Search

Undiscovered side effects of drugs can have a profound effect on the health of the nation, and electronic health-care databases offer opportunities to speed up the discovery of these side effects. We applied a “medication-wide association study” approach that combined multivariate analysis with exploratory visualization to study four health outcomes of interest in an administrative claims database of 46 million patients and a clinical database of 11 million patients. The technique had good predictive value, but there was no threshold high enough to eliminate false-positive findings. The visualization not only highlighted the class effects that strengthened the review of specific products but also underscored the challenges in confounding. These findings suggest that observational databases are useful for identifying potential associations that warrant further consideration but are unlikely to provide definitive evidence of causal effects.

Background Observational healthcare data offer the potential to enable identification of risks of medical products, and the medical literature is replete with analyses that aim to accomplish this objective. A number of established analytic methods dominate the literature but their operating characteristics in real-world settings remain unknown. Objectives To compare the performance of seven methods (new user cohort, case control, self-controlled case series, self-controlled cohort, disproportionality analysis, temporal pattern discovery, and longitudinal gamma poisson shrinker) as tools for risk identification in observational healthcare data. Research Design The experiment applied each method to 399 drug-outcome scenarios (165 positive controls and 234 negative controls across 4 health outcomes of interest) in 5 real observational databases (4 administrative claims and 1 electronic health record). Measures Method performance was evaluated through Area Under the receiver operator characteristics Curve (AUC), bias, mean square error, and confidence interval coverage probability. Results Multiple methods offer strong predictive accuracy, with AUC > 0.70 achievable for all outcomes and databases with more than one analytical approach. Selfcontrolled methods (self-controlled case series, temporal pattern discovery, self-controlled cohort) had higher predictive accuracy than cohort and case control methods across all databases and outcomes. Methods differed in the expected value and variance of the error distribution. All methods had lower coverage probability than the expe Conclusions Observational healthcare data can inform risk identification of medical product effects on acute liver injury, acute myocardial infarction, acute renal failure and gastrointestinal bleeding. However, effect estimates from all methods require calibration to address inconsistency in method operating characteristics. Further empirical evaluation is required to gauge the generalizability of these findings to other databases and outcomes.

To evaluate the performance of a disproportionality design, commonly used for analysis of spontaneous reports data such as the FDA Adverse Event Reporting System database, as a potential analytical method for an adverse drug reaction risk identification system using healthcare data. We tested the disproportionality design in 5 real observational healthcare databases and 6 simulated datasets, retrospectively studying the predictive accuracy of the method when applied to a collection of 165 positive controls and 234 negative controls across 4 outcomes: acute liver injury, acute myocardial infarction, acute kidney injury, and upper gastrointestinal bleeding. We estimate how well the method can be expected to identify true effects and discriminate from false findings and explore the statistical properties of the estimates the design generates. The primary measure was the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. For each combination of 4 outcomes and 5 databases, 48 versions of disproportionality analysis (DPA) were carried out and the AUC computed. The majority of the AUC values were in the range of 0.35

Causal assessment of adverse effects continues to evolve through the medical product lifecycle and requires clinical judgment to integrate evidence from multiple sources, including observational studies. This study describes a probabilistic framework quantifying how much can be learned from an epidemiological study, and how it varies by study design, database, or prior beliefs. We integrate new observational evidence with existing clinical knowledge by estimating the probability that a drug–outcome pair represents a true association, as a function of prior belief and estimated strength of association from an observational study. Incident user cohort with propensity score adjustment and self-controlled case series designs were applied in an automated fashion to 53 drug–outcome associations across a network of nine disparate databases covering over 100 million lives of patient-level data. When applying a incident user design to a large claims database for a drug with moderate sample, the performance of thes...

