A Gentle Introduction to Bayesian Network Meta-Analysis Using an Automated R Package.

Network meta-analysis is an extension of standard meta-analysis. It allows researchers to build a network of evidence to compare multiple interventions that may have not been compared directly in existing publications. With a Bayesian approach, network meta-analysis can be used to obtain a posterior probability distribution of all the relative treatment effects, which allows for the estimation of relative treatment effects to quantify the uncertainty of parameter estimates, and to rank all the treatments in the network. Ranking treatments using both direct and indirect evidence can provide guidance to policy makers and clinicians for making decisions. The purpose of this paper is to introduce fundamental concepts of Bayesian network meta-analysis (BNMA) to researchers in psychology and social sciences. We discuss several essential concepts of BNMA, including the assumptions of homogeneity and consistency, the fixed and random effects models, prior specification, and model fit evaluation strategies, while pointing out some issues and areas where researchers should use caution in the application of BNMA. Additionally, using an automated R package, we provide a step-by-step demonstration on how to conduct and report the findings of BNMA with a real dataset of psychological interventions extracted from PubMed.

Bayesian unanchored additive models for component network meta-analysis.

Component network meta-analysis (CNMA) models are an extension of standard network meta-analysis (NMA) models which account for the use of multicomponent treatments in the network. This article contributes innovatively to several statistical aspects of CNMA. First, by introducing a unified notation, we establish that currently available methods differ in the way they assume additivity, an important distinction that has been overlooked so far in the literature. In particular, one model uses a more restrictive form of additivity than the other which we term an anchored and unanchored model, respectively. We show that an anchored model can provide a poor fit to the data if it is misspecified. Second, given that Bayesian models are often preferred by practitioners, we develop two novel unanchored Bayesian CNMA models presented under the unified notation. An extensive simulation study examining bias, coverage probabilities, and treatment rankings confirms the favorable performance of the novel models. This is the first simulation study to compare the statistical properties of CNMA models in the literature. Finally, the use of our novel models is demonstrated on a real dataset, and the results of CNMA models on the dataset are compared.

BUGSnet: an R package to facilitate the conduct and reporting of Bayesian network Meta-analyses.

BACKGROUND: Several reviews have noted shortcomings regarding the quality and reporting of network meta-analyses (NMAs). We suspect that this issue may be partially attributable to limitations in current NMA software which do not readily produce all of the output needed to satisfy current guidelines. RESULTS: To better facilitate the conduct and reporting of NMAs, we have created an R package called "BUGSnet" (Bayesian inference Using Gibbs Sampling to conduct a Network meta-analysis). This R package relies upon Just Another Gibbs Sampler (JAGS) to conduct Bayesian NMA using a generalized linear model. BUGSnet contains a suite of functions that can be used to describe the evidence network, estimate a model and assess the model fit and convergence, assess the presence of heterogeneity and inconsistency, and output the results in a variety of formats including league tables and surface under the cumulative rank curve (SUCRA) plots. We provide a demonstration of the functions contained within BUGSnet by recreating a Bayesian NMA found in the second technical support document composed by the National Institute for Health and Care Excellence Decision Support Unit (NICE-DSU). We have also mapped these functions to checklist items within current reporting and best practice guidelines. CONCLUSION: BUGSnet is a new R package that can be used to conduct a Bayesian NMA and produce all of the necessary output needed to satisfy current scientific and regulatory standards. We hope that this software will help to improve the conduct and reporting of NMAs.

Adjusting for undercoverage of access-points in creel surveys with fewer overflights.

Creel surveys are used in recreational fisheries to estimate angling effort, catch, and harvest. Aerial-access creel surveys rely on two components: (1) a ground component in which fishing parties returning from their trips are interviewed at some access-points of the fishery; (2) an aerial component in which the number of fishing parties is counted. A common practice is to sample fewer aerial survey days than ground survey days. This is thought by practitioners to reduce the cost of the survey, but there is a lack of sound statistical methodology for this case. In this article, we propose various estimation methods to handle this situation and evaluate their asymptotic properties from a design-based perspective. We also propose formulas for the optimal allocation of the effort between the ground and the aerial portion of the survey, for given costs and budget. A simulation study investigates the performance of the estimators. Finally, we apply our methods to data from an annual Kootenay Lake survey (Canada).

Predictors of long-term use of prescription opioids in the community-dwelling population of adults without a cancer diagnosis: a retrospective cohort study.

