LocalControl: An R Package for Comparative Safety and Effectiveness Research.

The LocalControl R package implements novel approaches to address biases and confounding when comparing treatments or exposures in observational studies of outcomes. While designed and appropriate for use in comparative safety and effectiveness research involving medicine and the life sciences, the package can be used in other situations involving outcomes with multiple confounders. LocalControl is an open-source tool for researchers whose aim is to generate high quality evidence using observational data. The package implements a family of methods for non-parametric bias correction when comparing treatments in observational studies, including survival analysis settings, where competing risks and/or censoring may be present. The approach extends to bias-corrected personalized predictions of treatment outcome differences, and analysis of heterogeneity of treatment effect-sizes across patient subgroups.

Low-level radon exposure and lung cancer mortality.

BACKGROUND: It is agreed that high level radon exposure is harmful to humans. However, some published literature suggests that low levels of radon show no adverse effects or may even be protective. Claims made using traditional methods of analysis on observational data often fail to replicate. Here, we use a simple, alternative data-analytic strategy for examining effects of low-level indoor radon exposure on lung cancer mortality. One objective will be to demonstrate that local population characteristics can alter expected effects. METHODS: Observational data on indoor radon exposure levels and lung cancer mortality for 2881 U.S. counties were obtained from federal and state governmental agencies. A new "statistical thinking" step-by-step analysis strategy called Local Control (LC) allows us to perform analyses of observational data that are more objective and "fair" than regression-like methods. LC analytical strategy makes as few and as realistic assumptions as possible. As a result, key LC inferences are nonparametric, and estimates of potentially heterogeneous treatment effect-sizes are robust. RESULTS: Our LC analyses suggest that lung cancer mortality usually tends to decrease as background radon exposure increases. Local rank correlation (LRC) effect-sizes are shown to be predictable from confounding local characteristics like percentage of residents over 65, percentage of residents who currently smoke and percentage of obese residents. CONCLUSIONS: At low indoor radon exposure levels, reverse (negative) LRCs between radon exposure level and lung cancer mortality predominate. The strengths of these associations vary with local demographics.

The reliability of an environmental epidemiology meta-analysis, a case study.

Claims made in science papers are coming under increased scrutiny with many claims failing to replicate. Meta-analyses are questionable when based on data from observational studies which are often unreliable. We examine the reliability of the base studies used in an air quality/heart attack meta-analysis and the resulting meta-analysis. A meta-analysis study that includes 14 observational air quality/heart attack studies is examined for its statistical reliability. We use simple counting to evaluate the reliability of the base papers and a p-value plot of the p-values from the base studies to examine study heterogeneity. We find that the base papers have the potential for massive multiple testing and multiple modeling with no statistical adjustments. Statistics coming from the base papers are not guaranteed to be unbiased, a requirement for a valid meta-analysis. There is study heterogeneity for the base papers with strong evidence for so called p-hacking in some, possibly many, of the studies. We make two observations: there are many claims at issue in each of the 14 base studies so uncorrected multiple testing is a serious issue. We find that some of the base papers exhibit the characteristics of p-hacking and are therefore not reliable; the resulting meta-analysis is not reliable.

PM2.5 and ozone, indicators of air quality, and acute deaths in California, 2004-2007.

Since the London Great Smog of 1952 was estimated to have killed over 4000 people, scientists have studied the relationship between air quality and acute mortality. Currently, the association between air quality and acute deaths is usually taken as evidence for causality. As air quality has markedly improved since 1952, do contemporary datasets support this view? We use a large dataset, eight air basins in California for the years 2004-2007, to examine the possible association of ozone and PM2.5 with acute deaths after statistically removing seasonal and weather effects. Our analysis dataset is available on request. We conducted a regression-corrected, case-crossover analysis for all non-accidental deaths age 75 and older. We used stepwise regression to examine three causes of death. After seasonal and weather adjustments, there was essentially no predictive power of ozone or PM2.5 for acute deaths. The case-crossover analysis produced odds ratio very close to 1.000 (no effect). The very narrow confidence limits indicated good statistical power. We study recent air quality in both time-stratified, symmetric, bidirectional case-crossover and time series regression and both give consistent results. There is no statistically significant association between either ozone or PM2.5 and acute human mortality. In the absence of an association, causality is in question.

Systemic challenges in bipolar disorder management: A patient-centered approach.

