Quantification of multiple-sclerosis-related brain atrophy in two heterogeneous MRI datasets using mixed-effects modeling

Blake C. Jones, Govind Nair, Colin D. Shea, Ciprian M Crainiceanu, Irene C M Cortese, Daniel S. Reich

Research output: Contribution to journalArticle

Abstract

Brain atrophy, measured by MRI, has been proposed as a useful surrogate marker for disease progression in multiple sclerosis (MS). However, it is conventionally assumed that the accurate quantification of brain atrophy is made difficult, if not impossible, by changes in the parameters of the MRI acquisition, which are almost inevitable over the course of a longitudinal study since MRI technology changes rapidly. This state of affairs can negatively affect clinical trial design and limit the use of historical data. Here, we investigate whether we can coherently estimate brain atrophy rates in a heterogeneous MS sample via linear mixed-effects multivariable regression, incorporating three critical assumptions: (1) using age at time of scanning, rather than time since baseline, as the regressor of interest; (2) scanning individuals with a variety of techniques; and (3) introducing a simple additive correction for major differences in MRI protocol. We fit the model to several measures of brain volume as the outcome in two MS populations: 1123 scans from 195 cases acquired for over approximately 7 years in two natural history protocols (Cohort 1), and 1331 scans from 69 cases seen for over 11 years who were primarily treated with two specific MS disease-modifying therapies (Cohort 2). We compared the mixed-effects model with additive correction for MRI acquisition parameters to a model fit without this correction and performed sample-size calculations to provide an estimate of the number of participants in an MS clinical trial that might be required to see a therapeutic effect of treatment using the approach described here. The results show that without the additive correction for T1-weighted protocol parameters, atrophy was underestimated and subject-specific estimates were more narrowly distributed about the population mean. Ventricular CSF is the most consistently estimated brain volume, with a mean of 2.8%/year increase in Cohort 1 and 4.4%/year increase in Cohort 2. An interesting observation was that gray matter volume decreased and white matter volume remained essentially unchanged in both cohorts, suggesting that changes in ventricular CSF volume are a surrogate for changes in gray matter volume. In conclusion, the mixed-effects modeling framework presented here allows effective use of heterogeneously acquired and historical data in the study of brain atrophy in MS, potentially simplifying the design of future single- and multi-site clinical trials and natural history studies.

Original languageEnglish (US)
Pages (from-to)171-179
Number of pages9
JournalNeuroImage: Clinical
Volume3
DOIs
StatePublished - 2013

Fingerprint

Multiple Sclerosis
Atrophy
Brain
Clinical Trials
Natural History
Therapeutic Uses
Sample Size
Population
Longitudinal Studies
Disease Progression
Datasets
Biomarkers
Observation
Technology
Gray Matter

Keywords

  • Brain atrophy
  • Heterogeneous data
  • Mixed-effects model
  • MRI
  • Multiple sclerosis

ASJC Scopus subject areas

  • Clinical Neurology
  • Radiology Nuclear Medicine and imaging
  • Cognitive Neuroscience
  • Neurology

Cite this

Quantification of multiple-sclerosis-related brain atrophy in two heterogeneous MRI datasets using mixed-effects modeling. / Jones, Blake C.; Nair, Govind; Shea, Colin D.; Crainiceanu, Ciprian M; Cortese, Irene C M; Reich, Daniel S.

In: NeuroImage: Clinical, Vol. 3, 2013, p. 171-179.

