Summary
Background Brain atrophy occurs in both normal ageing and in multiple sclerosis (MS), but it occurs at a faster rate in MS, where it is the major driver of disability progression. Here, we employed a neuroimaging biomarker of structural brain ageing to explore how MS influences the brain ageing process.
Methods In a longitudinal, multi-centre sample of 3,565 MRI scans in 1,204 MS/clinically isolated syndrome (CIS) patients and 150 healthy controls (HCs) (mean follow-up time: patients 3⋅41 years, HCs 1⋅97 years) we measured ‘brain-predicted age’ using T1-weighted MRI. Brain-predicted age difference (brain-PAD) was calculated as the difference between the brain-predicted age and chronological age. Positive brain-PAD indicates a brain appears older than its chronological age. We compared brain-PAD between MS/CIS patients and HCs, and between disease subtypes. In patients, the relationship between brain-PAD and Expanded Disability Status Scale (EDSS) at study entry and over time was explored.
Findings Adjusted for age, sex, intracranial volume, cohort and scanner effects MS/CIS patients had markedly older-appearing brains than HCs (mean brain-PAD 11⋅8 years [95% CI 9⋅1—14⋅5] versus −0⋅01 [−3⋅0—3⋅0], p<0⋅0001). All MS subtypes had greater brain-PAD scores than HCs, with the oldest-appearing brains in secondary-progressive MS (mean brain-PAD 18⋅0 years [15⋅4—20⋅5], p<0⋅05). At baseline, higher brain-PAD was associated with a higher EDSS, longer time since diagnosis and a younger age at diagnosis. Brain-PAD at study entry significantly predicted time-to-EDSS progression (hazard ratio 1⋅02 [1⋅01—1⋅03], p<0⋅0001): for every 5 years of additional brain-PAD, the risk of progression increased by 14⋅2%.
Interpretation MS increases brain ageing across all MS subtypes. An older-appearing brain at baseline was associated with more rapid disability progression, suggesting ‘brain-age’ could be an individualised prognostic biomarker from a single, cross-sectional assessment.
Funding UK MS Society; National Institute for Health Research University College London Hospitals Biomedical Research Centre.
Research in context
Evidence before this study
We searched Pubmed and Scopus with the terms “multiple sclerosis” and “brain ageing” or “brain age” and “neuroimaging” or “MRI” for studies published before 15th March 2019. This searched return no studies of brain ageing in multiple sclerosis. We also searched the pre-print server for biology, bioRxiv, and found one manuscript deposited, though this study has yet to appear in a peer-reviewed journal. This study found a strong effect of multiple sclerosis on the apparent age of the brain, though was only cross-sectional, was from a single centre, did not consider disease subtypes and did not consider the relevance of clinical characteristics for brain ageing. Therefore, although there is strong prior evidence of the importance of brain atrophy in multiple sclerosis, there was no information on how the nature of this atrophy relates to brain ageing.
Added value of this study
Here we demonstrate for the first time that the progressive atrophy in multiple sclerosis patients results in an acceleration of age-related changes to brain structure. Using a large multi-centre study, our data strongly support the idea that brain ageing is increased in multiple sclerosis, and that this is apparent across disease subtypes, including those with very early disease - Clinically Isolated Syndrome. Of particular value is the demonstration that baseline brain-age can be used to predict future worsening of disability, suggesting that a general index of age-related brain health could have relevance in clinical practice for predicting which patients will go on to experience a more rapidly progressing disease course.
Implications of all the available evidence
Combined with the single other available study, this work shows robust evidence for a cross-sectional influence of multiple sclerosis on the apparent age of the brain, under the brain-age paradigm. This paradigm provides a new approach to considering how multiple sclerosis effects the structure of the brain during ageing, suggesting that multiple sclerosis may result in both disease-specific insults (e.g., lesions) alongside changes that are less specific (e.g., atrophy) and seen in ageing and other diseases. Potentially, treatments that improve brain health during normal ageing could be used to benefit patients with multiple sclerosis. Finally, brain-age may also have prognostic clinical value as a sensitive, if non-specific, biomarker of future health outcomes.