The relations among the classes of multivariate conditional independence models determined by directed acyclic graphs (DAG), undirected graphs (UDG), decomposable graphs (DEC), and finite distributive lattices (LCI) are investigated. First, LCI models that admit positive joint densities are characterized in terms of an appropriate factorization of the density. This factorization is then recognized as a particular form of the recursive factorization that characterizes DAG models, thereby establishing that the LCI models comprise a subclass of the class of DAG models. Precisely, the class of LCI models coincides with the subclass of transitive DAG models. Furthermore, the class of LCI models has nontrivial intersection with the class of DEC models. A series of examples illustrating these relations are presented.

The amounts of different phenolics adsorbed from beer and model solutions by various dosages of two commercial grades of polyvinylpolypyrrolidone (PVPP) were measured accurately by high-performance liquid chromatography, using electrochemical detection. Adsorption of simple phenolic acids and flavanols (catechins and proanthocyanidins) displayed classical Freundlich characteristics, and the isotherms obtained were compared to determine the extent of selectivity. Although PVPP preferentially adsorbed the more highly hydroxylated phenolic acids in model mixtures, a similar trend was not observed for the adsorption of dimeric flavanols in beer. The adsorptive capacity of commercial PVPP increased with decreasing particle size.

Both knowledge-based systems and statistical models are typically concerned with making predictions about future observables. Here we focus on assessment of predictive performance and provide two techniques for improving the predictive performance of Bayesian graphical models. First, we present Bayesian model averaging, a technique for accounting for model uncertainty. Second, we describe a technique for eliciting a prior distribution for competing models from domain experts. We explore the predictive performance of both techniques in the context of a urological diagnostic problem.

Characterization of relationships between time-varying drug exposures and adverse events (AEs) related to health outcomes represents the primary objective in postmarketing drug safety surveillance. Such surveillance increasingly utilizes large-scale longitudinal observational databases (LODs), containing time-stamped patient-level medical information including periods of drug exposure and dates of diagnoses for millions of patients. Statistical methods for LODs must confront computational challenges related to the scale of the data, and must also address confounding and other biases that can undermine efforts to estimate effect sizes. Methods that compare on-drug with off-drug periods within patient offer specific advantages over between patient analysis on both counts. To accomplish these aims, we extend the self-controlled case series (SCCS) for LODs. SCCS implicitly controls for fixed multiplicative baseline covariates since each individual acts as their own control. In addition, only exposed cases are required for the analysis, which is computationally advantageous. The standard SCCS approach is usually used to assess single drugs and therefore estimates marginal associations between individual drugs and particular AEs. Such analyses ignore confounding drugs and interactions and have the potential to give misleading results. In order to avoid these difficulties, we propose a regularized multiple SCCS approach that incorporates potentially thousands or more of time-varying confounders such as other drugs. The approach successfully handles the high dimensionality and can provide a sparse solution via an L₁ regularizer. We present details of the model and the associated optimization procedure, as well as results of empirical investigations.

We consider the problem of model selection and accounting for model uncertainty in high-dimensional contingency tables, motivated by expert system applications. The approach most used currently is a stepwise strategy guided by tests based on approximate asymptotic P values leading to the selection of a single model; inference is then conditional on the selected model. The sampling properties of such a strategy are complex, and the failure to take account of model uncertainty leads to underestimation of uncertainty about quantities of interest. In principle, a panacea is provided by the standard Bayesian formalism that averages the posterior distributions of the quantity of interest under each of the models, weighted by their posterior model probabilities. Furthermore, this approach is optimal in the sense of maximizing predictive ability. But this has not been used in practice, because computing the posterior model probabilities is hard and the number of models is very large (often greater than 1...

Sensitive methods for the determination of ferulic acid and 4-vinyl guaiacol in beers and worts are described. Samples were analyzed by direct-injection high-performance liquid chromatography using an octadecyl silica column. Detection was by either fluorescence monitoring (for 4-vinyl guaiacol alone) or electrochemical detection (for simultaneous determination of ferulic acid and 4-vinyl guaiacol). The methods were applied mainly to the analysis of production beers but were used also to monitor the decarboxylation of ferulic acid to 4-vinyl guaiacol in wort during a small-scale fermentation deliberately contaminated with a wild yeast strain.