OBJECTIVES: As part of a series of Patient-Centered Outcomes Research Institute-funded large-scale retrospective observational studies on bipolar disorder (BD) treatments and outcomes, we sought the input of patients with BD and their family members to develop research questions. We aimed to identify systemic root causes of patient-reported challenges with BD management in order to guide subsequent studies and initiatives. METHODS: Three focus groups were conducted where patients and their family members (total n = 34) formulated questions around the central theme, "What do you wish you had known in advance or over the course of treatment for BD?" In an affinity mapping exercise, participants clustered their questions and ranked the resulting categories by importance. The research team and members of our patient partner advisory council further rated the questions by expected impact on patients. Using a Theory of Constraints systems thinking approach, several causal models of BD management challenges and their potential solution were developed with patients using the focus group data. RESULTS: A total of 369 research questions were mapped to 33 categories revealing 10 broad themes. The top priorities for patient stakeholders involved pharmacotherapy and treatment alternatives. Analysis of causal relationships underlying 47 patient concerns revealed two core conflicts: for patients, whether or not to take pharmacotherapy, and for mental health services, the dilemma of care quality vs quantity. CONCLUSIONS: To alleviate the core conflicts identified, BD management requires a coordinated multidisciplinary approach including: improved access to mental health services, objective diagnostics, sufficient provider visit time, evidence-based individualized treatment, and psychosocial support.

Air quality environmental epidemiology studies are unreliable.

Ever since the London Great Smog of 1952 is estimated to have killed over 4000 people, scientists have studied the relationship between air quality and acute mortality. There are many hundreds of papers examining the question. There is a serious statistical problem with most of these papers. If there are many questions under consideration, and there is no adjustment for multiple testing or multiple modeling, then unadjusted p-values are totally unreliable making claims unreliable. Our idea is to determine the statistical reliability of eight papers published in Environmental Health Perspectives that were used in meta-analysis papers appearing in Lancet and JAMA. We counted the number of outcomes, air quality predictors, time lags and covariates examined in each paper. We estimate the multiplicity of questions that could be asked and the number of models that could be constructed. The results were that the median numbers of comparisons possible for multiplicity, models and search space were 135, 128, and 9568 respectively. Given the large search spaces, finding a small number of nominally significant results is not unusual at all. The claims in these eight papers are not statistically supported so these papers are unreliable as are the meta-analysis papers that use them.

Air quality and acute deaths in California, 2000-2012.

Many studies have shown an association between air quality and acute deaths, and such associations are widely interpreted as causal. Several factors call causation and even association into question, for example multiple testing and multiple modeling, publication bias and confirmation bias. Many published studies are difficult or impossible to reproduce because of lack of access to confidential data sources. Here we make publically available a dataset containing daily air quality levels, PM2.5 and ozone, daily temperature levels, minimum and maximum and daily maximum relative humidity levels for the eight most populous California air basins, thirteen years, >2M deaths, over 37,000 exposure days. The data are analyzed using standard time series analysis, and a sensitivity analysis is computed varying model parameters, locations and years. Our analysis finds little evidence for association between air quality and acute deaths. These results are consistent with those for the widely cited NMMAPS dataset when the latter are restricted to California. The daily death variability was mostly explained by time of year or weather variables; Neither PM2.5 nor ozone added appreciably to the prediction of daily deaths. These results call into question the widespread belief that association between air quality and acute deaths is causal/near-universal.

Local Control Strategy: Simple Analyses of Air Pollution Data Can Reveal Heterogeneity in Longevity Outcomes.

Claims from observational studies that use traditional model specification searches often fail to replicate, partially because the available data tend to be biased. There is an urgent need for an alternative statistical analysis strategy, that is not only simple and easily understood but also is more likely to give reliable insights when the available data have not been designed and balanced. The alternative strategy known as local control first generates local, nonparametric effect-size estimates (fair treatment comparisons) and only then asks whether the observed variation in these local estimates can be predicted from potential confounding factors. Here, we illustrate application of local control to a historical air pollution data set describing a "natural experiment" initiated by the federal Clean Air Act Amendments of 1970. Our reanalysis reveals subgroup heterogeneity in the effects of air quality regulation on elderly longevity (one size does not fit all), and we show that this heterogeneity is largely explained by socioeconomic and environmental confounders other than air quality.