Introduction
In multiple sclerosis (MS), age has been implicated as the dominant driver of disease progression.1 Older age increases the risk of progression,2 irrespective of disease duration, and once progression starts, disability accrual is independent of previous evolution of the disease (presence or not relapses, or relapse rates.3–5 This raises the possibility that MS interacts with some of the neurobiological drivers of brain ageing, leading to acceleration of the process, hastening brain atrophy in some individuals and leading to poorer long-term outcomes.6
That diseases may impact rates of biological ageing has been previously mooted outside of the context of MS. Potentially, a disease has both a specific impact but also may trigger a sequence of events which result in an acceleration of the biological processes seen in normal ageing, both systemically7 and in the brain.8,9
Recently, methods have been developed for measuring the biological ageing of the brain; the so-called ‘brain-age’ paradigm.10 Brain-age uses machine-learning analysis to generate a prediction of an individual’s age (their brain-predicted age), based solely on neuroimaging data (most commonly 3D T1-weighted MRI). The comparison of an individual’s brain-predicted age with their chronological age thus gives an index of whether their brain structure appears ‘older’ or ‘younger’ than would be expected for their age. By subtracting chronological age from brain-predicted age one can derive a brain-predicted age difference (brain-PAD); a simple numerical value in the unit years which shows promise as a biomarker of brain ageing. For example, brain-age has been shown to predict the likelihood of conversion from mild cognitive impairment to Alzheimer’s11,12 as well as the risk of mortality.13 Moreover, there is evidence for increased brain ageing in other neurological conditions contexts: traumatic brain injury,14 HIV,15 Down’s syndrome,16 and epilepsy.17
Here we employ brain-age to assess the relationship between MS disease progression and the brain ageing process. Using longitudinal neuroimaging and clinical outcomes in a large cohort of MS patients and healthy controls (HCs), we tested the following hypotheses: (i) MS patients have older-appearing brains than HCs; (ii) In MS patients, there is a relationship between brain-predicted age difference and disability at study entry; (iii) Brain-predicted age difference increases over time as disabilities worsen; and (iv) Brain-predicted age difference at baseline predicts future disability progression.
Methods
Participants
This study used data collected from seven European MS centres (MAGNIMS: www.magnims.eu) and Imperial College London on n=1,354 participants (table 1), largely overlapping with our previous work (detailed in the appendix table S1).18 Patients had all received a diagnosis of MS according to 2010 McDonald Criteria19 or CIS.20 MS/CIS patients were scored on the Expanded Disability Status Scale (EDSS).21 HCs without history of neurological or psychiatric disorders were also included (n=150). For longitudinal imaging analysis, participants were required to have undergone at least two high-resolution T1-weighted MRI acquired with the same protocol with an interval of ≥1 month.
The final protocol for this study was reviewed and approved by the European MAGNIMS collaboration for analysis of pseudo-anonymized scans and the Imperial NHS Trust (London Riverside Research Ethics Committee: 14/LO/0343).
EDSS progression
Time-to-event, where a progression event was an individual’s progression on the EDSS, was defined as per our previous work18: when a patient showed a longitudinal change of: a 1⋅5-point increase in EDSS if the baseline EDSS was 0; a 1-point increase if baseline EDSS was 1 to 6 inclusive; and a 0⋅5-point increase if EDSS was greater than 6.
Neuroimaging acquisition
Overall, 3,565 T1-weighted MRI scans were used in the study according to local MRI protocols, which used similar acquisition parameters. Thirteen different scanners (Siemens, GE, Philips) were used in patients recruited from 1998 onwards (see appendix table S1).
Machine-learning brain-predicted age analysis
Brain-predicted age calculation followed our previously established protocol.15 In brief, all structural images were pre-processed using SPM12 to generate grey matter (GM), white matter (WM) segmentations. Visual quality control was then conducted to verify segmentation accuracy; all images were included. Segmented GM and WM images were then non-linearly registered to a custom template (based on the training dataset). Finally, images were affine registered to MNI152 space (voxel size = 1⋅5mm3), modulated and smoothed (4mm). Summary volumetric measures of GM, WM, cerebrospinal fluid (CSF) and intracranial volume (ICV) were also generated.
Brain-predicted ages were generated using Pattern Recognition for Neuroimaging Toolbox (PRoNTo v2⋅0, www.mlnl.cs.ucl.ac.uk/pronto) software.22 First, a model of healthy brain ageing was defined: brain volumetric data (from in a separate training dataset, n=2001 healthy people, aged 18-90; appendix table S2) were used as the independent variables in a Gaussian Processes regression, with age as the dependent variable. This regression model achieved a mean absolute error (MAE) of 5⋅02 years, assessed using ten-fold cross-validation, which explained 88% of the variance in chronological age.