Research output: Contribution to journalArticle

Jones, Blake C. ; Nair, Govind ; Shea, Colin D. ; Crainiceanu, Ciprian M ; Cortese, Irene C M ; Reich, Daniel S. / Quantification of multiple-sclerosis-related brain atrophy in two heterogeneous MRI datasets using mixed-effects modeling. In: NeuroImage: Clinical. 2013 ; Vol. 3. pp. 171-179.
@article{606440443cc144a8a06cc04b7f509733,
title = "Quantification of multiple-sclerosis-related brain atrophy in two heterogeneous MRI datasets using mixed-effects modeling",
abstract = "Brain atrophy, measured by MRI, has been proposed as a useful surrogate marker for disease progression in multiple sclerosis (MS). However, it is conventionally assumed that the accurate quantification of brain atrophy is made difficult, if not impossible, by changes in the parameters of the MRI acquisition, which are almost inevitable over the course of a longitudinal study since MRI technology changes rapidly. This state of affairs can negatively affect clinical trial design and limit the use of historical data. Here, we investigate whether we can coherently estimate brain atrophy rates in a heterogeneous MS sample via linear mixed-effects multivariable regression, incorporating three critical assumptions: (1) using age at time of scanning, rather than time since baseline, as the regressor of interest; (2) scanning individuals with a variety of techniques; and (3) introducing a simple additive correction for major differences in MRI protocol. We fit the model to several measures of brain volume as the outcome in two MS populations: 1123 scans from 195 cases acquired for over approximately 7 years in two natural history protocols (Cohort 1), and 1331 scans from 69 cases seen for over 11 years who were primarily treated with two specific MS disease-modifying therapies (Cohort 2). We compared the mixed-effects model with additive correction for MRI acquisition parameters to a model fit without this correction and performed sample-size calculations to provide an estimate of the number of participants in an MS clinical trial that might be required to see a therapeutic effect of treatment using the approach described here. The results show that without the additive correction for T1-weighted protocol parameters, atrophy was underestimated and subject-specific estimates were more narrowly distributed about the population mean. Ventricular CSF is the most consistently estimated brain volume, with a mean of 2.8{\%}/year increase in Cohort 1 and 4.4{\%}/year increase in Cohort 2. An interesting observation was that gray matter volume decreased and white matter volume remained essentially unchanged in both cohorts, suggesting that changes in ventricular CSF volume are a surrogate for changes in gray matter volume. In conclusion, the mixed-effects modeling framework presented here allows effective use of heterogeneously acquired and historical data in the study of brain atrophy in MS, potentially simplifying the design of future single- and multi-site clinical trials and natural history studies.",
keywords = "Brain atrophy, Heterogeneous data, Mixed-effects model, MRI, Multiple sclerosis",
author = "Jones, {Blake C.} and Govind Nair and Shea, {Colin D.} and Crainiceanu, {Ciprian M} and Cortese, {Irene C M} and Reich, {Daniel S.}",
year = "2013",
doi = "10.1016/j.nicl.2013.08.001",
language = "English (US)",
volume = "3",
pages = "171--179",
journal = "NeuroImage: Clinical",
issn = "2213-1582",
publisher = "Elsevier BV",

}

TY - JOUR

T1 - Quantification of multiple-sclerosis-related brain atrophy in two heterogeneous MRI datasets using mixed-effects modeling

AU - Jones, Blake C.

AU - Nair, Govind

AU - Shea, Colin D.

AU - Crainiceanu, Ciprian M

AU - Cortese, Irene C M

AU - Reich, Daniel S.

