Evaluating different strategies for estimating treatment effects in observational studies

ABSTRACT

Since the introduction of the propensity score (PS), methods for estimating treatment effects with observational data have received growing attention in the literature. Recent research has added substantially to the number of available statistical approaches for controlling confounding in such analyses. However, researchers need guidance to decide on the optimal analytic strategy for any given scenario. To address this gap, we conducted simulations evaluating both well-established methods (regression, PS weighting, stratification, and matching) and more recently proposed approaches (tree-based methods, local control, entropy balancing, genetic matching, prognostic scoring). The simulation scenarios included tree-based and smooth regression models as true data-generation mechanisms. We evaluated an extensive number of analysis strategies combining different treatment choices and outcome models. Key findings include 1) the lack of a single best strategy across all potential scenarios; 2) the importance of appropriately addressing interactions in the treatment choice model and/or outcome model; and 3) a tree-structured treatment choice model and a polynomial outcome model with second-order interactions performed well. One limitation to this initial assessment is the lack of heterogeneous simulation scenarios allowing treatment effects to vary by patient.     

Statistical learning approach for predicting specific pharmacodynamic,pharmacokinetic, or toxicological properties of pharmaceutical agents

Pharmaceutical agents have been developed and tested for possessing desirable pharmacodynamic, pharmacokinetic, and minimal level of toxicological properties. Computational methods have been explored for predicting these properties aimed at the discovery of promising leads and the elimination of unsuitable ones in early stages of drug development. Statistical learning methods have shown their potential for predicting these properties for structurally diverse sets of agents by using both conventional (quantitative structure–activity and structure–property relationships) and more recently explored (such as neural networks and support vector machines) statistical models. These methods have been used for predicting agents of a variety of pharmacodynamic (such as inhibitors or agonists of a therapeutic target), pharmacokinetic (such as P‐glycoprotein substrates, human intestine absorption, and blood–brain barrier penetrating capabilities), and toxicological (such as genotoxicity) properties. The strategies, current progresses, and the underlying difficulties and future prospects of the application of the recently explored statistical learning methods are discussed. Drug Dev. Res. 66:245–259, 2006. © 2006 Wiley‐Liss, Inc.     

Simultaneous spectrophotometric determination of chlordiazepoxide and clidinium using multivariate calibration techniques

Three multivariate modelling approaches including partial least squares regression (PLS), genetic algorithm‐partial least squares regression (GA‐PLS), and principal components‐artificial neural network (PC‐ANN) analysis were investigated for their application to the simultaneous determination of chlordiazepoxide and clidinium levels in pharmaceuticals. A set of synthetic mixtures of drugs in ethanol and 0.1 M HCL was made, and the prediction abilities of the aforementioned methods were examined using RSE% (relative standard error of the prediction). The PLS and PC‐ANN methods were found to be comparable, and GA‐PLS produced slightly better results. The predictive models that we built were successfully applied to simultaneously determine the levels of chlordiazepoxide and clidinium in coated tablets. Copyright © 2010 John Wiley & Sons, Ltd.     

Dose as instrumental variable in exposure-safety analysis using count models

Confounding bias often occurs in the analysis of the exposure-safety relationship due to confounding factors that have impacts on both drug exposure and safety outcomes. Instrumental variable (IV) methods have been widely used to eliminate or to reduce the bias in observational studies in, for example, epidemiology. Recently applications of IV methods can also be found in clinical trials to deal with problems such as treatment non-compliance. IV methods have rarely been used in pharmacokinetic/pharmacodynamic analyses in clinical trials, although in a randomized trial with multiple dose levels dose may be a powerful IV. We consider modeling the relationship between pharmacokinetics as a measure of drug exposure and risk of adverse events with Poisson regression models and dose as an IV. We show that although IV methods for nonlinear models are in general complex, simple approaches are available for the combination of Poisson regression models and routinely used dose-exposure models. We propose two simple methods that are intuitive and easy to implement. Both methods consist of two stages with the first stage fitting the dose-exposure model; then the fitted model is used in fitting the Poisson regression model in two different ways. The properties of the two methods are compared under several practical scenarios with simulation. A numerical example is used to illustrate an application of the methods.     