Next, the coefficients from the full historical training model (n=2001) were applied to the current test data (i.e., MS/CIS patients and HCs), to generate brain-predicted ages. These values were adjusted to remove age-related variance, by subtracting 3⋅33 and then dividing by 0⋅91 (the intercept and slope of a linear regression of brain-predicted age on chronological age in the training dataset).
Finally, brain-PAD scores were calculated by subtracting chronological age from brain-predicted age and used for subsequent analysis. A positive brain-PAD score indicates that the individual’s brain is predicted to be ‘older’ than their chronological age.
Statistical analysis
Using brain-PAD values further statistical analysis was carried out to test our hypotheses, using R v3⋅4⋅3. A full list of R packages and versions is included in the accompanying R Notebook (appendix). We used linear mixed effects models, enabling incorporation of fixed and random effects predictors to model each given outcome measure. In these models, brain-PAD was used as the outcome variable. Each model included fixed effects of group (e.g., MS/CIS patient versus HCs; MS subtype [CIS, RRMS, SPMS, PPMS]), age, sex and ICV and random effects of MRI scanner field-strength and original study cohort (modelling intercept). Estimated marginal means and confidence intervals from linear models were calculated. This analysis was repeated using data from a single cohort from a single centre (UCL, London), where all MS subtypes were present.
A random effects meta-analysis was conducted to explore the heterogeneity of the group effects on brain-PAD across different study cohorts. Only cohorts that included HCs and MS or CIS patients were included in this analysis.
To establish whether brain volume measurements were driving the variability in brain-PAD, we performed a linear regression with hierarchical partitioning of variance, with brain-PAD as the outcome variable and age, sex, GM, WM and CSF volume as predictors.
Subsequent analyses were conducted to test for fixed-effect influences of EDSS score (MS and CIS patients), and time since clinical diagnosis and age at clinical diagnosis (MS patients only). Model fits were considered using F-tests and post-hoc pairwise comparisons using t-tests or Tukey tests where appropriate.
We explored how longitudinal changes in brain-PAD related to changes in disability over time in two ways: (i) by correlating annualised change in brain-PAD (i.e., the difference between first measured brain-PAD and last brain-PAD, divided by the interval in years) with the annualised change in EDSS score; (ii) by using linear mixed effects models to investigate group (MS/CIS vs., HCs; MS subtype) by time interactions. These analyses included a random effect of participant (modelling slope and intercept), alongside age, sex, ICV scanner and cohort effects.
Survival analysis, using a Cox proportional hazards regression, was used to test whether baseline brain-PAD predicted time-to-EDSS progression, including age at baseline MRI and sex as covariates.
We investigated the impact of MS lesions on brain-PAD in MS. Using cross-sectional data from a subset of n=575 MS/CIS patients, for which manually-annotated lesion maps were available, we explored the relationship between MS lesions and measurements of brain-PAD, using the FSL lesion-filling algorithm,23 by artificially removing lesions from T1-weighted MRI scans. Both ‘lesion-filled’ and ‘unfilled’ scans were run through the brain-age prediction procedure, then resulting brain-PAD scores compared.
Role of the funding source
The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.
Results
Multiple sclerosis is associated with older appearing brains
The MAGNIMS sample forms part of a well-characterised population (table 1). The combined cohort involves patients from six countries with a mean follow-up of 3⋅41 years in patients.
Patients with MS/CIS had markedly greater brain-PAD scores at time of initial MRI scan compared to HCs (estimated marginal means 11⋅8 years, [95% CI 9⋅1–14⋅5] versus −0⋅01 [95% CI −3⋅0–3⋅0]). When adjusted for the age, sex, intracranial volume, cohort and scanner effects, there was a statistically significant group mean difference in brain-PAD of 11⋅8 years (95% CI 9⋅9–13⋅8, p<0⋅0001).
Though there is considerable heterogeneity between the study cohorts, due to clinical characteristics and technical factors (e.g., MRI scanner system), the difference between MS/CIS and HCs was robust in a random-effects meta-analysis of a subset of the data; six London cohorts that included both MS/CIS patients and HCs (figure 1A). The heterogeneity in the group differences was substantial (I2 = 59%, [95% CI 3–91%]).
