Dealing with excess of zeros in the statistical analysis of magnetic resonance imaging lesion count in multiple sclerosis

Mercier, François; Peter, Chin; Francis, Gordon

doi:10.1002/pst.1529

Cited by 10 publications

(15 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…identifies model (6) as being nested within model (4), which allows hypothesis testing for model comparison. A Wald test statistic for testing whether the constraints hold is based on the p − 2 dimensional vector g(…”

Section: A Shared-parameter Marginalized Zero-inflated Negative Binommentioning

confidence: 99%

Marginalized zero‐inflated negative binomial regression with application to dental caries

Preisser

Das

Long

et al. 2015

Statistics in Medicine

View full text Add to dashboard Cite

The zero-inflated negative binomial regression model (ZINB) is often employed in diverse fields such as dentistry, health care utilization, highway safety, and medicine to examine relationships between exposures of interest and overdispersed count outcomes exhibiting many zeros. The regression coefficients of ZINB have latent class interpretations for a susceptible subpopulation at risk for the disease/condition under study with counts generated from a negative binomial distribution and for a non-susceptible subpopulation that provides only zero counts. The ZINB parameters, however, are not well-suited for estimating overall exposure effects, specifically, in quantifying the effect of an explanatory variable in the overall mixture population. In this paper, a marginalized zero-inflated negative binomial regression (MZINB) model for independent responses is proposed to model the population marginal mean count directly, providing straightforward inference for overall exposure effects based on maximum likelihood estimation. Through simulation studies, the finite sample performance of MZINB is compared to marginalized zero-inflated Poisson, Poisson, and negative binomial regression. The MZINB model is applied in the evaluation of a school-based fluoride mouthrinse program on dental caries in 677 children.

show abstract

Section: A Shared-parameter Marginalized Zero-inflated Negative Binommentioning

confidence: 99%

Marginalized zero‐inflated negative binomial regression with application to dental caries

Preisser

Das

Long

et al. 2015

Statistics in Medicine

View full text Add to dashboard Cite

show abstract

“…These data refer to the number of Gd-enhanced lesions counted on brain magnetic resonance imaging scans at baseline and months 6, 12, and 24. From table 1 of Francois et al [24], we see that the means for fingolimod and placebo at baseline and months 6, 12, and 24 are small (between 0.22 and 1.74). These means are very small (between 0.19 and 0.30) for the two different doses of the fingolimod treatment at months 6, 12, and 24.…”

Section: Introductionmentioning

confidence: 83%

“…In addition, in some situations, especially when the means are small, interval estimation of the MR is often preferable [22,23]. For instance, Francois et al [24] analyzed the lesion count in multiple sclerosis, to assess the effect of the fingolimod treatment in the FREEDOMS trial. These data refer to the number of Gd-enhanced lesions counted on brain magnetic resonance imaging scans at baseline and months 6, 12, and 24.…”

Section: Introductionmentioning

confidence: 99%

Estimating the Treatment Effect in the Analysis of Extra-Dispersed Count Response Data from Clinical Trials

Saha¹

2013

J Biomet Biostat

View full text Add to dashboard Cite

Responses in the form of counts arise in many clinical trials and epidemiological studies, and are usually extradispersed. When one wishes to estimate the treatment effect in comparison with a placebo in clinical trials, confidence intervals are frequently used. It is of common interest in many clinical trials and epidemiological studies, to obtain the confidence interval for one of the two quantities, mean difference and mean ratio. The preference of one measure over the other depends on the design of the study. In many situations, the mean ratio is more relevant than the difference of means. Confidence interval procedures for the mean difference between treatment and control groups in the analysis of such extra-dispersed counts have been studied recently, but no attention has been paid to investigating the problem of confidence interval construction for the mean ratio. In this article, we develop several asymptotic confidence interval procedures for the mean ratio, by using the delta method, to extend the variance of a single mean estimate to the variance of the mean ratio estimate. The simulation studies indicate that all procedures perform reasonably well in terms of coverage. However, the interval based on the generalized estimating equation approach, using the logarithmic transformation, performs uniformly best in terms of coverage, expected width and location, and is preferable to the other intervals, in most of the situations considered here. Finally, three real-life examples from clinical trials are analyzed to illustrate the proposed confidence interval procedures.

