulation studies suggested that such a risk score may have poor accuracy when the proportional hazards assumption fails. We propose the use of a nonparametric transformation model (Han, 1987) as a working model to derive an optimal composite score with theoretical justification. We demonstrate that the proposed score is the optimal score when the model holds and is optimal "on average" among linear scores even if the model fails. Time-dependent sensitivity, specificity, and receiver operating characteristic curve functions are used to quantify the accuracy of the resulting composite score. We provide consistent and asymptotically Gaussian estimators of these accuracy measures. A simple model-free resampling procedure is proposed to obtain all consistent variance estimators. We illustrate the new proposals with simulation studies and an analysis of a breast cancer gene expression data set.

Regression analysis of multivariate panel count data

He, X., Tong, X., Sun, J., Cook, R. J. — 2008-03-17

We consider panel count data which are frequently obtained in prospective studies involving recurrent events that are only detected and recorded at periodic assessment times. The data take the form of counts of the cumulative number of events detected at each inspection time, along with explanatory covariates. Examples arise in diverse areas such as epidemiological studies, medical follow-up studies, reliability studies, and tumorigenicity experiments. This article is concerned with regression analysis of multivariate panel count data which arise if more than one type of recurrent event is of interest and individuals are only observed intermittently. We present a class of marginal mean models which leave the dependence structures for related types of recurrent events completely unspecified. Estimating equations are developed for regression parameters, and the resulting estimates are shown to be consistent and asymptotically normal. Simulation studies show that the proposed estimation procedures work well for practical situations. The methodology is applied to a motivating study of patients with psoriatic arthritis in which the events of interest are the onset of joint damage according to 2 different criteria.

A penalized latent class model for ordinal data

2008-03-17

Latent class models provide a useful framework for clustering observations based on several features. Application of latent class methodology to correlated, high-dimensional ordinal data poses many challenges. Unconstrained analyses may not result in an estimable model. Thus, information contained in ordinal variables may not be fully exploited by researchers. We develop a penalized latent class model to facilitate analysis of high-dimensional ordinal data. By stabilizing maximum likelihood estimation, we are able to fit an ordinal latent class model that would otherwise not be identifiable without application of strict constraints. We illustrate our methodology in a study of schwannoma, a peripheral nerve sheath tumor, that included 3 clinical subtypes and 23 ordinal histological measures.

The 2-sample problem for failure rates depending on a continuous mark: an application to vaccine efficacy

Gilbert, P. B., McKeague, I. W., Sun, Y. — 2008-03-17

The efficacy of an HIV vaccine to prevent infection is likely to depend on the genetic variation of the exposing virus. This paper addresses the problem of using data on the HIV sequences that infect vaccine efficacy trial participants to (1) test for vaccine efficacy more powerfully than procedures that ignore the sequence data and (2) evaluate the dependence of vaccine efficacy on the divergence of infecting HIV strains from the HIV strain that is contained in the vaccine. Because hundreds of amino acid sites in each HIV genome are sequenced, it is natural to treat the genetic divergence as a continuous mark variable that accompanies each failure (infection) time. Problems (1) and (2) can then be approached by testing whether the ratio of the mark-specific hazard functions for the vaccine and placebo groups is unity or independent of the mark. We develop nonparametric and semiparametric tests for these null hypotheses and nonparametric techniques for estimating the mark-specific relative risks. The asymptotic properties of the procedures are established. In addition, the methods are studied in simulations and are applied to HIV genetic sequence data collected in the first HIV vaccine efficacy trial.

Principal stratification with predictors of compliance for randomized trials with 2 active treatments

Roy, J., Hogan, J. W., Marcus, B. H. — 2008-03-17

In behavioral medicine trials, such as smoking cessation trials, 2 or more active treatments are often compared. Noncompliance by some subjects with their assigned treatment poses a challenge to the data analyst. The principal stratification framework permits inference about causal effects among subpopulations characterized by potential compliance. However, in the absence of prior information, there are 2 significant limitations: (1) the causal effects cannot be point identified for some strata and (2) individuals in the subpopulations (strata) cannot be identified. We propose to use additional information—compliance-predictive covariates—to help identify the causal effects and to help describe characteristics of the subpopulations. The probability of membership in each principal stratum is modeled as a function of these covariates. The model is constructed using marginal compliance models (which are identified) and a sensitivity parameter that captures the association between the 2 marginal distributions. We illustrate our methods in both a simulation study and an analysis of data from a smoking cessation trial.