Dose as Instrumental Variable in Exposure–Safety Analysis Using Count Models

Confounding bias often occurs in the analysis of the exposure–safety relationship due to confounding factors that have impacts on both drug exposure and safety outcomes. Instrumental variable (IV) methods have been widely used to eliminate or to reduce the bias in observational studies in, for example, epidemiology. Recently applications of IV methods can also be found in clinical trials to deal with problems such as treatment non-compliance. IV methods have rarely been used in pharmacokinetic/pharmacodynamic analyses in clinical trials, although in a randomized trial with multiple dose levels dose may be a powerful IV. We consider modeling the relationship between pharmacokinetics as a measure of drug exposure and risk of adverse events with Poisson regression models and dose as an IV. We show that although IV methods for nonlinear models are in general complex, simple approaches are available for the combination of Poisson regression models and routinely used dose-exposure models. We propose two simple methods that are intuitive and easy to implement. Both methods consist of two stages with the first stage fitting the dose-exposure model; then the fitted model is used in fitting the Poisson regression model in two different ways. The properties of the two methods are compared under several practical scenarios with simulation. A numerical example is used to illustrate an application of the methods.     

Multivariate spectral calibration technique based on regression analysis for the quantitative multiresolution of a ternary mixture containing paracetamol, ascorbic acid and acetylsalicylic acid in effervescent tablets

Multivariate spectral calibration techniques based on regression analysis were established for the quantitative multiresolution of a ternary mixture containing parecetamol (PAR) ascorbic acid (AA) and acetylsalicylic acid (ASP) having closely overlapping spectra. The mathematical algorithms of multivariate spectral calibrations as namely tri-linear regression calibration (TLRC) and multi-linear regression calibration (MLRC) are based on the use of the linear regression equations at a three-wavelength set and a ten-wavelength set in the range of 215-305 nm. These calibration techniques do not require any chemical pre-treatment and a graphical procedure of the overlapping spectra. The mathematical content of TLRC and MLRC approaches were briefly formulated for the quantitative analysis of three- or multi-component mixtures. The applicability of the formulated calibration models were tested by analysing the various synthetic ternary mixtures consisting of these active compounds and then these models were applied to real pharmaceutical formulations. It was observed that TLRC and MLRC models give a successful quantitative multiresolution. The experimental results of these techniques were compared with each other as well as with those obtained by literature methods.     

Sensation seeking needs among 8th and 11th graders: characteristics associated with cigarette and marijuana use

This cross-sectional school-based study explored the relationship between adolescent use of cigarettes and marijuana and the sensation seeking personality factors of (1) Disinhibition and (2) Thrill and Adventure Seeking. The study population included a representative sample of both male and female 8th and 11th graders in the state of Delaware. Analytic methods utilized included correlational analysis and multivariate logistic regression. In the multivariate logistic regression models, the Disinhibition personality factor accounted for cigarette and marijuana using behaviors with odds ratios ranging between 2 and 3. Thrill and Adventure Seeking was not a significant explanatory variable in any of the final multivariate models. Potential confounders (age, gender and race) were considered in all analyses. Of all the two-way interactions assessed, none was significant. The findings from this study utilizing a large general community sample indicate that sensation seeking needs are a potential risk factor for adolescent substance use.     

Simultaneous determination of Fe(II) and Fe(III) in pharmaceutical formulations with chromogenic mixed reagent by using principal component artificial neural network and multivariate calibration

The use of chemometric approaches for the simultaneous determination of Fe(II) and Fe(III) ions has been explored by means of a two component reagent. Mixed reagents of 1,10-phenanthroline and thiocyanate were used as a selective chromogenic system for speciation of Fe(II) and Fe(III). Although the complexes of Fe(II) and Fe(III) with mixed reagent show a spectral overlap, they have been simultaneously determined with chemometric approaches, such as principal component artificial neural network (PC-ANN), principal component regression (PCR) and partial least squares (PLS). A set of synthetic mixtures of Fe(II) and Fe(III) was evaluated and the results obtained by the applications of these chemometric approaches were discussed and compared. It was found that the PC-ANN and PLS methods afforded better precision relatively than its of PCR. PC-ANN and PLS methods were also applied satisfactorily in determination of Fe(II) and Fe(III) in pharmaceutical samples.     