MS subtype (CIS, RRMS, SPMS, PPMS) significantly influenced brain-PAD (F3,802⋅25 = 29⋅9, p<0⋅0001, figure 1B). Estimated marginal mean brain-PAD per subtype were: CIS 6⋅3 years [95% CI 3⋅9–8⋅8], RRMS 12⋅4 years [95% CI 10⋅3–14⋅5], SPMS 18⋅0 years [95% CI 15⋅4–20⋅5], and PPMS 12⋅4 years [95% CI 9⋅7–15⋅2]. Post-hoc pairwise group comparison (appendix table S3) showed statistically significant differences (p<0⋅05) in brain-PAD between each subtype and HCs, and between CIS patients and each of the three MS groups (RRMS, SPMS, PPMS). SPMS patients showed significantly greater brain-PAD compared to both RRMS and PPMS patients. The difference in brain-PAD between PPMS and RRMS was not statistically significant (p=0⋅62). The findings of differences in brain-PAD between MS subtypes were replicated in a single cohort from a single centre, where all subtypes were present (cohort UCL3, figure 1C). Brain-PAD scores and corresponding T1-weighted MRI scans of individual female participants with similar ages, but with different subtypes of MS, are illustrated in figure 1D.
The relationship between lesions, brain volume and brain-PAD
We considered the impact of lesions of brain-PAD, by comparing brain-PAD values on a single MRI scan from n=575 patients with both a lesion-filled and unfilled version of the same image. The correlation between brain-predicted age using filled and unfilled scans was r=0⋅99, p<0⋅0001 (appendix figure S1A) suggesting that the presence of lesions did not overly influence the brain-PAD values used throughout the study (which were unfilled). A Bland-Altman plot showed a mean difference between filled and unfilled scans was −0⋅28 ±1⋅29 years with no systemic bias caused by lesion filling evident, though there was increased variability between ages 60-70 years (appendix figure S1B).
When we examined whether brain volume measurements were driving the variability in brain-PAD, we found that the combination of chronological age, sex, GM, WM and CSF volume explained about half of the variation in brain-PAD (adjusted R2=0⋅48) (appendix table S4). Age (9% variance explained), GM (15%) and CSF (20%) volume were major contributors to variance in brain-PAD.
Brain-PAD at baseline is associated with disability, age at diagnosis, and time since clinical diagnosis
At baseline, a higher brain-PAD was associated with higher disability, as measured by the EDSS, when adjusting for age, sex, ICV, scanner and cohort: for every 1⋅74 years increase in brain-PAD, the EDSS increased by one point (95% CI 1⋅39–2⋅09], p<0⋅0001). This effect was consistent across the MS subtypes with no statistically significant interaction between subtype and EDSS score (F3,1159⋅6 = 1⋅12, p=0⋅34; figure 2A). With the same adjustments, a higher brain-PAD was associated with both younger age at diagnosis and longer time since diagnosis: for every year increase in brain-PAD, the age at diagnosis was reduced by 0⋅45 years (95% CI −0⋅55–−0⋅36], p<0⋅0001); for every 0⋅48 year increase in brain-PAD, the time since diagnosis increased by one year (95% CI 0⋅40–0⋅57, p<0⋅0001). There was an interaction between subtype (RRMS, PPMS and SPMS) and age at diagnosis (F2,883⋅9 = 3⋅20, p=0⋅041; figure 2B), driven by the presence of stronger relationships between brain-PAD and age at diagnosis in PPMS (slope beta −0⋅51) and SPMS (beta−0⋅57) compared to RRMS (beta −0⋅36), though all were significant (p<0⋅001). For time since diagnosis, the interaction was also significant (F2,690⋅5 = 3⋅61, p=0⋅028; figure 2C), driven by the presence of relationships in RRMS (beta 0⋅48, p<0⋅0001) and SPMS (beta 0⋅26, p=0⋅01), not observed in PPMS (beta 0⋅12, p=0⋅47).
Brain-PAD increase over time correlates with EDSS worsening
In patients who had two or more scans (n=1155), there was a significant positive correlation between annualised change in brain-PAD and annualised change in EDSS (Pearson’s r=0⋅26, p<0⋅0001). There was a significant interaction between EDSS change and disease subtype, when predicting brain-PAD change in linear model (F4,1092 = 24⋅5, p=0⋅009). The slopes were positive in CIS (beta 0⋅84, p=0⋅0001) and RRMS (beta 1⋅25, p<0⋅0001), though flatter in PPMS (beta 0⋅59, p=0⋅090) and negative (though not significant) in SPMS (beta −0⋅70, p=0⋅29; figure 3). To explore the latter finding post-hoc, correlated baseline brain-PAD with the number of follow-up scans completed. This showed a significant inverse correlation (n=104, Spearman’s rho=−0⋅29, p=0⋅0028).