Stochastic segmentation models for array-based comparative genomic hybridization data analysis

Lai, T. L., Xing, H., Zhang, N. — 2008-03-17

Array-based comparative genomic hybridization (array-CGH) is a high throughput, high resolution technique for studying the genetics of cancer. Analysis of array-CGH data typically involves estimation of the underlying chromosome copy numbers from the log fluorescence ratios and segmenting the chromosome into regions with the same copy number at each location. We propose for the analysis of array-CGH data, a new stochastic segmentation model and an associated estimation procedure that has attractive statistical and computational properties. An important benefit of this Bayesian segmentation model is that it yields explicit formulas for posterior means, which can be used to estimate the signal directly without performing segmentation. Other quantities relating to the posterior distribution that are useful for providing confidence assessments of any given segmentation can also be estimated by using our method. We propose an approximation method whose computation time is linear in sequence length which makes our method practically applicable to the new higher density arrays. Simulation studies and applications to real array-CGH data illustrate the advantages of the proposed approach.

Joint inference for nonlinear mixed-effects models and time to event at the presence of missing data

Wu, L., Hu, X. J., Wu, H. — 2008-03-17

In many longitudinal studies, the individual characteristics associated with the repeated measures may be possible covariates of the time to an event of interest, and thus, it is desirable to model the time-to-event process and the longitudinal process jointly. Statistical analyses may be further complicated in such studies with missing data such as informative dropouts. This article considers a nonlinear mixed-effects model for the longitudinal process and the Cox proportional hazards model for the time-to-event process. We provide a method for simultaneous likelihood inference on the 2 models and allow for nonignorable data missing. The approach is illustrated with a recent AIDS study by jointly modeling HIV viral dynamics and time to viral rebound.

Small-sample estimation of negative binomial dispersion, with applications to SAGE data

Robinson, M. D., Smyth, G. K. — 2008-03-17

We derive a quantile-adjusted conditional maximum likelihood estimator for the dispersion parameter of the negative binomial distribution and compare its performance, in terms of bias, to various other methods. Our estimation scheme outperforms all other methods in very small samples, typical of those from serial analysis of gene expression studies, the motivating data for this study. The impact of dispersion estimation on hypothesis testing is studied. We derive an "exact" test that outperforms the standard approximate asymptotic tests.

Cross-study validation and combined analysis of gene expression microarray data

Garrett-Mayer, E., Parmigiani, G., Zhong, X., Cope, L., Gabrielson, E. — 2008-03-17

Investigations of transcript levels on a genomic scale using hybridization-based arrays have led to formidable advances in our understanding of the biology of many human illnesses. At the same time, these investigations have generated controversy because of the probabilistic nature of the conclusions and the surfacing of noticeable discrepancies between the results of studies addressing the same biological question. In this article, we present simple and effective data analysis and visualization tools for gauging the degree to which the findings of one study are reproduced by others and for integrating multiple studies in a single analysis. We describe these approaches in the context of studies of breast cancer and illustrate that it is possible to identify a substantial biologically relevant subset of the human genome within which hybridization results are reliable. The subset generally varies with the platforms used, the tissues studied, and the populations being sampled. Despite important differences, it is also possible to develop simple expression measures that allow comparison across platforms, studies, laboratories and populations. Important biological signals are often preserved or enhanced. Cross-study validation and combination of microarray results requires careful, but not overly complex, statistical thinking and can become a routine component of genomic analysis.

Efficient resampling methods for nonsmooth estimating functions

Zeng, D., Lin, D. Y. — 2008-03-17

We propose a simple and general resampling strategy to estimate variances for parameter estimators derived from nonsmooth estimating functions. This approach applies to a wide variety of semiparametric and nonparametric problems in biostatistics. It does not require solving estimating equations and is thus much faster than the existing resampling procedures. Its usefulness is illustrated with heteroscedastic quantile regression and censored data rank regression. Numerical results based on simulated and real data are provided.

A modified sign test for comparing paired ROC curves

Braun, T. M., Alonzo, T. A. — 2008-03-17

We develop a permutation test for assessing a difference in the areas under the curve (AUCs) in a paired setting where both modalities are given to each diseased and nondiseased subject. We propose that permutations be made between subjects specifically by shuffling the diseased/nondiseased labels of the subjects within each modality. As these permutations are made within modality, the permutation test is valid even if both modalities are measured on different scales. We show that our permutation test is a sign test for the symmetry of an underlying discrete distribution whose size remains valid under the assumption of equal AUCs. We demonstrate the operating characteristics of our test via simulation and show that our test is equal in power to a permutation test recently proposed by Bandos and others (2005).

