Search

Birol Emir,1 Dhammika Amaratunga,2 Mohan Beltangady,3 Javier Cabrera,4 Roy Freeman,5 David Madigan,6 Ha Nguyen,1 Ed Whalen11Primary Care Business Unit, Pfizer, Inc., New York, NY, USA; 2Nonclinical Statistics and Computing, Johnson & Johnson, Raritan, NJ, USA; 3Pfizer, Inc., New London, CT, USA; 4Rutgers University, Piscataway, NJ, USA; 5Harvard Medical School, Harvard University, Cambridge, MA, USA; 6Department of Statistics, Columbia University, New York, NY, USAAbstract: Due to the ever-increasing complexity of scientific technologies and resulting data, consulting statisticians are becoming more involved in the design, conduct, and analysis of biomedical research. This requires extensive collaboration between the consulting statistician and nonstatisticians, such as researchers, clinicians, and corporate executives. Consequently, a successful consulting career is becoming ever more dependent on the statistician's ability to effectively communicate with nonstatisticians. This is especially true when more complex, nontraditional analytical methods are required. In this paper, we examine the collaboration between statisticians and nonstatisticians from three different professional perspectives. Integrating these perspectives, we discuss ways to help the consulting statistician generate productive dialogue with clients. Finally, we examine how universities can better prepare students for careers in statistical consulting by incorporating more communication-based elements into their curriculum and by offering students ample opportunities to collaborate with nonstatisticians. Overall, we designed this exercise to help the consulting statistician generate dialogue with clients that results in more productive collaborations and a more satisfying work experience.Keywords: statistical consulting, nontraditional analysis, communication

To the Editor: In their article, “Performance of Pharmacovigilance Signal-Detection Algorithms for the FDA Adverse Event Reporting System,” Harpaz et al.1 report an evaluation of signal-detection a ...

When you next pass a motorway being repaired or widened and you see blocks of concrete being maneuvred into position by a large crane a few feet away from traffic passing through a contraflow at 50mph, think of CSCW. For in this world of ‘muck and bullets’ a revolution is about to occur. In a very few years, the hard hat and safety boots are going to be joined on site by the electronic hard hat and the laptop computer.

We consider the problem of model selection for Bayesian graphical models, and embed it in the larger context of accounting for model uncertainty. Data analysts typically select a single model from some class of models, and then condition all subsequent inference on this model. However, this approach ignores model uncertainty, leading to poorly calibrated predictions: it will often be seen in retrospect that one’s uncertainty bands were not wide enough. The Bayesian analyst solves this problem by averaging over all plausible models when making inferences about quantities of interest. In many applications, however, because of the size of the model space and awkward integrals, this averaging will not be a practical proposition, and approximations are required. Here we examine the predictive performance of two recently proposed model averaging schemes. In the examples considered, both schemes outperform any single model that might reasonably have been selected.

In a hierarchical Bayesian expert system, the probabilities relating the variables are not known precisely; rather, imprecise knowledge of these probabilities is described by placing prior distributions on them. After obtaining data, one would like to update those distributions to reflect the new information gained; however, this can prove difficult computationally if the observed data are incomplete. This paper describes a way around these difficulties—use of Markov chain Monte Carlo methods.

A team of experts assemble a graphical belief network from many small pieces. This paper catalogs the types of knowledge that comprise a graphical belief network and proposes a way in which they can be stored in libraries. This promotes reuse of model components both within the team and between projects.

In recent years, much attention has been directed at various graphical “conditional independence” models and at the application of such graphical models to probabilistic expert systems. However, there exists a broad range of statistical problems to which Bayesian graphical models, in particular, can be applied.