Background: Expanded availability of observational healthcare data (both administrative claims and electronic health records) has prompted the development of statistical methods for identifying adverse events associated with medical products, but the operating characteristics of these methods when applied to the real-world data are unknown. Methods: We studied the performance of eight analytic methods for estimating of the strength of association-relative risk (RR) and associated standard error of 53 drug–adverse event outcome pairs, both positive and negative controls. The methods were applied to a network of ten observational healthcare databases, comprising over 130 million lives. Performance measures included sensitivity, specificity, and positive predictive value of methods at RR thresholds achieving statistical significance of p 1.5, as well as threshold-free measures such as area under receiver operating characteristic curve (AUC). Results: Although no specific method demonstrated superior performance, the aggregate results provide a benchmark and baseline expectation for risk identification method performance. At traditional levels of statistical significance (RR > 1, p 18%, with positive predictive value

In sequential event prediction, we are given a "sequence database" of past event sequences to learn from, and we aim to predict the next event within a current event sequence. We focus on applications where the set of the past events has predictive power and not the specific order of those past events. Such applications arise in recommender systems, equipment maintenance, medical informatics, and in other domains. Our formalization of sequential event prediction draws on ideas from supervised ranking. We show how specific choices within this approach lead to different sequential event prediction problems and algorithms. In recommender system applications, the observed sequence of events depends on user choices, which may be influenced by the recommendations, which are themselves tailored to the user's choices. This leads to sequential event prediction algorithms involving a non-convex optimization problem. We apply our approach to an online grocery store recommender system, email recipient recommendation, and a novel application in the health event prediction domain.

We propose a statistical modeling technique, called the Hierarchical Association Rule Model (HARM), that predicts a patient’s possible future medical conditions given the patient’s current and past history of reported conditions. The core of our technique is a Bayesian hierarchical model for selecting predictive association rules (such as “condition 1 and condition 2 → condition 3”) from a large set of candidate rules. Because this method “borrows strength” using the conditions of many similar patients, it is able to provide predictions specialized to any given patient, even when little information about the patient’s history of conditions is available., National Science Foundation (U.S.) (NSF Grant IIS-10-53407), Google (Firm) (Ph.D. fellowship in statistics)

Data mining disproportionality methods (PRR, ROR, EBGM, IC, etc.) are commonly used to identify drug safety signals in spontaneous report system (SRS) databases. Newer data sources such as longitudinal observational databases (LOD) provide time-stamped patient-level information and overcome some of the SRS limitations such as an absence of the denominator, total number of patients who consume a drug, and limited temporal information. Application of the disproportionality methods to LODs has not been widely explored. The scale of the LOD data provides an interesting computational challenge. Larger health claims databases contain information on more than 50 million patients and each patient has records for up to 10 years. In this article we systematically explore the application of commonly used disproportionality methods to simulated and real LOD data.

To determine the association of high-risk alleles in the complement factor H (CFH; Y402H, rs1061170) and age-related maculopathy susceptibility (ARMS2; A69S, rs10490924) genes with reticular macular disease (RMD), a major clinical subphenotype of age-related macular degeneration (AMD).Using retinal images from the Columbia Macular Genetics Study, we identified 67 subject individuals with RMD. A comparison group of 64 subjects with AMD without RMD was matched by ethnicity, age, sex, and AMD clinical stage.In the RMD group, 53 of 67 subjects (79.1%) were female, the mean age was 83 years, and 47 of 67 (70.1%) had late AMD, with closely matched values in the non-RMD group. The frequencies of the CFH 402H allele were 39.6% in the RMD group (53 of 134 individuals) and 58.6% in the non-RMD group (75 of 128 individuals) (χ(2) = 8.8; P = .003; odds ratio, 0.46 [95% confidence interval, 0.28-0.76]). The corresponding frequencies of the risk allele for ARMS2 were 44.0% (40 of 128 individuals) and 31.3% (40 of 128 individuals), respectively (χ(2) = 4.0; P = .045; odds ratio, 1.73 [95% confidence interval, 1.04-2.90]). Homozygosity for 402H was particularly associated with the absence of RMD, occurring in 8 of 67 subjects (11.9%) with RMD vs 24 of 64 subjects (37.5%) without RMD (P.001). Retinal macular disease also was associated with hypertension among male patients.The AMD-associated CFH 402H risk variant is significantly associated with the absence of RMD but enhanced risk for RMD is conferred by the ARMS2 69S AMD risk allele. These results are consistent with the hypothesis that 402H may confer a survival benefit against certain infections, some of which may cause RMD.Reticular macular disease may be genetically distinct from the rest of AMD.

