Bayesian Methods for Prevention Research.

Bayesian statistics represents a paradigm shift in statistical reasoning and an approach to analysis that is applicable to prevention trials with small samples. This paper introduces the reader to the philosophy behind Bayesian statistics. This introduction is followed by a review of some issues that arise in sampling statistics and how Bayesian methods address them. Finally, the article provides an extended illustration of the application of Bayesian statistics to data from a prevention trial that tested a family-focused intervention.

Understanding intra-neighborhood patterns in PM2.5 and PM10 using mobile monitoring in Braddock, PA.

BACKGROUND: Braddock, Pennsylvania is home to the Edgar Thomson Steel Works (ETSW), one of the few remaining active steel mills in the Pittsburgh region. An economically distressed area, Braddock exceeds average annual (>15 µg/m3) and daily (>35 µg/m3) National Ambient Air Quality Standards (NAAQS) for particulate matter (PM2.5). METHODS: A mobile air monitoring study was designed and implemented in morning and afternoon hours in the summer and winter (2010-2011) to explore the within-neighborhood spatial and temporal (within-day and between-day) variability in PM2.5 and PM10. RESULTS: Both pollutants displayed spatial variation between stops, and substantial temporal variation within and across study days. For summer morning sampling runs, site-specific mean PM2.5 ranged from 30.0 (SD = 3.3) to 55.1 (SD = 13.0) µg/m3. Mean PM10 ranged from 30.4 (SD = 2.5) to 69.7 (SD = 51.2) µg/m3, respectively. During summer months, afternoon concentrations were significantly lower than morning for both PM2.5 and PM10, potentially owing to morning subsidence inversions. Winter concentrations were lower than summer, on average, and showed lesser diurnal variation. Temperature, wind speed, and wind direction predicted significant variability in PM2.5 and PM10 in multiple linear regression models. CONCLUSIONS: Data reveals significant morning versus afternoon variability and spatial variability in both PM2.5 and PM10 concentrations within Braddock. Information obtained on peak concentration periods, and the combined effects of industry, traffic, and elevation in this region informed the design of a larger stationary monitoring network.

Identifying likely PM2.5 sources on days of elevated concentration: a simple statistical approach.

A simple statistical method is described for identifying the likely importance of local sources of PM2.5 in a region on days when the National Ambient Air Quality Standard is exceeded. The method requires only PM2.5 mass concentration and wind direction data, and makes use of the EPA database on PM2.5 emissions in the local region of interest. The method has been illustrated using data from the Pittsburgh Air Quality Study, and suggests that local sources can be very important in affecting PM2.5 exceedances. The results have implications for many of the urban areas in the eastern United States downwind of large sources in the Midwest, and shows that simple statistical tests can be of value in identifying regions where further testing with sophisticated air quality dispersion models and source-receptor models is warranted.

Optimal Bayesian design for patient selection in a clinical study.

Bayesian experimental design for a clinical trial involves specifying a utility function that models the purpose of the trial, in this case the selection of patients for a diagnostic test. The best sample of patients is selected by maximizing expected utility. This optimization task poses difficulties due to a high-dimensional discrete design space and, also, to an expected utility formula of high complexity. A simulation-based optimal design method is feasible in this case. In addition, two deterministic algorithms that perform a systematic search over the design space are developed to address the computational issues.

Adjusting for verification bias in diagnostic test evaluation: a Bayesian approach.

Obtaining accurate estimates of the performance of a diagnostic test for some population of patients might be difficult when the sample of subjects used for this purpose is not representative for the whole population. Thus, in the motivating example of this paper a test is evaluated by comparing its results with those given by a gold standard procedure, which yields the disease status verification. However, this procedure is invasive and has a non-negligible risk of serious complications. Moreover, subjects are selected to undergo the gold standard based on some risk factors and the results of the test under study. The test performance estimates based on the selected sample of subjects are biased. This problem was presented in previous studies under the name of verification bias. The current paper introduces a Bayesian method to adjust for this bias, which can be regarded as a missing data problem. In addition, it addresses the case of non-ignorable verification bias. The proposed Bayesian estimation approach provides test performance estimates that are consistent with the results obtained using likelihood-based approach. In addition, the paper studies how valuable the statistical findings are from the perspective of clinical decision making.

Comment on "Phylogenetic MCMC algorithms are misleading on mixtures of trees".

Mossel and Vigoda (Reports, 30 September 2005, p. 2207) show that nearest neighbor interchange transitions, commonly used in phylogenetic Markov chain Monte Carlo (MCMC) algorithms, perform poorly on mixtures of dissimilar trees. However, the conditions leading to their results are artificial. Standard MCMC convergence diagnostics would detect the problem in real data, and correction of the model misspecification would solve it.

Diabetes case identification methods applied to electronic medical record systems: their use in HIV-infected patients.