Brain-predicted age difference at first scan predicts EDSS worsening
In patients who had EDSS assessed at ≥2 time-points (n=1147), baseline brain-PAD significantly predicted EDSS worsening. Of these patients, 303 (26⋅5%) experienced EDSS worsening during the follow-up period. Using a Cox proportional-hazards regression model, adjusted for age and sex, the hazard ratio for brain-PAD was 1⋅027 (95% CI 1⋅016–1⋅038, p<0⋅0001). In other words, for every 5 years of additional brain-PAD, there was a 14⋅1% increased chance of EDSS progression during follow-up. Survival curves grouped by a median split of baseline brain-PAD illustrate the differing rates of ‘survival’ prior to EDSS progression (figure 4).
MS accelerates longitudinal increase in brain-PAD
A total of 1266 participants had two or more MRI scans (MS/CIS=1155, HCs=111). This included 573 with three or more scans (MS/CIS=509, HCs=64). When using these data, we found a significant interaction between group and time (F1,1325⋅6 = 5⋅37, p=0⋅021) and between MS subtypes and time (F4,938⋅25 = 5⋅35, p<0⋅0001), when adjusting for age, gender, ICV, cohort and scanner status (figure 5). This indicated that the annual rate of increase in brain-PAD over time was faster in MS/CIS than in HCs, and significantly different between MS subtypes. The estimate marginal mean annualised rates of increase in brain-PAD per group was as follows: HCs −0⋅98 [95% CI −2⋅03–0⋅07], CIS −0⋅14 [95% CI −1⋅07–0⋅78], RRMS 0⋅93 [95% CI 0⋅21–1⋅66], SPMS 0⋅34 [95% CI −0⋅69–1⋅37], PPMS 1⋅21 [95% CI 0⋅16–2⋅25], all CIS/MIS 0⋅70 [95% CI 0⋅01–1⋅39].
Discussion
By assessing the relationship between MS disease progression and the normal brain ageing process, we have found that patients with MS have an older appearing brain (i.e., higher brain-PAD) compared to controls. As the disease develops from a clinically isolated episode to relapsing and then secondary progressive MS, brain-PAD increases. A single baseline brain-PAD was independently associated with higher disability (measured by EDSS), younger age at diagnosis and longer time since diagnosis, irrespectively of disease phenotype. Using scans performed at multiple sites in different scanners we observed that longitudinal brain-PAD increases correlate with worsening disability and that measures of brain-PAD at baseline predict future disability accumulation. In the whole cohort, we show that measures of brain-PAD over time increase with respect to chronological age, implying an accelerated ageing process, particularly in RRMS and PPMS.
In a life-long disease, the accumulation of neurological disability is the main clinical and societal burden,24 estimated to cost $10⋅6 billion/year in the USA.25 Tracking disease evolution is hampered by the lack of a simple and powerful outcome measure. MRI-assessed brain atrophy is a surrogate outcome for this process, but the need for precise longitudinal assessments, usually over at least a 12-month interval, reduces the feasibility of use. Here, we demonstrate that with a single T1-weighted MRI, brain-PAD values can index elements of MS disease progression. Firstly, we show that a single point estimate can place a patient’s disease and disability in context of their age. This has been lacking with current techniques but is achieved because brain-PAD measures change relative to a model of the healthy ageing process. Our results suggest that the ‘brain-age’ framework can provide informative data without the need for longitudinal scans.26 Secondly, we demonstrate that a single measure can give prognostic value for disability accumulation. This can allow us to better contextualise the impact of the disease on an individual, measured at a single time point, and then chart different pathways of neurodegeneration in MS. Brain atrophy has undoubted utility in capturing elements of disease progression, but is currently difficult to utilise in clinical practice.27 Here we demonstrate that machine learning technique provides a biomarker of structural brain ageing that enables prediction of disability worsening. Thus, the ability to make prognostic predictions from cross-sectional data could prove highly valuable to facilitate early use of therapy to prevent future disability accumulation.28
The brain-age paradigm has been applied widely in neuropsychiatric diseases,10 though only recently in MS.29 Kaufmann and colleague’s analysis (n=254) showed a strong effect of MS on brain-age (mean increase 5⋅6 years), though was only cross-sectional and did not explore subtypes separately. Here we go further, utilising serial MRI scans that were carried out over 15 years in a wide range of settings – different countries, institutions and scanners. The mean magnitude of the apparent brain ageing we observed MS (11.8 years) is greater than has been reported in dementia (9 years),11 epilepsy (4⋅5 years)17 or after a traumatic brain injury (4⋅7 years).14 We show that brain-PAD increases faster than chronological age in MS/CIS patients, suggesting an accelerating neurodegenerative process. Interestingly, brain-PAD did not increase longitudinally in SPMS patients; potentially due to a survivor bias or a floor effect in this group, whereby those patients with rapidly deteriorating disease did not return for longitudinal follow-up. Evidence for this comes from the inverse correlation between brain-PAD at baseline and the number of follow-up scans acquired in SPMS patients.