show abstract

“…Zero-inflated models also give a good fit to lesion count data. 27 Zero-inflated distributions are used to model count data with many zero counts; they are 2-component mixture models that combine a "point mass" function at zero with a count distribution (in this case, a Poisson or an NB distribution). Furthermore, a new parametric model (the NB time-to-event [NBT] model) has been developed that fits the distribution of time to relapse better than the conventional exponential distribution.…”

mentioning

confidence: 99%

“…New T2 lesion counts and relapse rates in our cohort were modeled with an NB distribution. This model has been shown to be an appropriate model in the context of MS. [24][25][26] For MRI lesion counts, we extended the NB model to zero-inflated distributions-a model that has recently been evaluated for white-matter 27 and cortical 28 MRI lesions. We have confirmed that the NB and ZINB distributions better fit the T2 lesion count data than the Poisson and ZIP in a pediatric MS population, and therefore, we generated sample size estimates by resampling the NB distribution.…”

mentioning

confidence: 99%

Clinical and MRI activity as determinants of sample size for pediatric multiple sclerosis trials

et al. 2013

View full text Add to dashboard Cite

Objective: To estimate sample sizes for pediatric multiple sclerosis (MS) trials using new T2 lesion count, annualized relapse rate (ARR), and time to first relapse (TTFR) endpoints.Methods: Poisson and negative binomial models were fit to new T2 lesion and relapse count data, and negative binomial time-to-event and exponential models were fit to TTFR data of 42 children with MS enrolled in a national prospective cohort study. Simulations were performed by resampling from the best-fitting model of new T2 lesion count, number of relapses, or TTFR, under various assumptions of the effect size, trial duration, and model parameters.Results: Assuming a 50% reduction in new T2 lesions over 6 months, 90 patients/arm are required, whereas 165 patients/arm are required for a 40% treatment effect. Sample sizes for 2-year trials using relapse-related endpoints are lower than that for 1-year trials. For 2-year trials and a conservative assumption of overdispersion (q), sample sizes range from 70 patients/arm (using ARR) to 105 patients/arm (TTFR) for a 50% reduction in relapses, and 230 patients/arm (ARR) to 365 patients/arm (TTFR) for a 30% relapse reduction. Assuming a less conservative q, 2-year trials using ARR require 45 patients/arm (60 patients/arm for TTFR) for a 50% reduction in relapses and 145 patients/arm (200 patients/arm for TTFR) for a 30% reduction.Conclusion: Six-month phase II trials using new T2 lesion count as an endpoint are feasible in the pediatric MS population; however, trials powered on ARR or TTFR will need to be 2 years in duration and will require multicentered collaboration. Therapies to treat pediatric-onset multiple sclerosis (MS) 1-4 are used off-label, because all approved pharmacologic agents for MS have been studied exclusively in adult patients. With the recent approval of the first oral disease-modifying therapies (DMTs) for MS 5,6 and several promising new therapies likely to be approved over the next 10 years, pediatric practitioners face the challenge of recommending suitable therapies for pediatric patients with MS despite a lack of evidence on safety and efficacy of these drugs in the pediatric population.7 Clinical trials for pediatric-onset MS are being planned, but data on appropriate trial endpoints and required sample sizes to inform trial design are lacking.Annualized relapse rate (ARR) is a common primary efficacy endpoint in clinical trials of adults with MS, and reduction in new T2 and gadolinium-enhancing lesions are frequently used secondary endpoints in phase III trials. 5,[8][9][10][11][12] Time to first relapse (TTFR) has recently been shown in adults with MS to be a feasible and powerful trial endpoint; using TTFR has the advantage of permitting patients initially randomized to placebo to be switched to active therapy at the time of first relapse.

show abstract

Dealing with excess of zeros in the statistical analysis of magnetic resonance imaging lesion count in multiple sclerosis

Cited by 10 publications

References 44 publications

Marginalized zero‐inflated negative binomial regression with application to dental caries

Marginalized zero‐inflated negative binomial regression with application to dental caries

Estimating the Treatment Effect in the Analysis of Extra-Dispersed Count Response Data from Clinical Trials

Clinical and MRI activity as determinants of sample size for pediatric multiple sclerosis trials

Contact Info

Product

Resources

About