PY - 2013

Y1 - 2013

N2 - Brain atrophy, measured by MRI, has been proposed as a useful surrogate marker for disease progression in multiple sclerosis (MS). However, it is conventionally assumed that the accurate quantification of brain atrophy is made difficult, if not impossible, by changes in the parameters of the MRI acquisition, which are almost inevitable over the course of a longitudinal study since MRI technology changes rapidly. This state of affairs can negatively affect clinical trial design and limit the use of historical data. Here, we investigate whether we can coherently estimate brain atrophy rates in a heterogeneous MS sample via linear mixed-effects multivariable regression, incorporating three critical assumptions: (1) using age at time of scanning, rather than time since baseline, as the regressor of interest; (2) scanning individuals with a variety of techniques; and (3) introducing a simple additive correction for major differences in MRI protocol. We fit the model to several measures of brain volume as the outcome in two MS populations: 1123 scans from 195 cases acquired for over approximately 7 years in two natural history protocols (Cohort 1), and 1331 scans from 69 cases seen for over 11 years who were primarily treated with two specific MS disease-modifying therapies (Cohort 2). We compared the mixed-effects model with additive correction for MRI acquisition parameters to a model fit without this correction and performed sample-size calculations to provide an estimate of the number of participants in an MS clinical trial that might be required to see a therapeutic effect of treatment using the approach described here. The results show that without the additive correction for T1-weighted protocol parameters, atrophy was underestimated and subject-specific estimates were more narrowly distributed about the population mean. Ventricular CSF is the most consistently estimated brain volume, with a mean of 2.8%/year increase in Cohort 1 and 4.4%/year increase in Cohort 2. An interesting observation was that gray matter volume decreased and white matter volume remained essentially unchanged in both cohorts, suggesting that changes in ventricular CSF volume are a surrogate for changes in gray matter volume. In conclusion, the mixed-effects modeling framework presented here allows effective use of heterogeneously acquired and historical data in the study of brain atrophy in MS, potentially simplifying the design of future single- and multi-site clinical trials and natural history studies.

AB - Brain atrophy, measured by MRI, has been proposed as a useful surrogate marker for disease progression in multiple sclerosis (MS). However, it is conventionally assumed that the accurate quantification of brain atrophy is made difficult, if not impossible, by changes in the parameters of the MRI acquisition, which are almost inevitable over the course of a longitudinal study since MRI technology changes rapidly. This state of affairs can negatively affect clinical trial design and limit the use of historical data. Here, we investigate whether we can coherently estimate brain atrophy rates in a heterogeneous MS sample via linear mixed-effects multivariable regression, incorporating three critical assumptions: (1) using age at time of scanning, rather than time since baseline, as the regressor of interest; (2) scanning individuals with a variety of techniques; and (3) introducing a simple additive correction for major differences in MRI protocol. We fit the model to several measures of brain volume as the outcome in two MS populations: 1123 scans from 195 cases acquired for over approximately 7 years in two natural history protocols (Cohort 1), and 1331 scans from 69 cases seen for over 11 years who were primarily treated with two specific MS disease-modifying therapies (Cohort 2). We compared the mixed-effects model with additive correction for MRI acquisition parameters to a model fit without this correction and performed sample-size calculations to provide an estimate of the number of participants in an MS clinical trial that might be required to see a therapeutic effect of treatment using the approach described here. The results show that without the additive correction for T1-weighted protocol parameters, atrophy was underestimated and subject-specific estimates were more narrowly distributed about the population mean. Ventricular CSF is the most consistently estimated brain volume, with a mean of 2.8%/year increase in Cohort 1 and 4.4%/year increase in Cohort 2. An interesting observation was that gray matter volume decreased and white matter volume remained essentially unchanged in both cohorts, suggesting that changes in ventricular CSF volume are a surrogate for changes in gray matter volume. In conclusion, the mixed-effects modeling framework presented here allows effective use of heterogeneously acquired and historical data in the study of brain atrophy in MS, potentially simplifying the design of future single- and multi-site clinical trials and natural history studies.

KW - Brain atrophy

KW - Heterogeneous data

KW - Mixed-effects model

KW - MRI

KW - Multiple sclerosis

UR - http://www.scopus.com/inward/record.url?scp=84883353846&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84883353846&partnerID=8YFLogxK

U2 - 10.1016/j.nicl.2013.08.001

DO - 10.1016/j.nicl.2013.08.001

M3 - Article

C2 - 24179861

AN - SCOPUS:84883353846

VL - 3

SP - 171

EP - 179

JO - NeuroImage: Clinical

JF - NeuroImage: Clinical

SN - 2213-1582

ER -