Bayesian modeling of embryonic growth using latent variables

Slaughter, J. C., Herring, A. H., Hartmann, K. E. — 2008-03-17

In a growth model, individuals move progressively through a series of states in which each state is indicative of developmental status. Interest lies in estimating the rate of progression through each state while incorporating covariates that might affect the transition rates. We develop a Bayesian discrete-time multistate growth model for inference from cross-sectional data with unknown initiation times. For each subject, data are collected at only one time point at which we observe the state as well as covariates that measure developmental progress. We link the developmental progress variables to an underlying latent growth variable that can also affect the state transition rates. A subject with slow latent growth will then have relatively small developmental progress covariates and move through state transitions slowly. We then examine the association between latent growth and the probability of future events in a novel study of embryonic development and pregnancy loss. Using a Markov chain Monte Carlo (MCMC) algorithm for posterior computation, we found evidence in favor of a previously hypothesized but unproven association between slow growth early in pregnancy and increased risk of future spontaneous abortion.

A hybrid model for reducing ecological bias

Salway, R., Wakefield, J. — 2007-12-14

A major drawback of epidemiological ecological studies, in which the association between area-level summaries of risk and exposure is used to make inference about individual risk, is the difficulty in characterizing within-area variability in exposure and confounder variables. To avoid ecological bias, samples of individual exposure/confounder data within each area are required. Unfortunately, these may be difficult or expensive to obtain, particularly if large samples are required. In this paper, we propose a new approach suitable for use with small samples. We combine a Bayesian nonparametric Dirichlet process prior with an estimating functions’ approach and show that this model gives a compromise between 2 previously described methods. The method is investigated using simulated data, and a practical illustration is provided through an analysis of lung cancer mortality and residential radon exposure in counties of Minnesota. We conclude that we require good quality prior information about the exposure/confounder distributions and a large between- to within-area variability ratio for an ecological study to be feasible using only small samples of individual data.

Spatial smoothing and hot spot detection for CGH data using the fused lasso

Tibshirani, R., Wang, P. — 2007-12-14

We apply the "fused lasso" regression method of (TSRZ2004) to the problem of "hot- spot detection", in particular, detection of regions of gain or loss in comparative genomic hybridization (CGH) data. The fused lasso criterion leads to a convex optimization problem, and we provide a fast algorithm for its solution. Estimates of false-discovery rate are also provided. Our studies show that the new method generally outperforms competing methods for calling gains and losses in CGH data.

Penalized logistic regression for detecting gene interactions

Park, M. Y., Hastie, T. — 2007-12-14

We propose using a variant of logistic regression (LR) with $${L}_{2}$$-regularization to fit gene–gene and gene–environment interaction models. Studies have shown that many common diseases are influenced by interaction of certain genes. LR models with quadratic penalization not only correctly characterizes the influential genes along with their interaction structures but also yields additional benefits in handling high-dimensional, discrete factors with a binary response. We illustrate the advantages of using an $${L}_{2}$$-regularization scheme and compare its performance with that of "multifactor dimensionality reduction" and "FlexTree," 2 recent tools for identifying gene–gene interactions. Through simulated and real data sets, we demonstrate that our method outperforms other methods in the identification of the interaction structures as well as prediction accuracy. In addition, we validate the significance of the factors selected through bootstrap analyses.

Integrating quantitative information from ChIP-chip experiments into motif finding

Shim, H., Keles, S. — 2007-12-14

Identifying binding locations of transcription factors (TFs) within long segments of noncoding DNA is a challenging task. Recent chromatin immunoprecipitation on microarray (ChIP-chip) experiments utilizing tiling arrays are especially promising for this task since they provide high-resolution genome-wide maps of the interactions between the TFs and the DNA. Data from these experiments are invaluable for characterizing DNA recognition profiles (regulatory motifs) of TFs. A 2-step paradigm is commonly used for performing motif searches based on ChIP-chip data. First, candidate bound sequences that are in the order of 500–1000 bp are identified from ChIP-chip data. Then, motif searches are performed among these sequences. These 2 steps are typically carried out in a disconnected fashion in the sense that the quantitative nature of the ChIP-chip information is ignored in the second step. More specifically, all bound regions are assumed to equally likely have the motif(s), and the motifs are assumed to reside at any position of the bound regions with equal probability. We develop a conditional two-component mixture (CTCM) model that relaxes both these common assumptions by adaptively incorporating ChIP-chip information. The performances of the new and existing methods are compared using simulated data and ChIP-chip data from recently available ENCODE studies (Consortium, 2004). These studies indicate that CTCM efficiently utilizes the information available in the ChIP-chip experiments and has superior sensitivity and specificity especially when the motif of interest has low abundance among the ChIP-chip bound regions and/or low information content.

