Generalized Linear Mixed Models for Binary Data: Are Matching Results from Penalized Quasi-Likelihood and Numerical Integration Less Biased?

Benedetti, Andrea; Platt, Robert W.; Atherton, Juli

doi:10.1371/journal.pone.0084601

Cited by 19 publications

(12 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We could see that already in the first step, a bias was introduced, where the mixed model was fitted, and thus an NMA based on the estimated treatment effects from the first step can not be unbiased. This is in line with the results that are presented, for example, in Benedetti et al (), where the issue of biased estimates from mixed models for binary data is discussed. As a side note, in very few situations, the mixed model led to estimated (log) odds ratios with very high estimated variances which leads to doubts about the convergence of the model.…”

Section: Discussionsupporting

confidence: 92%

“…To estimate the (adjusted) log odds ratio in the first step of the two‐step approach, that is, the log odds ratio obtained in the mixed‐effects model, mainly two different approaches can be used, the Gaussian Hermite Quadrature or the penalized quasi‐likelihood method. An abundant amount of discussion exists in the literature about these estimating procedures (for example, Benedetti et al, ; Austin, ). We decided to use the penalized quasi‐likelihood procedure implemented in the R‐package mass (Venables and Ripley, , version 7.3‐22), because it is less CPU‐intensive.…”

Section: Simulation Studymentioning

confidence: 99%

See 1 more Smart Citation

Bayesian network meta‐analysis for cluster randomized trials with binary outcomes

Uhlmann

Jensen

Kieser

2016

Research Synthesis Methods

View full text Add to dashboard Cite

Network meta-analysis is becoming a common approach to combine direct and indirect comparisons of several treatment arms. In recent research, there have been various developments and extensions of the standard methodology. Simultaneously, cluster randomized trials are experiencing an increased popularity, especially in the field of health services research, where, for example, medical practices are the units of randomization but the outcome is measured at the patient level. Combination of the results of cluster randomized trials is challenging. In this tutorial, we examine and compare different approaches for the incorporation of cluster randomized trials in a (network) meta-analysis. Furthermore, we provide practical insight on the implementation of the models. In simulation studies, it is shown that some of the examined approaches lead to unsatisfying results. However, there are alternatives which are suitable to combine cluster randomized trials in a network meta-analysis as they are unbiased and reach accurate coverage rates. In conclusion, the methodology can be extended in such a way that an adequate inclusion of the results obtained in cluster randomized trials becomes feasible. Copyright © 2016 John Wiley & Sons, Ltd.

show abstract

Section: Discussionsupporting

confidence: 92%

Section: Simulation Studymentioning

confidence: 99%

Bayesian network meta‐analysis for cluster randomized trials with binary outcomes

Uhlmann

Jensen

Kieser

2016

Research Synthesis Methods

View full text Add to dashboard Cite

show abstract

“…The approaches yielded similar results in relation to statistical significance and direction. The results of the QUAD function were retained for analysis because some literature suggests it may yield less biased results when dealing with small numbers of observations within clusters or for binary response variables, as is the case here ( 36 ). A traditional generalized linear model with no random effects was also examined, and it was attempted to account for spatial bias by removing entirely people who walked or rode a bicycle, as they were tightly clustered in areas directly adjacent to campus.…”

Section: Methodsmentioning

confidence: 99%

Commuters’ Assessment of Public Transport as a “Reasonable” Option in Montreal, QC

DeWeese

El-Geneidy

2020

Transportation Research Record

View full text Add to dashboard Cite

Retaining and increasing public transport ridership is a centerpiece of many strategies to address both the climate crisis and public health challenges. Understanding how and why commuters choose or reject public transport as a viable option or actual mode is, thus, central to policymakers’ efforts. This study makes use of a detailed travel-behavior survey conducted at McGill University in Montreal, Quebec, to answer two key questions: 1) What factors influence travelers’ perception of public transport as a reasonable commuting option? and 2) From among those travelers that do consider public transport to be reasonable, what factors influence their final decision to use it? One important finding is that there is sometimes a disconnect between the factors that influence a person’s initial assessment of reasonableness and subsequent mode choice. For example, car owners were paradoxically more likely to consider public transport a reasonable option but significantly less likely to use it. More generally, another important finding of this study is that there may be a sizeable contingent of travelers who consider public transport to be a reasonable or viable option but nonetheless decline to use it. These travelers may be easier to convert to public transport than those who do not already consider it a reasoanble option, making it important for policymakers to understand their motivations. Ultimately, public transport agencies may be able to use this type of information to develop policies better targeted at bolstering ridership.

show abstract

“…According to previous literature, these factors have all been identified as influencing model performance [ 6 ]. While several simulation studies have been published, these have mainly limited their attention to simple models with only random intercepts [ 13 , 16 ]. Thus, the performance of the random effects models including both a random intercept and a random slope are less well known.…”

Section: Introductionmentioning

confidence: 99%

A comparison of analytic approaches for individual patient data meta-analyses with binary outcomes

Thomas

Platt

Benedetti

2017

BMC Med Res Methodol

Self Cite

View full text Add to dashboard Cite

BackgroundIndividual patient data meta-analyses (IPD-MA) are often performed using a one-stage approach-- a form of generalized linear mixed model (GLMM) for binary outcomes. We compare (i) one-stage to two-stage approaches (ii) the performance of two estimation procedures (Penalized Quasi-likelihood-PQL and Adaptive Gaussian Hermite Quadrature-AGHQ) for GLMMs with binary outcomes within the one-stage approach and (iii) using stratified study-effect or random study-effects.MethodsWe compare the different approaches via a simulation study, in terms of bias, mean-squared error (MSE), coverage and numerical convergence, of the pooled treatment effect (β 1) and between-study heterogeneity of the treatment effect (τ 12). We varied the prevalence of the outcome, sample size, number of studies and variances and correlation of the random effects.ResultsThe two-stage and one-stage methods produced approximately unbiased β 1 estimates. PQL performed better than AGHQ for estimating τ 12 with respect to MSE, but performed comparably with AGHQ in estimating the bias of β 1 and of τ 12. The random study-effects model outperformed the stratified study-effects model in small size MA.ConclusionThe one-stage approach is recommended over the two-stage method for small size MA. There was no meaningful difference between the PQL and AGHQ procedures. Though the random-intercept and stratified-intercept approaches can suffer from their underlining assumptions, fitting GLMM with a random-intercept are less prone to misfit and has good convergence rate.Electronic supplementary materialThe online version of this article (doi:10.1186/s12874-017-0307-7) contains supplementary material, which is available to authorized users.

show abstract

Generalized Linear Mixed Models for Binary Data: Are Matching Results from Penalized Quasi-Likelihood and Numerical Integration Less Biased?

Cited by 19 publications

References 19 publications

Bayesian network meta‐analysis for cluster randomized trials with binary outcomes

Bayesian network meta‐analysis for cluster randomized trials with binary outcomes

Commuters’ Assessment of Public Transport as a “Reasonable” Option in Montreal, QC

A comparison of analytic approaches for individual patient data meta-analyses with binary outcomes

Contact Info

Product

Resources

About