We present a framework and generalization analysis for the use of association rules in the setting of supervised learning. We are specifically interested in a sequential event prediction problem where data are revealed one by one, and the goal is to determine what will next be revealed. In the context of this problem, algorithms based on association rules have a distinct advantage over classical statistical and machine learning methods; however, to our knowledge there has not previously been a theoretical foundation established for using association rules in supervised learning. We present two simple algorithms that incorporate association rules. These algorithms can be used both for sequential event prediction and for supervised classification. We provide generalization guarantees on these algorithms based on algorithmic stability analysis from statistical learning theory. We include a discussion of the strict minimum support threshold often used in association rule mining, and introduce an "adjusted confidence" measure that provides a weaker minimum support condition that has advantages over the strict minimum support. The paper brings together ideas from statistical learning theory, association rule mining and Bayesian analysis.

In many healthcare settings, patients visit healthcare professionals periodically and report multiple medical conditions, or symptoms, at each encounter. We propose a statistical modeling technique, called the Hierarchical Association Rule Model (HARM), that predicts a patient’s possible future symptoms given the patient’s current and past history of reported symptoms. The core of our technique is a Bayesian hierarchical model for selecting predictive association rules (such as “symptom 1 and symptom 2 → symptom 3 ”) from a large set of candidate rules. Because this method “borrows strength” using the symptoms of many similar patients, it is able to provide predictions specialized to any given patient, even when little information about the patient’s history of symptoms is available.

Directed acyclic independence graphs (DAIGs) play an important role in recent developments in probabilistic expert systems and influence diagrams (Chyu [1]). The purpose of this note is to show that DAIGs can usefully be grouped into equivalence classes where the members of a single class share identical Markov properties. These equivalence classes can be identified via a simple graphical criterion. This result is particularly relevant to model selection procedures for DAIGs (see, e.g., Cooper and Herskovits [2] and Madigan and Raftery [4]) because it reduces the problem of searching among possible orientations of a given graph to that of searching among the equivalence classes.

OBJECTIVES: This study explored reproductive health care in rural Washington State, reasons given by providers for not offering abortions, and providers' willingness to use medical abortifacients. METHODS: Physicians, midwives, nurse practitioners, and physician assistants in rural Washington completed an inventory of reproductive health services that they provide, whether and why they do not perform abortions, and whether they would use medical abortifacients. RESULTS: Of the respondents, 89.2% reported providing reproductive health care. Only 1.2% reported performing surgical abortions, and 26.1% indicated that they would probably prescribe medical abortifacients. CONCLUSIONS: Few providers offer surgical abortions in rural Washington. Greater numbers report a willingness to prescribe medical abortifacients.