Model-based clustering on the unit sphere with an illustration using gene expression profiles

Dortet-Bernadet, J.-L., Wicker, N. — 2007-12-14

We consider model-based clustering of data that lie on a unit sphere. Such data arise in the analysis of microarray experiments when the gene expressions are standardized so that they have mean 0 and variance 1 across the arrays. We propose to model the clusters on the sphere with inverse stereographic projections of multivariate normal distributions. The corresponding model-based clustering algorithm is described. This algorithm is applied first to simulated data sets to assess the performance of several criteria for determining the number of clusters and to compare its performance with existing methods and second to a real reference data set of standardized gene expression profiles.

Retrospective analysis of haplotype-based case control studies under a flexible model for gene environment association

Chen, Y.-H., Chatterjee, N., Carroll, R. J. — 2007-12-14

Genetic epidemiologic studies often involve investigation of the association of a disease with a genomic region in terms of the underlying haplotypes, that is the combination of alleles at multiple loci along homologous chromosomes. In this article, we consider the problem of estimating haplotype–environment interactions from case–control studies when some of the environmental exposures themselves may be influenced by genetic susceptibility. We specify the distribution of the diplotypes (haplotype pair) given environmental exposures for the underlying population based on a novel semiparametric model that allows haplotypes to be potentially related with environmental exposures, while allowing the marginal distribution of the diplotypes to maintain certain population genetics constraints such as Hardy–Weinberg equilibrium. The marginal distribution of the environmental exposures is allowed to remain completely nonparametric. We develop a semiparametric estimating equation methodology and related asymptotic theory for estimation of the disease odds ratios associated with the haplotypes, environmental exposures, and their interactions, parameters that characterize haplotype–environment associations and the marginal haplotype frequencies. The problem of phase ambiguity of genotype data is handled using a suitable expectation–maximization algorithm. We study the finite-sample performance of the proposed methodology using simulated data. An application of the methodology is illustrated using a case–control study of colorectal adenoma, designed to investigate how the smoking-related risk of colorectal adenoma can be modified by "NAT2," a smoking-metabolism gene that may potentially influence susceptibility to smoking itself.

Group additive regression models for genomic data analysis

Luan, Y., Li, H. — 2007-12-14

One important problem in genomic research is to identify genomic features such as gene expression data or DNA single nucleotide polymorphisms (SNPs) that are related to clinical phenotypes. Often these genomic data can be naturally divided into biologically meaningful groups such as genes belonging to the same pathways or SNPs within genes. In this paper, we propose group additive regression models and a group gradient descent boosting procedure for identifying groups of genomic features that are related to clinical phenotypes. Our simulation results show that by dividing the variables into appropriate groups, we can obtain better identification of the group features that are related to the phenotypes. In addition, the prediction mean square errors are also smaller than the component-wise boosting procedure. We demonstrate the application of the methods to pathway-based analysis of microarray gene expression data of breast cancer. Results from analysis of a breast cancer microarray gene expression data set indicate that the pathways of metalloendopeptidases (MMPs) and MMP inhibitors, as well as cell proliferation, cell growth, and maintenance are important to breast cancer–specific survival.

A score test for linkage analysis of ordinal traits based on IBD sharing

Feng, R., Zhang, H. — 2007-12-14

Statistical methods for linkage analysis are well established for both binary and quantitative traits. However, numerous diseases including cancer and psychiatric disorders are rated on discrete ordinal scales. To analyze pedigree data with ordinal traits, we recently proposed a latent variable model which has higher power to detect linkage using ordinal traits than methods using the dichotomized traits. The challenge with the latent variable model is that the likelihood is usually very complicated, and as a result, the computation of the likelihood ratio statistic is too intensive for large pedigrees. In this paper, we derive a computationally efficient score statistic based on the identity-by-decent sharing information between relatives. Using simulation studies, we examined the asymptotic distribution of the test statistic and the power of our proposed test under various levels of heritability. We compared the computing time as well as power of the score test with the likelihood ratio test. We then applied our method for the Collaborative Study on the Genetics of Alcoholism and performed a genome scan to map susceptibility genes for alcohol dependence. We found a strong linkage signal on chromosome 4.