Hypothyroidism risk compared among nine common bipolar disorder therapies in a large US cohort.

OBJECTIVES: Thyroid abnormalities in patients with bipolar disorder (BD) have been linked to lithium treatment for decades, yet other drugs have been less well studied. Our objective was to compare hypothyroidism risk for lithium versus the anticonvulsants and second-generation antipsychotics commonly prescribed for BD. METHODS: Administrative claims data on 24,574 patients with BD were analyzed with competing risk survival analysis. Inclusion criteria were (i) one year of no prior hypothyroid diagnosis nor BD drug treatment, (ii) followed by at least one thyroid test during BD monotherapy on lithium carbonate, mood-stabilizing anticonvulsants (lamotrigine, valproate, oxcarbazepine, or carbamazepine) or antipsychotics (aripiprazole, olanzapine, risperidone, or quetiapine). The outcome was cumulative incidence of hypothyroidism per drug, in the presence of the competing risk of ending monotherapy, adjusted for age, sex, physician visits, and thyroid tests. RESULTS: Adjusting for covariates, the four-year cumulative risk of hypothyroidism for lithium (8.8%) was 1.39-fold that of the lowest risk therapy, oxcarbazepine (6.3%). Lithium was non-statistically significantly different from quetiapine. While lithium conferred a higher risk when compared to all other treatments combined as a group, hypothyroidism risk error bars overlapped for all drugs. Treatment (p = 3.86e-3), age (p = 6.91e-10), sex (p = 3.93e-7), and thyroid testing (p = 2.79e-87) affected risk. Patients taking lithium were tested for hypothyroidism 2.26-3.05 times more frequently than those on other treatments. CONCLUSIONS: Thyroid abnormalities occur frequently in patients with BD regardless of treatment. Therefore, patients should be regularly tested for clinical or subclinical thyroid abnormalities on all therapies and treated as indicated to prevent adverse effects of hormone imbalances on mood.

Power and sample size calculation for microarray studies.

Microarray is a technology to screen a large number of genes to discover those differentially expressed between clinical subtypes or different conditions of human diseases. Gene discovery using microarray data requires adjustment for the large-scale multiplicity of candidate genes. The family-wise error rate (FWER) has been widely chosen as a global type I error rate adjusting for the multiplicity. Typically in microarray data, the expression levels of different genes are correlated because of coexpressing genes and the common experimental conditions shared by the genes on each array. To accurately control the FWER, the statistical testing procedure should appropriately reflect the dependency among the genes. Permutation methods have been used for accurate control of the FWER in analyzing microarray data. It is important to calculate the required sample size at the design stage of a new (confirmatory) microarray study. Because of the high dimensionality and complexity of the correlation structure in microarray data, however, there have been no sample size calculation methods accurately reflecting the true correlation structure of real microarray data. We propose sample size and power calculation methods that are useful when pilot data are available to design a confirmatory experiment. If no pilot data are available, we recommend a two-stage sample size recalculation based on our proposed method using the first stage data as pilot data. The calculated sample sizes are shown to accurately maintain the power through simulations. A real data example is taken to illustrate the proposed method.

ChemModLab: a web-based cheminformatics modeling laboratory.

ChemModLab, written by the ECCR @ NCSU consortium under NIH support, is a toolbox for fitting and assessing quantitative structure-activity relationships (QSARs). Its elements are: a cheminformatic front end used to supply molecular descriptors for use in modeling; a set of methods for fitting models; and methods for validating the resulting model. Compounds may be input as structures from which standard descriptors will be calculated using the freely available cheminformatic front end PowerMV; PowerMV also supports compound visualization. In addition, the user can directly input their own choices of descriptors, so the capability for comparing descriptors is effectively unlimited. The statistical methodologies comprise a comprehensive collection of approaches whose validity and utility have been accepted by experts in the fields. As far as possible, these tools are implemented in open-source software linked into the flexible R platform, giving the user the capability of applying many different QSAR modeling methods in a seamless way. As promising new QSAR methodologies emerge from the statistical and data-mining communities, they will be incorporated in the laboratory. The web site also incorporates links to public-domain data sets that can be used as test cases for proposed new modeling methods. The capabilities of ChemModLab are illustrated using a variety of biological responses, with different modeling methodologies being applied to each. These show clear differences in quality of the fitted QSAR model, and in computational requirements. The laboratory is web-based, and use is free. Researchers with new assay data, a new descriptor set, or a new modeling method may readily build QSAR models and benchmark their results against other findings. Users may also examine the diversity of the molecules identified by a QSAR model. Moreover, users have the choice of placing their data sets in a public area to facilitate communication with other researchers; or can keep them hidden to preserve confidentiality.