Accurate adjustment for injury severity is needed to evaluate the effectiveness of trauma management. While the choice of injury coding scheme used for modeling affects performance, the impact of combining coding schemes on performance has not been evaluated. The purpose of this study was to use Bayesian logistic regression to develop models predicting hospital mortality in injured children and to compare the performance of models developed using different injury coding schemes.Records of children (age15 years) admitted after injury were obtained from the National Trauma Data Bank (NTDB) and the National Pediatric Trauma Registry (NPTR) and used to train Bayesian logistic regression models predicting mortality using three injury coding schemes (International Classification of Disease-9th revision [ICD-9] injury codes, the Abbreviated Injury Scale [AIS] severity scores, and the Barell matrix) and their combinations. Model performance was evaluated using independent data from the NTDB and the Kids' Inpatient Database 2003 (KID).Discrimination was optimal when modeling both ICD-9 and AIS severity codes (area under the receiver operating curve [AUC] = 0.921 [NTDB] and 0.967 [KID], Hosmer-Lemeshow [HL] h-statistic = 115 [NTDB] and 147 [KID]), while calibration was optimal when modeling coding based on the Barell matrix (AUC = 0.882 [NTDB] and 0.936 [KID], HL h-statistic = 19 [NTDB] and 69 [KID]). When compared to models based on ICD-9 codes alone, models that also included AIS severity scores and coding from the Barell matrix showed improved discrimination and calibration.Mortality models that incorporate additional injury coding schemes perform better than those based on ICD-9 codes alone in the setting of pediatric trauma. Combining injury coding schemes may be an effective approach for improving the predictive performance of empirically derived estimates of injury mortality.

Purpose To compare the results of drug safety data mining with three different algorithms, when adverse events are identified using MedDRA 1 Preferred Terms (PT) vs. High Level Terms (HLT) vs. Standardised MedDRA Queries (SMQ). Methods For a representative set of 26 drugs, data from the FDA Adverse Event Reporting System (AERS) database from 2001 through 2005 was mined for signals of disproportionate reporting (SDRs) using three different data mining algorithms (DMAs): the Gamma Poisson Shrinker (GPS), the urn-model algorithm (URN), and the proportional reporting rate (PRR) algorithm. Results were evaluated using a previously described Reference Event Database (RED) which contains documented drug–event associations for the 26 drugs. Analysis emphasized the percentage of SDRs in the “unlabeled supported” category, corresponding to those adverse events that were not described in the U.S. prescribing information for the drug at the time of its approval, but which were supported by some published evidence for an association with the drug. Results Based on a logistic regression analysis, the percentage of unlabeled supported SDRs was smallest at the PT level, intermediate at the HLT level, and largest at the SMQ level, for all three algorithms. The GPS and URN methods detected comparable percentages of unlabeled supported SDRs while the PRR method detected a smaller percentage, at all three MedDRA levels. No evidence of a method/level interaction was seen. Conclusions Use of HLT and SMQ groupings can improve the percentage of unlabeled supported SDRs in data mining results. The trade-off for this gain is the medically less-specific language of HLTs and SMQs compared to PTs, and the need for the added step in data mining of examining the component PTs of each HLT or SMQ that results in a signal of disproportionate reporting.

The emergence of real-time ultrasonic imaging for extracorporeal shockwave lithotripsy poses questions regarding the factors and techniques which facilitate stone imaging for clinicians with no previous practical experience in ultrasonography. The ability of these clinicians to assess when stone disintegration has been achieved also needs to be confirmed. A wide range of data was recorded from each of 2688 lithotripsy treatments performed over a 2-year period using the EDAP LT.01 ultrasound-imaged piezoelectric lithotriptor. An analysis of these data was performed using a comprehensive microcomputer-based statistics package. The mean time taken for stone imaging and positioning was reduced from 11.2 to 7.5 min over the 2-year period. Obese patients and those with renal pelvic stones were best imaged in a lateral position. Overall there was no difference in percentage stone disintegration or clearance between treatments in the supine or lateral positions, but a significant reduction in the clearance of small caliceal stones resulted when the lateral position was used. Factors associated with a significantly greater percentage of stone disintegration and clearance included pain experienced by the patient during fine adjustment of the processing head during treatment, acoustic focus attenuation and widening and acoustic shadow widening as detected by the urologist at the end of treatment. Among the factors not associated with significant alterations in the percentage of stone disintegration or clearance were the lithotriptor operator, the side or site of the calculus, obesity and shockwave frequency or power. This study confirmed the ability of urologists to develop expertise in ultrasonography for renal stone imaging and to interpret successfully the subtle signs of stone disintegration.