BACKGROUND: Long-term opioid use is a known risk factor for opioid-related harms. We aimed to identify risk factors for and predictors of long-term use of prescription opioids in the community-dwelling population of adults without a diagnosis of cancer, to inform practice change at the point of care. METHODS: Using Quebec administrative claims databases, we conducted a retrospective cohort study in a random sample of adult members (≥ 18 yr) of the public drug plan who did not have a cancer diagnosis and who initiated a prescription opioid in the outpatient setting between Jan. 1, 2012, and Dec. 31, 2016. The outcome of interest was long-term opioid use (≥ 90 consecutive days or ≥ 120 cumulative days over 12 mo). Potential predictors included sociodemographic factors, medical history, characteristics of the initial opioid prescription and prescriber's specialty. We used multivariable logistic regression to assess the association between each characteristic and long-term use. We used the area under the receiver operating characteristic curve to determine the predictive performance of full and parsimonious models. RESULTS: Of 124 664 eligible patients who initiated opioid therapy, 4172 (3.3%) progressed to long-term use of prescription opioids. The most important associated factors in the adjusted analysis were long-term prescription of acetaminophen-codeine (odds ratio [OR] 6.30, 95% confidence interval [CI] 4.99 to 7.96), prescription of a long-acting opioid at initiation (OR 6.02, 95% CI 5.31 to 6.84), initial supply of 30 days or more (OR 4.22, 95% CI 3.81 to 4.69), chronic pain (OR 2.41, 95% CI 2.16 to 2.69) and initial dose of at least 90 morphine milligram equivalents (MME) per day (OR 1.24, 95% CI 1.04 to 1.47). Our predictive model, including only the initial days' supply and chronic pain diagnosis, had area under the curve of 0.7618. INTERPRETATION: This study identified factors associated with long-term prescription opioid use. Limiting the initial supply to no more than 7 days and limiting doses to 90 MME/day or less are actions that could be undertaken at the point of care.

Probabilistic Graphical Modeling for Estimating Risk of Coronary Artery Disease: Applications of a Flexible Machine-Learning Method.

Objectives. Coronary artery disease (CAD) is the leading cause of death and disease burden worldwide, causing 1 in 7 deaths in the United States alone. Risk prediction models that can learn the complex causal relationships that give rise to CAD from data, instead of merely predicting the risk of disease, have the potential to improve transparency and efficacy of personalized CAD diagnosis and therapy selection for physicians, patients, and other decision makers. Methods. We use Bayesian networks (BNs) to model the risk of CAD using the Z-Alizadehsani data set-a published real-world observational data set of 303 Iranian patients at risk for CAD. We also describe how BNs can be used for incorporation of background knowledge, individual risk prediction, handling missing observations, and adaptive decision making under uncertainty. Results. BNs performed on par with machine-learning classifiers at predicting CAD and showed better probability calibration. They achieved a mean 10-fold area under the receiver-operating characteristic curve (AUC) of 0.93 ± 0.04, which was comparable with the performance of logistic regression with L1 or L2 regularization (AUC: 0.92 ± 0.06), support vector machine (AUC: 0.92 ± 0.06), and artificial neural network (AUC: 0.91 ± 0.05). We describe the use of BNs to predict with missing data and to adaptively calculate prognostic values of individual variables under uncertainty. Conclusion. BNs are powerful and versatile tools for risk prediction and health outcomes research that can complement traditional statistical techniques and are particularly useful in domains in which information is uncertain or incomplete and in which interpretability is important, such as medicine.

Characteristics and methods of incorporating randomized and nonrandomized evidence in network meta-analyses: a scoping review.

OBJECTIVES: The objective of the study was to conduct a scoping review of the published literature on methods used to combine randomized and nonrandomized evidence (NRE) in network meta-analyses (NMAs) and their respective characteristics. STUDY DESIGN AND SETTING: We conducted a scoping review using a list of NMAs which incorporated NRE that were identified from a previous review. All NMAs that included NRE in the analysis of main outcomes or sensitivity analyses were eligible for inclusion. Two reviewers independently screened studies for inclusion and performed data abstraction. Data analysis involved quantitative (frequencies and percentages) and qualitative (narrative synthesis) methods. RESULTS: A total of 23 NMAs met the predefined inclusion criteria, of which 74% (n = 17) used naïve pooling, 0% used NRE as informative priors, 9% (n = 2) used the 3-level Bayesian hierarchical model, 9% (n = 2) used all methods, and 9% (n = 2) used other methods. Most NMAs were supplemented with additional analyses to investigate the effect estimates when only randomized evidence was included. CONCLUSION: Although most studies provided justification for the inclusion of NRE, transparent reporting of the method used to combine randomized evidence and NRE was unclear in most published networks. Most NMAs used naïve pooling for combining randomized evidence and NRE.

Network meta-analysis of disconnected networks: How dangerous are random baseline treatment effects?

In network meta-analysis, the use of fixed baseline treatment effects (a priori independent) in a contrast-based approach is regularly preferred to the use of random baseline treatment effects (a priori dependent). That is because, often, there is not a need to model baseline treatment effects, which carry the risk of model misspecification. However, in disconnected networks, fixed baseline treatment effects do not work (unless extra assumptions are made), as there is not enough information in the data to update the prior distribution on the contrasts between disconnected treatments. In this paper, we investigate to what extent the use of random baseline treatment effects is dangerous in disconnected networks. We take 2 publicly available datasets of connected networks and disconnect them in multiple ways. We then compare the results of treatment comparisons obtained from a Bayesian contrast-based analysis of each disconnected network using random normally distributed and exchangeable baseline treatment effects to those obtained from a Bayesian contrast-based analysis of their initial connected network using fixed baseline treatment effects. For the 2 datasets considered, we found that the use of random baseline treatment effects in disconnected networks was appropriate. Because those datasets were not cherry-picked, there should be other disconnected networks that would benefit from being analyzed using random baseline treatment effects. However, there is also a risk for the normality and exchangeability assumption to be inappropriate in other datasets even though we have not observed this situation in our case study. We provide code, so other datasets can be investigated.