Background In September 2004, rofecoxib was voluntarily withdrawn from the worldwide market. Our objective was to determine whether and when analysis of published and unpublished placebo-controlled trials could have revealed cardiovascular risk associated with rofecoxib before its withdrawal as an example to inform future postmarket pharmaceutical safety surveillance efforts. Methods We conducted a cumulative subject-level pooled analysis of data from all randomized, placebo-controlled trials of rofecoxib conducted by the manufacturer before September 2004. Our main outcome measurement was incidence of any investigator-reported death from any cause or cardiovascular thromboembolic (CVT) adverse event. Results We identified 30 randomized, placebo-controlled trials of rofecoxib that enrolled a combined 20 152 subjects. Trial duration ranged from 4 weeks to 4 years; enrollment ranged from 17 to 2586 subjects prescribed either rofecoxib or placebo; and rofecoxib dose ranged from 12.5 mg to 50 mg. As of December 2000, 21 of these trials had been completed (70%), and the risk of a CVT adverse event or death was greater among subjects assigned to the rofecoxib group (rate ratio [RR], 2.18; 95% confidence interval [CI], 0.93-5.81) ( P = .07), raising concerns from a safety standpoint. Subsequently collected data through June 2001 showed that rofecoxib was associated with a 35% increased risk of a CVT adverse event or death (RR, 1.35; 95% CI, 1.00-1.96) ( P = .05). Analyzing data available as of April 2002, we found a 39% increased risk (RR, 1.39; 95% CI, 1.07-1.80) ( P = .02), and using data available as of September 2004, we found a 43% increased risk (RR,1.43; 95% CI, 1.16-1.76) ( P Conclusion Cumulative pooled analysis of all randomized, placebo-controlled trials demonstrates a trend toward increased cardiovascular risk associated with rofecoxib compared with placebo as early as December 2000, the comparison reaching a P value of .05 by June 2001, nearly 3½ years before the manufacturer's voluntary market withdrawal.

The availability and outcome of abortion services as of 1983-84 and 1993-94 in rural Washington State were investigated. The population data include birth fetal death and pregnancy termination which came from the vital statistics data compiled by Washington State. Results showed that birth rates and abortion rates decreased throughout the state from 1983-84 to 1993-94. The magnitude of the drop in abortion rates was significantly greater in rural than in urban women (p < 0.01). The rural abortion rate fell 27% compared with a 17% drop in the urban rate. The declination in the abortion rate was larger for adolescents than other age groups. The abortion rate for adolescents aged 10-19 years dropped 35% in rural areas and 28% in urban areas. 12 miles increased the distance that rural women traveled to obtain abortion. The proportion of rural women having abortions decreased significantly from 25% to 3%. During 1983-84 62% traveled 50 miles to obtain abortion compared with 73% in 1993-94. In both time periods the proportion of rural women who traveled out of state for an abortion increased from 8% to 14%. Furthermore the proportion of women terminating their pregnancy after the first trimester increased from 8% in 1983-84 to 15% in 1993-94. More work is needed to understand the relationships among provider availability other factors influencing decision-making and pregnancy outcomes.

Belief networks provide an important bridge between statistical modeling and expert systems. This article presents methods for visualizing probabilistic “evidence flows” in belief networks, thereby enabling belief networks to explain their behavior. Building on earlier research on explanation in expert systems, we present a hierarchy of explanations, ranging from simple colorings to detailed displays. Our approach complements parallel work on textual explanations in belief networks. Graphical-Belief, Mathsoft Inc.'s belief network software, implements the methods.

A method for the determination of ascorbic acid in beer by using HPLC with dual electrode electrochemical detection is described. Ascorbic acid was separated by using ion suppression chromatography on a C18 column and detected by a dual-electrode system consisting of a porous graphite high efficiency electrode followed in series by a glassy carbon amperometric electrode. It was easily detectable at sub-ppm levels in beer. The use of dual electrode detection provided a constant measurement of both peak purity and electrode efficiency.