Microarray learning with ABC

Amaratunga, D., Cabrera, J., Kovtun, V. — 2007-12-14

Standard clustering algorithms when applied to DNA microarray data often tend to produce erroneous clusters. A major contributor to this divergence is the feature characteristic of microarray data sets that the number of predictors (genes) in such data far exceeds the number of samples by many orders of magnitude, with only a small percentage of predictors being truly informative with regards to the clustering while the rest merely add noise. An additional complication is that the predictors exhibit an unknown complex correlational configuration embedded in a small subspace of the entire predictor space. Under these conditions, standard clustering algorithms fail to find the true clusters even when applied in tandem with some sort of gene filtering or dimension reduction to reduce the number of predictors. We propose, as an alternative, a novel method for unsupervised classification of DNA microarray data. The method, which is based on the idea of aggregating results obtained from an ensemble of randomly resampled data (where both samples and genes are resampled), introduces a way of tilting the procedure so that the ensemble includes minimal representation from less important areas of the gene predictor space. The method produces a measure of dissimilarity between each pair of samples that can be used in conjunction with (a) a method like Ward's procedure to generate a cluster analysis and (b) multidimensional scaling to generate useful visualizations of the data. We call the dissimilarity measures ABC dissimilarities since they are obtained by aggregating bundles of clusters. An extensive comparison of several clustering methods using actual DNA microarray data convincingly demonstrates that classification using ABC dissimilarities offers significantly superior performance.

Combining assays for estimating prevalence of human herpesvirus 8 infection using multivariate mixture models

Pfeiffer, R. M., Carroll, R. J., Wheeler, W., Whitby, D., Mbulaiteye, S. — 2007-12-14

For many diseases, it is difficult or impossible to establish a definitive diagnosis because a perfect "gold standard" may not exist or may be too costly to obtain. In this paper, we propose a method to use continuous test results to estimate prevalence of disease in a given population and to estimate the effects of factors that may influence prevalence. Motivated by a study of human herpesvirus 8 among children with sickle-cell anemia in Uganda, where 2 enzyme immunoassays were used to assess infection status, we fit 2-component multivariate mixture models. We model the component densities using parametric densities that include data transformation as well as flexible transformed models. In addition, we model the mixing proportion, the probability of a latent variable corresponding to the true unknown infection status, via a logistic regression to incorporate covariates. This model includes mixtures of multivariate normal densities as a special case and is able to accommodate unusual shapes and skewness in the data. We assess model performance in simulations and present results from applying various parameterizations of the model to the Ugandan study.

Inverse sampling of controls in a matched case control study

Keogh, R. H. — 2007-12-14

A method of inverse sampling of controls in a matched case–control study is described in which, for each case, controls are sampled until a discordant set is achieved. For a binary exposure, inverse sampling is used to determine the number of controls for each case. When most individuals in a population have the same exposure, standard case–control sampling may result in many case–control sets being concordant with respect to exposure and thus uninformative in the conditional logistic analysis. The method using inverse control sampling is proposed as a solution to this problem in situations when it is practically feasible. In many circumstances, inverse control sampling is found to offer improved statistical efficiency relative to a comparable study with a fixed number of controls per case.

Nonlinear growth generates age changes in the moments of the frequency distribution: the example of height in puberty

Cole, T. J., Cortina-Borja, M., Sandhu, J., Kelly, F. P., Pan, H. — 2007-12-14

Higher moments of the frequency distribution of child height and weight change with age, particularly during puberty, though why is not known. Our aims were to confirm that height skewness and kurtosis change with age during puberty, to devise a model to explain why, and to test the model by analyzing the data longitudinally. Heights of 3245 Christ's Hospital School boys born during 1927–1956 were measured twice termly from 9 to 20 years ($$n=129508$$). Treating the data as independent, the mean, standard deviation (SD), skewness, and kurtosis were calculated in 40 age groups and plotted as functions of age t. The data were also analyzed longitudinally using the nonlinear random-effects growth model $$H\left(t\right)=h(t-\epsilon )+\alpha $$, with $$H\left(t\right)$$ the cross-sectional data, $$h\left(t\right)$$ the individual mean curve, and $$\epsilon $$ and $$\alpha $$ subject-specific random effects reflecting variability in age and height at peak height velocity (PHV). Mean height increased monotonically with age, while the SD, skewness, and kurtosis changed cyclically with, respectively, 1, 2, and 3 turning points. Surprisingly, their age curves corresponded closely in shape to the first, second, and third derivatives of the mean height curve. The growth model expanded as a Taylor series in $$\epsilon $$ predicted such a pattern, and the longitudinal analysis showed that adjusting for age at PHV on a multiplicative scale largely removed the trends in the higher moments. A nonlinear growth process where subjects grow at different rates, such as in puberty, generates cyclical changes in the higher moments of the frequency distribution.