SUMMARY Asmussen & Edwards (1983) defined necessary and sufficient conditions for collapsibility of a hierarchical log linear model for a multidimensional contingency table. We have shown that for decomposable log linear models these conditions can be combined with various graph-theoretic algorithms to provide useful classes of sub-tables which are collapsible onto. In particular, the SAHR algorithm finds the minimal set onto which the model can be collapsed and which contains a sub-table of interest. In the context of expert systems, by reducing a probabilistic influence network onto only the relevant nodes, the algorithms reduce the required computation and simplify interpretation.

Following work of Stroud and Saeger and Anand et al., we formulate a port of entry inspection sequencing task as a problem of finding an optimal binary decision tree for an appropriate Boolean decision function. We report on new algorithms that are more efficient computationally than those presented by Stroud and Saeger and Anand et al. We achieve these efficiencies through a combination of specific numerical methods for finding optimal thresholds for sensor functions and a novel binary decision tree search algorithm that operates on a space of potentially acceptable binary decision trees.

The objective of this report is to provide a basis to inform decisions about priorities for developing statistical research initiatives in the field of public health surveillance for emerging threats. Rapid information system advances have created a vast opportunity of secondary data sources for information to enhance the situational and health status awareness of populations. While the field of medical informatics and initiatives to standardize healthcare-seeking encounter records continue accelerating, it is necessary to adapt analytic and statistical methodologies to mature in sync with sibling information science technologies. One major right-of-passage for statistical inference is to advance the optimal application of analytic methodologies for using multiple data streams in detecting and characterizing public health population events of importance. This report first describes the problem in general and the data context, then delineates more specifically the practical nature of the problem and the related issues. Approaches currently applied to data with time-series, statistical process control and traditional inference concepts are described with examples in the section on Statistics and the Role of the Analytic Surveillance Data Monitor. These are the techniques that are providing substance to surveillance professionals and enabling use of multiple data streams. The next section describes use of a more complex approach that takes temporal as well as spatial dimensions into consideration for detection and situational awareness regarding event distributions. The space-time statistic has successfully been used to detect and track public health events of interest. Important research questions which are summarized at the end of this report are described in more detail with respect to the methodological application in the respective sections. This was thought to help elucidate the research requirements as summarized later in the report. Following the description of the space-time scan statistical application; this report extends to a less traditional area of promise given what has been observed in recent application of analytic methods. Bayesian networks (BNs) represent a conceptual step with advantages of flexibility for the public health surveillance community. Progression from traditional to the more extending statistical concepts in the context of the dynamic status quo of responsibility and challenge, leads to a conclusion consisting of categorical research needs. The report is structured by design to inform judgment about how to build on practical systems to achieve better analytic outcomes for public health surveillance. There are references to research issues throughout the sections with a summarization at the end, which also includes items previously unmentioned in the report.

We present a class of sparse generalized linear models that include probit and logistic regression as special cases and offer some extra flexibility. We provide an EM algorithm for learning the parameters of these models from data. We apply our method in text classification and in simulated data and show that our method outperforms the logistic and probit models and also the elastic net, in general by a substantial margin., Comment: Published at http://dx.doi.org/10.1214/074921707000000067 in the IMS Lecture Notes Monograph Series (http://www.imstat.org/publications/lecnotes.htm) by the Institute of Mathematical Statistics (http://www.imstat.org)

Classification problems with functionally structured in- put variables arise naturally in many applications. In a clinical domain, for example, input variables could include a time series of blood pressure measurements. In a financial setting, different time series of stock returns might serve as predictors. In an archaeological application, the 2-D pro- file of an artifact may serve as a key input variable. In such domains, accuracy of the classifier is not the only reason- able goal to strive for; classifiers that provide easily inter- pretable results are also of value. In this work, we present an intuitive scheme for extending decision trees to handle functional input variables. Our results show that such deci- sion trees are both accurate and readily interpretable.