Pendant plus d'un demi-siecle, les graphes ont ete utilises pour representer des modeles statistiques en analyse de donnees. En particulier, les graphes decrivant l'independance conditionnelle sont apparus comme une classe importante de ces modeles. Des applications en analyse d'image, analyse de pedigree, ou encore en systeme expert sont a l'origine de leur developpement, et plusieurs livres de synthese ont deja ete publies a ce sujet. Le developpement d'un cadre Bayesien de ces modeles est en revanche moins connu, et les applications en systemes experts ont motive la recherche dans ce domaine. La possibilite de construire des modeles capables de se remettre a jour au fur et a mesure que de nouvelles donnees sont disponibles est a l'origine d'un interet intense de la part de la communaute travaillant en intelligence artificielle. Cependent, leur application a une classe plus vaste d'analyse de donnees a ete largement negligee. L'objet de cet article est de montrer comment les modeles Bayesiens de graphes permettent d'unifier et de simplifier des problemes standards, tels que les modeles log-lineaires Bayesiens (avec des donnees completes ou non), l'estimation d'une population fermee ou le double echantillonnage. Dans la mesure ou le choix d'un modele conventionel unique echoue dans ce type de situation, nous construisons des distributions a posteriori des quantites d'interet en moyennant sur les modeles possibles. Plus particulierement, nous introduisons la composition de chaines de Markov-Monte Carlo, une methode de Monte-Carlo permettant de moyenner sur les modeles retenus.

SUMMARY In double-sampling schemes, a large sample is classified by using one method, and a subsample is also classified with a supplementary method. In the application discussed here, we are attempting to identify infants in Norway born with Down's syndrome by using both a national birth registry and a regional registry. Usual methods for analysing such data assume that one classification method is perfect, which is not the case here. We develop a Bayesian approach that allows for error in both registries, includes covariates (here, the age of the mother) and explicitly accounts for our lack of knowledge about the complexity of the relationships between the variables considered. Markov chain Monte Carlo methods are used to approximate the posterior. In the data considered here, the error rates of the two registries appear to be substantial. Despite a strong relationship between maternal age and risk of Down's syndrome, the inclusion of the maternal age covariate does not substantially change the overall estimates.

A method was developed to determine the flavanols prodelphinidin B3, procyanidin B3, (+)-catechin, and (-)-epicatechin by high-performance liquid chromatography, using dual-channel electrochemical detection. This method was especially suited to the direct analysis of beer samples and to analysis of acetone extracts of barley samples, and was capable of determining proanthocyanidins and catechins at levels of 0.1-5.0 mg l-1. The use of dual-channel electrochemical detection also enabled unambiguous peak identification by measurement of collection efficiencies. This method offered improved sensitivity and selectivity compared with ultraviolet detection, and sample preparation procedures were greatly simplified. The method was applied to the comparison of stabilized and unstabilized lagers, and to the analysis of different barley varieties grown in Ireland.

Markov Chain Monte Carlo (MCMC) techniques revolutionized statistical practice in the 1990s by providing an essential toolkit for making the rigor and flexibility of Bayesian analysis computationally practical. At the same time the increasing prevalence of massive datasets and the expansion of the field of data mining has created the need to produce statistically sound methods that scale to these large problems. Except for the most trivial examples, current MCMC methods require a complete scan of the dataset for each iteration eliminating their candidacy as feasible data mining techniques.In this article we present a method for making Bayesian analysis of massive datasets computationally feasible. The algorithm simulates from a posterior distribution that conditions on a smaller, more manageable portion of the dataset. The remainder of the dataset may be incorporated by reweighting the initial draws using importance sampling. Computation of the importance weights requires a single scan of the remaining observations. While importance sampling increases efficiency in data access, it comes at the expense of estimation efficiency. A simple modification, based on the "rejuvenation" step used in particle filters for dynamic systems models, sidesteps the loss of efficiency with only a slight increase in the number of data accesses.To show proof-of-concept, we demonstrate the method on a mixture of transition models that has been used to model web traffic and robotics. For this example we show that estimation efficiency is not affected while offering a 95% reduction in data accesses.