Systematic decrements in QTc between the first and second day of contiguous daily ECG recordings under controlled conditions.

BACKGROUND: Many thorough QT (TQT) studies use a baseline day and double delta analysis to account for potential diurnal variation in QTc. However, little is known about systematic changes in the QTc across contiguous days when normal volunteers are brought into a controlled inpatient environment. METHODS: Two separate crossover TQT studies included 2 days of no treatment lead-in days with ECG collection preceding periods of drug treatment . In the first study, there were two pairs of such contiguous days with 10 replicate electrocardiograms (ECGs) collected at six time points, and in the second study, there were four pairs of contiguous days with nine replicate ECGs collected at five time points. These lead-in day pairs provided the opportunity to evaluate any systematic changes across contiguous first and second days of an inpatient environment. Within-patient consistency of change across pairs of days as well as within day, diurnal variation could also be evaluated. RESULTS: Modest (4.2 ms [range 1.9-6.5 ms]) but consistent decreases (significant [P < 0.05] for all 32 comparisons) were observed (probability: ≤5.4 × 10(-16)). Although group behavior with respect to QTc was consistent, individual subjects demonstrated substantial variability across pairs of days. Evidence of diurnal variation was weak and inconsistent. Magnitude of any diurnal variation was less than magnitude of change across days. CONCLUSIONS: Subjects show a systematic decrease in QTc from first day to second day of inpatient status and do not demonstrate a significant diurnal pattern. The magnitude of this systematic change is sufficient to influence QTc study interpretation.

Assessing Methods for Evaluating the Number of Components in Non-Negative Matrix Factorization.

Non-negative matrix factorization is a relatively new method of matrix decomposition which factors an m×n data matrix X into an m×k matrix W and a k×n matrix H, so that X≈W×H. Importantly, all values in X, W, and H are constrained to be non-negative. NMF can be used for dimensionality reduction, since the k columns of W can be considered components into which X has been decomposed. The question arises: how does one choose k? In this paper, we first assess methods for estimating k in the context of NMF in synthetic data. Second, we examine the effect of normalization on this estimate's accuracy in empirical data. In synthetic data with orthogonal underlying components, methods based on PCA and Brunet's Cophenetic Correlation Coefficient achieved the highest accuracy. When evaluated on a well-known real dataset, normalization had an unpredictable effect on the estimate. For any given normalization method, the methods for estimating k gave widely varying results. We conclude that when estimating k, it is best not to apply normalization. If underlying components are known to be orthogonal, then Velicer's MAP or Minka's Laplace-PCA method might be best. However, when orthogonality of the underlying components is unknown, none of the methods seemed preferable.

Clinically approved heterocyclics act on a mitochondrial target and reduce stroke-induced pathology.

Substantial evidence indicates that mitochondria are a major checkpoint in several pathways leading to neuronal cell death, but discerning critical propagation stages from downstream consequences has been difficult. The mitochondrial permeability transition (mPT) may be critical in stroke-related injury. To address this hypothesis, identify potential therapeutics, and screen for new uses for established drugs with known toxicity, 1,040 FDA-approved drugs and other bioactive compounds were tested as potential mPT inhibitors. We report the identification of 28 structurally related drugs, including tricyclic antidepressants and antipsychotics, capable of delaying the mPT. Clinically achievable doses of one drug in this general structural class that inhibits mPT, promethazine, were protective in both in vitro and mouse models of stroke. Specifically, promethazine protected primary neuronal cultures subjected to oxygen-glucose deprivation and reduced infarct size and neurological impairment in mice subjected to middle cerebral artery occlusion/reperfusion. These results, in conjunction with new insights provided to older studies, (a) suggest a class of safe, tolerable drugs for stroke and neurodegeneration; (b) provide new tools for understanding mitochondrial roles in neuronal cell death; (c) demonstrate the clinical/experimental value of screening collections of bioactive compounds enriched in clinically available agents; and (d) provide discovery-based evidence that mPT is an essential, causative event in stroke-related injury.