An alternative model for bivariate random-effects meta-analysis when the within-study correlations are unknown

Riley, R. D., Thompson, J. R., Abrams, K. R. — 2007-12-14

Multivariate meta-analysis models can be used to synthesize multiple, correlated endpoints such as overall and disease-free survival. A hierarchical framework for multivariate random-effects meta-analysis includes both within-study and between-study correlation. The within-study correlations are assumed known, but they are usually unavailable, which limits the multivariate approach in practice. In this paper, we consider synthesis of 2 correlated endpoints and propose an alternative model for bivariate random-effects meta-analysis (BRMA). This model maintains the individual weighting of each study in the analysis but includes only one overall correlation parameter, , which removes the need to know the within-study correlations. Further, the only data needed to fit the model are those required for a separate univariate random-effects meta-analysis (URMA) of each endpoint, currently the common approach in practice. This makes the alternative model immediately applicable to a wide variety of evidence synthesis situations, including studies of prognosis and surrogate outcomes. We examine the performance of the alternative model through analytic assessment, a realistic simulation study, and application to data sets from the literature. Our results show that, unless $$\widehat{\rho }$$ is very close to 1 or –1, the alternative model produces appropriate pooled estimates with little bias that (i) are very similar to those from a fully hierarchical BRMA model where the within-study correlations are known and (ii) have better statistical properties than those from separate URMAs, especially given missing data. The alternative model is also less prone to estimation at parameter space boundaries than the fully hierarchical model and thus may be preferred even when the within-study correlations are known. It also suitably estimates a function of the pooled estimates and their correlation; however, it only provides an approximate indication of the between-study variation. The alternative model greatly facilitates the utilization of correlation in meta-analysis and should allow an increased application of BRMA in practice.

Identification of SNP interactions using logic regression

Schwender, H., Ickstadt, K. — 2007-12-14

Interactions of single nucleotide polymorphisms (SNPs) are assumed to be responsible for complex diseases such as sporadic breast cancer. Important goals of studies concerned with such genetic data are thus to identify combinations of SNPs that lead to a higher risk of developing a disease and to measure the importance of these interactions. There are many approaches based on classification methods such as CART and random forests that allow measuring the importance of single variables. But none of these methods enable the importance of combinations of variables to be quantified directly. In this paper, we show how logic regression can be employed to identify SNP interactions explanatory for the disease status in a case–control study and propose 2 measures for quantifying the importance of these interactions for classification. These approaches are then applied on the one hand to simulated data sets and on the other hand to the SNP data of the GENICA study, a study dedicated to the identification of genetic and gene–environment interactions associated with sporadic breast cancer.

Biostatistics - Referees of Manuscripts Submitted Mid-2006 to Mid-2007

2007-12-14

Estimation of the benchmark dose by structural equation models

Budtz-Jorgensen, E. — 2007-09-25

While epidemiological data typically contain a multivariate response and often also multiple exposure parameters, current methods for safe dose calculations, including the widely used benchmark approach, rely on standard regression techniques. In practice, dose–response modeling and calculation of the exposure limit are often based on the seemingly most sensitive outcome. However, this procedure ignores other available data, is inefficient, and fails to account for multiple testing. Instead, risk assessment could be based on structural equation models, which can accommodate both a multivariate exposure and a multivariate response function. Furthermore, such models will allow for measurement error in the observed variables, which is a requirement for unbiased estimation of the benchmark dose. This methodology is illustrated with the data on neurobehavioral effects in children prenatally exposed to methylmercury, where results based on standard regression models cause an underestimation of the true risk.

The influence of competing-risks setting on the choice of hypothesis test for treatment effect

Williamson, P. R., Kolamunnage-Dona, R., Tudur Smith, C. — 2007-09-25

There is considerable debate regarding the choice of test for treatment difference in a randomized clinical trial in the presence of competing risks. This question arose in the study of standard and new antiepileptic drugs (SANAD) trial comparing new and standard antiepileptic drugs. This paper provides simulation results for the log-rank test comparing cause-specific hazard rates and Gray's test comparing cause-specific cumulative incidence curves. To inform the analysis of the SANAD trial, competing-risks settings were considered where both events are of interest, events may be negatively correlated, and the degree of correlation may differ in the 2 treatment groups. In settings where there are effects in opposite directions for the 2 event types, a likely situation for the SANAD trial, Gray's test has greater power to detect treatment differences than log-rank analysis. For the epilepsy application, conclusions were qualitatively similar for both log-rank and Gray's tests.