OBJECTIVE: New onset diabetes mellitus type 2 is increasing among HIV-infected patients in the era of potent antiretroviral therapy. Accurately identifying HIV-infected patients with a diagnosis of diabetes in electronic medical record systems will facilitate the study of patients with this disease. STUDY DESIGN AND SETTING: We examined electronic medical record data for all patients who initiated care at an HIV clinic between 1/1/1997 and 12/31/2001 to identify potential cases of diabetes. Case identification methods included clinician-coded diagnoses, medications, and HbA1c and glucose levels. Diabetes diagnoses were verified by clinician documentation in an electronic medical record progress note. Test characteristics of each case identification method were calculated. RESULTS: 53 cases of diabetes were identified among the cohort of 1,441 patients. Use of clinician-coded diagnoses alone or combined with other methods was the most sensitive method for identifying diabetes cases. Clinician-coded diagnoses were also the best method as assessed by standard receiver operator characteristic plots. A significant attenuation of odds ratios for associations with diabetes were found for case identification methods with imperfect specificity such as serum glucose levels. CONCLUSIONS: This study demonstrates that electronic medical record data can be used to accurately identify HIV-infected patients with diabetes. The optimal method applied will depend on the goals of a particular study.

A Bayesian approach to the estimation of ancestral genome arrangements.

We describe a Bayesian approach to estimate phylogeny and ancestral genome arrangements on the basis of genome arrangement data using a model in which gene inversion is the sole mechanism of change. While we have described a similar method to estimate phylogenetic relationships in the statistics literature, the novel contribution of the present work is the description of a method to compute probability distributions of ancestral genome arrangements. We assess the robustness of posterior distributions to different specifications of prior distributions and provide an empirical means to selecting a prior distribution. We note that parsimony approaches to ancestral reconstruction in the literature focus on the development of computationally efficient algorithms for searching for optimal ancestral genome arrangements, but, unlike Bayesian approaches, do not include assessment of uncertainty in these estimates. We compare and contrast a Bayesian approach with a parsimony approach to infer phylogenies and ancestral arrangements from genome arrangement data by re-analyzing a number of previously published data sets.

Bayesian methods for health-related decision making.

This article reviews the basics of Bayesian decision theory, and comments on its use in medical decision making. It emphasizes the subjectivity of the probability and utility inputs, and the desirability, in certain contexts, of representing several decision makers, each with his or her own probabilities and utilities. Applications and ethical considerations are also discussed. A brief bibliography gives pointers to the literature.

A bayesian analysis of metazoan mitochondrial genome arrangements.

Genome arrangements are a potentially powerful source of information to infer evolutionary relationships among distantly related taxa. Mitochondrial genome arrangements may be especially informative about metazoan evolutionary relationships because (1) nearly all animals have the same set of definitively homologous mitochondrial genes, (2) mitochondrial genome rearrangement events are rare relative to changes in sequences, and (3) the number of possible mitochondrial genome arrangements is huge, making convergent evolution of genome arrangements appear highly unlikely. In previous studies, phylogenetic evidence in genome arrangement data is nearly always used in a qualitative fashion-the support in favor of clades with similar or identical genome arrangements is considered to be quite strong, but is not quantified. The purpose of this article is to quantify the uncertainty among the relationships of metazoan phyla on the basis of mitochondrial genome arrangements while incorporating prior knowledge of the monophyly of various groups from other sources. The work we present here differs from our previous work in the statistics literature in that (1) we incorporate prior information on classifications of metazoans at the phylum level, (2) we describe several advances in our computational approach, and (3) we analyze a much larger data set (87 taxa) that consists of each unique, complete mitochondrial genome arrangement with a full complement of 37 genes that were present in the NCBI (National Center for Biotechnology Information) database at a recent date. In addition, we analyze a subset of 28 of these 87 taxa for which the non-tRNA mitochondrial genomes are unique where the assumption of our inversion-only model of rearrangement is more plausible. We present summaries of Bayesian posterior distributions of tree topology on the basis of these two data sets.

Placebos that harm: sham surgery controls in clinical trials.

Recent debates over the use of sham surgery as a control for studies of fetal tissue transplantation for Parkinson's disease have focused primarily on rival interpretations of the US federal regulations governing human-subjects research. Using the core ethical and methodological considerations that underwrite the equipoise requirement, we find strong prima facie reasons against using sham surgery as a control in studies of cellular-based therapies for Parkinson's disease and more broadly in clinical research. Additionally, we believe that these reasons can be generalized to apply to the use of other placebo controls that carry significant risks of positive harms in and of themselves. As a result, our arguments are centrally relevant to the emerging drive to subject therapies with a surgical component to the same rigorous standards of evaluation as those governing the approval of new pharmaceuticals.