For Bayesian analysis of massive data, Markov chain Monte Carlo (MCMC) techniques often prove infeasible due to computational resource constraints. Standard MCMC methods generally require a complete scan of the dataset for each iteration. Ridgeway and Madigan (2002) and Chopin (2002b) recently presented importance sampling algorithms that combined simulations from a posterior distribution conditioned on a small portion of the dataset with a reweighting of those simulations to condition on the remainder of the dataset. While these algorithms drastically reduce the number of data accesses as compared to traditional MCMC, they still require substantially more than a single pass over the dataset. In this paper, we present "1PFS," an efficient, one-pass algorithm. The algorithm employs a simple modification of the Ridgeway and Madigan (2002) particle filtering algorithm that replaces the MCMC based "rejuvenation" step with a more efficient "shrinkage" kernel smoothing based step. To show proof-of-concept and to enable a direct comparison, we demonstrate 1PFS on the same examples presented in Ridgeway and Madigan (2002), namely a mixture model for Markov chains and Bayesian logistic regression. Our results indicate the proposed scheme delivers accurate parameter estimates while employing only a single pass through the data.

A principle concern of pharmacovigilance is the timely detection of adverse drug reactions that are novel by virtue of their clinical nature, severity and/or frequency. The cornerstone of this process is the scientific acumen of the pharmacovigilance domain expert. There is understandably an interest in developing database screening tools to assist human reviewers in identifying associations worthy of further investigation (i.e., signals) embedded within a database consisting largely of background 'noise' containing reports of no substantial public health significance. Data mining algorithms are, therefore, being developed, tested and/or used by health authorities, pharmaceutical companies and academic researchers. After a focused review of postapproval drug safety signal detection, the authors explain how the currently used algorithms work and address key questions related to their validation, comparative performance, deployment in naturalistic pharmacovigilance settings, limitations and potential for misuse. Suggestions for further research and development are offered.

Motivated by high‐dimensional applications in authorship attribution, we describe a Bayesian multinomial logistic regression model together with an associated learning algorithm.

Failure to adhere to complex antiretroviral regimens can lead to resistance and treatment failure among HIV-positive persons. In this study of the feasibility of an automated two-way messaging system to improve adherence, participants received multiple short daily messages designed to remind, educate, encourage adherence, and solicit responses concerning side effects and self-reported adherence. Twenty-five participants remained in the study for a median of 208 days, receiving 17,440 messages and replying to 14,677 (84%). Participants reported missing one or more doses on 36% of 743 queries and reported medication side effects on 26% of 729 queries. Participants expressed high satisfaction with the messaging system and reported that it helped with medication adherence. The study suggests that it is feasible to use an automated wireless two-way messaging system to communicate with HIV-positive patients over an extended period of time.

Correction of table 2 on page 1757 of 'Separation and completeness properties for AMP chain graph Markov models', Annals of Statistics, volume 29 (2001).

There are massive numbers of aviation inspection reports collected each year in the USA. These reports record findings from aviation surveillance inspections as well as accident or incident investigations. The goal of this paper is to apply text classification to the mining of these reports, and to show that the text classification methodology can be a critical element of the aviation safety decision support system. The performances of several text classification models are evaluated in the context of mining aviation inspection reports. The evaluation is given in terms of misclassification rates. Further breakdowns of the misclassification rates and related findings from the dataset suggest ways for improving data quality and for gathering information which are more pertinent for filing inspection reports.