When should one subtract background fluorescence in 2-color microarrays?

Scharpf, R. B., Iacobuzio-Donahue, C. A., Sneddon, J. B., Parmigiani, G. — 2007-09-25

Two-color microarrays are a powerful tool for genomic analysis, but have noise components that make inferences regarding gene expression inefficient and potentially misleading. Background fluorescence, whether attributable to nonspecific binding or other sources, is an important component of noise. The decision to subtract fluorescence surrounding spots of hybridization from spot fluorescence has been controversial, with no clear criteria for determining circumstances that may favor, or disfavor, background subtraction. While it is generally accepted that subtracting background reduces bias but increases variance in the estimates of the ratios of interest, no formal analysis of the bias–variance trade off of background subtraction has been undertaken. In this paper, we use simulation to systematically examine the bias–variance trade off under a variety of possible experimental conditions. Our simulation is based on data obtained from 2 self versus self microarray experiments and is free of distributional assumptions. Our results identify factors that are important for determining whether to background subtract, including the correlation of foreground to background intensity ratios. Using these results, we develop recommendations for diagnostic visualizations that can help decisions about background subtraction.

Joint frailty models for recurring events and death using maximum penalized likelihood estimation: application on cancer events

2007-09-25

The observation of repeated events for subjects in cohort studies could be terminated by loss to follow-up, end of study, or a major failure event such as death. In this context, the major failure event could be correlated with recurrent events, and the usual assumption of noninformative censoring of the recurrent event process by death, required by most statistical analyses, can be violated. Recently, joint modeling for 2 survival processes has received considerable attention because it makes it possible to study the joint evolution over time of 2 processes and gives unbiased and efficient parameters. The most commonly used estimation procedure in the joint models for survival events is the expectation maximization algorithm. We show how maximum penalized likelihood estimation can be applied to nonparametric estimation of the continuous hazard functions in a general joint frailty model with right censoring and delayed entry. The simulation study demonstrates that this semiparametric approach yields satisfactory results in this complex setting. As an illustration, such an approach is applied to a prospective cohort with recurrent events of follicular lymphomas, jointly modeled with death.

Impact of nonignorable coarsening on Bayesian inference

Zhang, J., Heitjan, D. F. — 2007-09-25

The coarse data model of Heitjan and Rubin (1991) generalizes the missing data model of Rubin (1976) to cover other forms of incompleteness such as censoring and grouping. The model has 2 components: an ideal data model describing the distribution of the quantity of interest and a coarsening mechanism that describes a distribution over degrees of coarsening given the ideal data. The coarsening mechanism is said to be nonignorable when the degree of coarsening depends on an incompletely observed ideal outcome, in which case failure to properly account for it can spoil inferences. A theme in recent research is to measure sensitivity to nonignorability by evaluating the effect of a small departure from ignorability on the maximum likelihood estimate (MLE) of a parameter of the ideal data model. One such construct is the "index of local sensitivity to nonignorability" (ISNI) (Troxel and others, 2004), which is the derivative of the MLE with respect to a nonignorability parameter evaluated at the ignorable model. In this paper, we adapt ISNI to Bayesian modeling by instead defining it as the derivative of the posterior expectation. We propose the application of ISNI as a first step in judging the robustness of a Bayesian analysis to nonignorable coarsening. We derive formulas for a range of models and apply the method to evaluate sensitivity to nonignorable coarsening in 2 real data examples, one involving missing CD4 counts in an HIV trial and the other involving potentially informatively censored relapse times in a leukemia trial.

A moment-based method for estimating the proportion of true null hypotheses and its application to microarray gene expression data

Lai, Y. — 2007-09-25

Due to advances in experimental technologies, it is feasible to collect measurements for a large number of variables. When these variables are simultaneously screened by a statistical test, it is necessary to consider the adjustment for multiple hypothesis testing. The false discovery rate has been proposed and widely used to address this issue. A related problem is the estimation of the proportion of true null hypotheses. The long-standing difficulty to this problem is the identifiability of the nonparametric model. In this study, we propose a moment-based method coupled with sample splitting for estimating this proportion. If the p values from the alternative hypothesis are homogeneously distributed, then the proposed method will solve the identifiability and give its optimal performances. When the p values from the alternative hypothesis are heterogeneously distributed, we propose to approximate this mixture distribution so that the identifiability can be achieved. Theoretical aspects of the approximation error are discussed. The proposed estimation method is completely nonparametric and simple with an explicit formula. Simulation studies show the favorable performances of the proposed method when it is compared to the other existing methods. Two microarray gene expression data sets are considered for applications.