Dose-response modeling of in vivo genotoxicity data for use in risk assessment: some approaches illustrated by an analysis of acrylamide

Methods for dose-response modeling of in vivo genotoxicity data are introduced and applied to a case study of acrylamide. Genetic toxicity results are typically summarized as being either positive or negative, with no further consideration of the dose-response patterns that can be estimated from such studies. This analysis explores the use of three modeling approaches: Poisson regression of counts of genetic effects per cell; dynamic modeling of the time-course of micronucleus production and loss as a function of exposure; and categorical regression of sets of genetic toxicity experiments, the results of which are recoded in terms of severities of response. Estimates derived from these models (benchmark doses and predictions of response rates for predetermined doses of interest) are then used to assess the relevance and role of the genetic toxicity results in a risk assessment. With respect to the acrylamide data base, the results suggest that the genetic damage studies do not appear to be consistent or congruent with the thyroid tumor endpoints observed in two long-term bioassays in rats. This suggests that acrylamide's mechanism of action with respect to production of such tumors may not be genotoxic, and that a cancer risk assessment that applied a linear, no-threshold approach to such endpoints might be inappropriate. Benchmark doses derived from the genetic toxicity data base do not appear to be the critical ones for acrylamide risk assessment. Dose metric and modeling issues associated with the proposed dose-response approach to evaluation of genetic toxicity data are explored, and it is recommended that further advancements of the methodology be developed and employed for optimal use of such data for risk assessment purposes.     

Models and methods for predicting drug transfer into human milk

Adverse effects in infants due to the ingestion of drugs and other xenobiotics remain an area of concern. A key parameter in assessing infant exposure via breast milk, the milk to plasma concentration ratio (M/P), has not been determined in vivo in humans for most drugs. There are various methods for predicting M/P, which involve in vitro experiments in mammary cell monolayers, assessment of drug binding to plasma and milk protein and lipid, in vivo experiments in animals, and regression models based on a compound's physicochemical characteristics. This article reviews these approaches in terms of their utility, advantages and disadvantages. Some combination of these methods is necessary for reasonably accurate prediction of M/P in humans.     

Approaches to enhancing immune responses stimulated by CpG oligodeoxynucleotides

CpG oligodeoxynucleotides (ODN) activate the immune system and are promising immunotherapeutic agents against infectious diseases, allergy/asthma and cancer. It has become apparent that while CpG ODN are potent immune activators in mice, their immune stimulatory effects are often less dramatic in humans and large animals. This disparity between rodents and mammals has been attributed to the differences in TLR9 expression in different species. This along with the sometimes transient activity of ODN may limit its potential immunotherapeutic applications. Several approaches to enhance the activity of CpG ODN have been explored including formulation of ODN in depot-forming adjuvants, and more recently, coadministration with polyphosphazenes, inhibitors of cytokines that downregulate TLR9 activation, and simultaneous activation with multiple TLR agonists. We will discuss these approaches and the mechanisms involved, with emphasis on what we have learned from large animal models.     

Ecotoxicology and Population Genetics: The Emergence of ‘Phylogeographic and Evolutionary Ecotoxicology’

Genetics of ecotoxicology has recently emerged as a priority research field. The advent of polymerase chain reaction and molecular population genetics has made it possible to examine the genetics in even the smallest individuals. Although a potentially powerful technique, current approaches oversimplify the relationship of change in gene frequency to contaminant exposure. Many of these approaches cannot control for random correlation or accessory abiotic factors that impinge on the system tested. Indeed, the gestalt approaches of laboratory exposure or natural field experiments may ignore significant genome-level interactions that are important within a given system. At the very least, these approaches would benefit by a biogeographic survey of genetic variation to understand geographic microevolutionary patterns, or phylogeography, within a species to reduce spurious correlations and erroneous conclusions. Other single locus approaches can be chosen to enhance this approach if genetic/environmental interactions have been characterized for laboratory populations or for other model systems.     

QSAR based therapeutic management of <Emphasis Type="Italic">M. tuberculosis</Emphasis>

Mycobacterium tuberculosis is responsible for severe mortality and morbidity worldwide but, under-developed and developing countries are more prone to infection. In search of effective and wide-spectrum anti-tubercular agents, interdisciplinary approaches are being explored. Of the several approaches used, computer based quantitative structure activity relationship (QSAR) have gained momentum. Structure-based drug design and discovery implies a combined knowledge of accurate prediction of ligand poses with the good prediction and interpretation of statistically validated models derived from the 3D-QSAR approach. The validated models are generally used to screen a small combinatorial library of potential synthetic candidates to identify hits which further subjected to docking to filter out compounds as novel potential emerging drug molecules to address multidrug-resistant tuberculosis. Several newer models are integrated to QSAR methods which include different types of chemical and biological data, and simultaneous prediction of pharmacological activities including toxicities and/or other safety profiles to get new compounds with desired activity. In the process, several newer molecules have been identified which are now being assessed for their clinical efficacy. Present review deals with the advances made in the field highlighting overall future prospects of the development of anti-tuberculosis drugs.