Modeling lesion counts in multiple sclerosis when patients have been selected for baseline activity

Abstract

The number of new gadolinium-enhancing lesions discovered via magnetic resonance imaging is a well-established outcome for multiple sclerosis studies, especially Phase II Studies. Due to the high cost of magnetic resonance imaging scans, many investigators select participants for the presence of lesions. While this selection procedure is thought to improve the power of inferences, the effect of screening for baseline activity on parameter estimation and interval coverage has not yet been examined. The objective of this study was to investigate the performance of the negative binomial distribution for modeling lesion count data in multiple sclerosis when patients have been selected for activity on a baseline scan. We performed computer simulations to investigate the influence of the screening process on inferences made using a negative binomial model about treatment effects in two independent samples. We also demonstrate how the statistical properties of screening can be incorporated into trial design. We demonstrate that when the negative binomial distribution is used to model lesion counts, while screening for baseline activity improves point estimation, this practice also has the potential to decrease interval coverage and inflate the Type I error rate. For data that is to be modeled using a negative binomial distribution, screening for baseline activity can create a trade-off between cost effectiveness and a higher than desired false positive rate that must be carefully considered in planning Phase II trials.

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

“…Recently, Morgan et al . have discussed the pros and cons of ‘enriched’ study design where patients with no lesions would be excluded from the trial prior to randomization. Such a recruitment strategy would help in overcoming the problem of increasing sample size due to the decreasing level of disease activity (hence, number of events) in the MS study populations.…”

“…Optimizing this model would not only improve the accuracy of prospective sample size estimations but also help in better understanding the disease and the effect of a therapeutic treatment on its progression. Recently, Morgan et al [24] have discussed the pros and cons of 'enriched' study design where patients with no lesions would be excluded from the trial prior to randomization. Such a recruitment strategy would help in overcoming the problem of increasing sample size due to the decreasing level of disease activity (hence, number of events) in the MS study populations.…”

Identification of treatment responders based on multiple longitudinal outcomes with applications to multiple sclerosis patients

“…The posterior probabilities estimated from SINGLE-T25FW are most strongly correlated with the change in MSFC, followed by those from TRIPLE, while those from SINGLE-PASAT and SINGLE-T25FW are not as strongly correlated. Patients 8,23,32,84,and 86 in the left bottom corner of Figure 5 (a) are not identified as responders by TRIPLE but have a relatively large decrease in MSFC scores (< −3). As seen in Figure 4 (a), these patients are identified as responders only by SINGLE-T25FW.…”

A flexible mixed‐effect negative binomial regression model for detecting unusual increases in MRI lesion counts in individual multiple sclerosis patients

“…The Poisson distribution, commonly used to model count data, restricts the mean and the variance of the count variable to be the same. The NB distribution allows the variance to be larger than the mean and has been the main candidate to model the distribution of CEL counts [6][7][8].…”