One of the main objectives in pharmacovigilance is the detection of adverse drug events (ADEs) through mining of healthcare databases, such as electronic health records or administrative claims data. Although different approaches have been shown to be of great value, research is still focusing on the enhancement of signal detection to gain efficiency in further assessment and follow-up. We applied similarity-based modeling techniques, using 2D and 3D molecular structure, ADE, target, and ATC (anatomical therapeutic chemical) similarity measures, to the candidate associations selected previously in a medication-wide association study for four ADE outcomes. Our results showed an improvement in the precision when we ranked the subset of ADE candidates using similarity scorings. This method is simple, useful to strengthen or prioritize signals generated from healthcare databases, and facilitates ADE detection through the identification of the most similar drugs for which ADE information is available.

Squashing is a lossy data compression technique that preserves statistical information. Specifically, squashing compresses a massive dataset to a much smaller one so that outputs from statistical analyses carried out on the smaller (squashed) dataset reproduce outputs from the same statistical analyses carried out on the original dataset. Likelihood-based data squashing (LDS) differs from a previously published squashing algorithm insofar as it uses a statistical model to squash the data. The results show that LDS provides excellent squashing performance even when the target statistical analysis departs from the model used to squash the data.

We use data on Washington State abortions and births for 1983-1984 and 1993-1994 to analyze trends for urban and rural women, using the demographic measures total abortion and total fertility rates. These express pacing of childbearing in a single number which is simple to calculate and interpret, and is age-standardized. We find significant urban-rural differences. Total abortion rates decline and total fertility rates increase in both areas. However, the relative magnitudes of pacing decreases in abortions for rural women and increases in births for urban women are striking. The demographic measures are useful interpretive tools, and can be applied to a broad range of questions.

A printing malfunction caused all minus signs and some left parentheses to be omitted from the paper “Bayesian Model Averaging:A Tutorial ” by Jennifer A. Hoeting, David Madigan, Adrian E. Raftery and Chris T. Volinsky in the November 1999 issue of Statistical Science (volume 14, pages 382–417).These errors occurred after the proof stage and were not the fault of the au- thors. Corrections to the paper are listed below. A corrected version of the paper is also available at http://www.stat.washington.edu/www/research/online/hoeting1999.pdf.Please cite this article as follows: Hoeting, J. A., Madigan, D., Raftery, A. E. and Volinsky, C. T. (1999) “Bayesian Model Averaging: A Tutorial (with discussion)” Statistical Science 14:4, 382–417. Corrected version available at http://www.stat.washington.edu/www/research/online/hoeting1999.pdf.

Standard statistical practice ignores model uncertainty. Data analysts typically select a model from some class of models and then proceed as if the selected model had generated the data. This approach ignores the uncertainty in model selection, leading to over-confident inferences and decisions that are more risky than one thinks they are. Bayesian model averaging (BMA)provides a coherent mechanism for accounting for this model uncertainty. Several methods for implementing BMA have recently emerged. We discuss these methods and present a number of examples.In these examples, BMA provides improved out-of-sample predictive performance. We also provide a catalogue of currently available BMA software.

Undirected graphs and acyclic digraphs (ADGs), as well as their mutual extension to chain graphs, are widely used to describe dependencies among variables in multivariate distributions. In particular, the likelihood functions of ADG models admit convenient recursive factorizations that often allow explicit maximum likelihood estimates and that are well suited to building Bayesian networks for expert systems. Whereas the undirected graph associated with a dependence model is uniquely determined, there may, however, be many ADGs that determine the same dependence (= Markov) model. Thus, the family of all ADGs with a given set of vertices is naturally partitioned into Markov-equivalence classes, each class being associated with a unique statistical model. Statistical procedures, such as model selection or model averaging, that fail to take into account these equivalence classes, may incur substantial computational or other inefficiencies. Here it is shown that each Markov-equivalence class is uniquely determined by a single chain graph, the essential graph, that is itself simultaneously Markov equivalent to all ADGs in the equivalence class. Essential graphs are characterized, a polynomial-time algorithm for their construction is given, and their applications to model selection and other statistical questions are described.