Identifying latent clusters of variability in longitudinal data

Elliott, M. R. — 2007-09-25

Means or other central tendency measures are by far the most common focus of statistical analyses. However, as Carroll (2003) noted, "systematic dependence of variability on known factors" may be "fundamental to the proper solution of scientific problems" in certain settings. We develop a latent cluster model that relates underlying "clusters" of variability to baseline or outcome measures of interest. Because estimation of variability is inextricably linked to estimation of trend, assumptions about underlying trends are minimized by using nonparametric regression estimates. The resulting residual errors are then clustered into unobserved clusters of variability that are in turn related to subject-level predictors of interest. An application is made to psychological affect data.

The effect of miss-specified baseline characteristics on inference for longitudinal trends in linear mixed models

Verbeke, G., Fieuws, S. — 2007-09-25

The main advantage of longitudinal studies is that they can distinguish changes over time within individuals (longitudinal effects) from differences among subjects at the start of the study (baseline characteristics, cross-sectional effects). Often, especially in observational studies, longitudinal trends are studied after correction for many potentially important baseline differences between subjects. We show that, in the context of linear mixed models, inference for longitudinal trends is in general biased if a wrong model for the baseline characteristics is used. However, we will argue that this bias is small in most practical situations and completely vanishes in the special case of a growth curve model for complete balanced data. In the latter case, inference for longitudinal trends is completely independent of additional baseline covariates that might have been omitted from the model.

Bayesian regularization of diffusion tensor images

2007-09-25

Diffusion tensor imaging (DTI) is a powerful tool in the study of the course of nerve fiber bundles in the human brain. Using DTI, the local fiber orientation in each image voxel can be described by a diffusion tensor which is constructed from local measurements of diffusion coefficients along several directions. The measured diffusion coefficients and thereby the diffusion tensors are subject to noise, leading to possibly flawed representations of the 3-dimensional (3D) fiber bundles. In this paper, we develop a Bayesian procedure for regularizing the diffusion tensor field, fully utilizing the available 3D information of fiber orientation. The use of the procedure is exemplified on synthetic and in vivo data.

On the potential for illogic with logically defined outcomes

Li, X., Caffo, B., Scharfstein, D. — 2007-09-25

Logically defined outcomes are commonly used in medical diagnoses and epidemiological research. When missing values in the original outcomes exist, the method of handling the missingness can have unintended consequences, even if the original outcomes are missing completely at random. In this note, we consider 2 binary original outcomes, which are missing completely at random. For estimating the prevalence of a logically defined "or" outcome, we discuss the properties of 4 estimators: the complete-case estimator, the available-case estimator, the maximum likelihood estimator (MLE), and a moment-based estimator. With the exception of the available-case case estimator, all the estimators are consistent. The MLE exhibits superior performance and should be generally adopted.

A temporal hidden Markov regression model for the analysis of gene regulatory networks

Gupta, M., Qu, P., Ibrahim, J. G. — 2007-09-25

We propose a novel hierarchical hidden Markov regression model for determining gene regulatory networks from genomic sequence and temporally collected gene expression microarray data. The statistical challenge is to simultaneously determine the groupings of genes and subsets of motifs involved in their regulation, when the groupings may vary over time, and a large number of potential regulators are available. We devise a hybrid Monte Carlo methodology to estimate parameters under 2 classes of latent structure, one arising due to the unobservable state identity of genes and the other due to the unknown set of covariates influencing the response within a state. The effectiveness of this method is demonstrated through a simulation study and an application on an yeast cell-cycle data set.

Fitting semiparametric random effects models to large data sets

Pennell, M. L., Dunson, D. B. — 2007-09-25

For large data sets, it can be difficult or impossible to fit models with random effects using standard algorithms due to memory limitations or high computational burdens. In addition, it would be advantageous to use the abundant information to relax assumptions, such as normality of random effects. Motivated by data from an epidemiologic study of childhood growth, we propose a 2-stage method for fitting semiparametric random effects models to longitudinal data with many subjects. In the first stage, we use a multivariate clustering method to identify G<<N groups of subjects whose data have no scientifically important differences, as defined by subject matter experts. Then, in stage 2, group-specific random effects are assumed to come from an unknown distribution, which is assigned a Dirichlet process prior, further clustering the groups from stage 1. We use our approach to model the effects of maternal smoking during pregnancy on growth in 17 518 girls.

Index

2007-09-25