We addressed some potential issues with the use of a non-specific ageing biomarker like brain age for the assessment of MS. Brain lesions, the overt MRI marker of MS disease activity, had minimal impact of the brain-PAD measurement in MS. Brain volumes, perhaps unsurprisingly, were strongly correlated with brain-PAD; GM, WM and CSF volume measures combined explained ~49% of the variance in brain-PAD. Evidently, a substantial proportion of variation in brain-PAD is not explained by demographic and MRI characteristics and might be unique to ‘brain-age’. In particular, ‘brain-age’ incorporates voxelwise MRI data in the statistical model, thereby capturing more information than when using summary statistics. This means that more widespread and distributed patterns of features (i.e., voxelwise GM and WM volumes) can contribute to the age-prediction model, capturing elements of cortical thinning, sulcal widening and ventricular enlargement, alongside more macroscopic loss of tissue volume.
Our study has some strengths and weaknesses. The sample size for both training and test sets is relatively large but one potential limitation is the multiple sources of training data, though previous work has shown high between-scanner reliability.30 Thus, if it is to be used as a single value this would need to be in the context of individual scanner performance. Comprehensive biomedical data were not available on all these individuals in the training dataset, meaning some may have had undetected health conditions. However, individuals in this sample were screened according to various criteria to ensure the absence of manifest neurological, psychiatric or major medical health issues. We were not able to determine the impact of therapy in this study as it was not a randomised trial and worsening disease drives use of therapy, the effectiveness of which is challenging to determine. However, the majority of the current study sample were on not receiving therapy at baseline, thus therapeutic effects are unlikely to have confounded our results.
This work supports the use of the ‘brain-age’ paradigm in MS. We propose that brain-predicted age has potential value for: 1) MS disease monitoring; potentially capturing the progressive processes that start early on in all disease phenotypes including CIS. 2) Integrating MRI measures of brain injury in MS in a wide range of centres and different scanners. 3) Conveying complex neuroanatomical information in a conceptually simple and intuitive manner. 4) Assessing both current brain health and prognosis. 5) Aiding clinical trial design, by stratifying enrolment based on high brain-PAD, or using brain-PAD as a surrogate outcome measure, reflecting age-associated neurodegeneration. Further work is needed to determine its utility in larger clinical cohorts, but its ease of use makes it an exciting candidate for such cohorts. Further work is needed to improve the anatomical interpretability of brain-age, both in general and specifically to MS. Ultimately, this may offer insight into an individual’s disease course, in line with the move towards precision medicine in the treatment of MS.
Appendix
MAGNIMS Study Group: Steering Committee Members
Alex Rovira (co-chair): MR Unit and Section of Neuroradiology, Department of Radiology, Hospital Universitari Vall d’Hebron, Universitat Autònoma de Barcelona, Barcelona, Spain
Christian Enzinger (co-chair): Department of Neurology, Medical University of Graz, Graz, Austria
Frederik Barkhof: Queen Square Multiple Sclerosis Centre, UCL Institute of Neurology, University College London, London, UK
Olga Ciccarelli: Queen Square Multiple Sclerosis Centre, UCL Institute of Neurology, University College London, London, UK
Massimo Filippi: Neuroimaging Research Unit, Institute of Experimental Neurology, Division of Neuroscience, San Raffaele Scientific Institute, Vita-Salute San Raffaele University, Milan, Italy
Nicola De Stefano: Department of Medicine, Surgery and Neuroscience, University of Siena, Siena, Italy
Ludwig Kappos: Department of Neurology, University Hospital, Kantonsspital, Basel, Switzerland
Jette Frederiksen: The MS Clinic, Department of Neurology, University of Copenhagen, Glostrup Hospital, Denmark
Jaqueline Palace: Centre for Functional Magnetic Resonance Imaging of the Brain, University of Oxford, UK
Maria A Rocca: Neuroimaging Research Unit, Institute of Experimental Neurology, Division of Neuroscience, San Raffaele Scientific Institute, Vita-Salute San Raffaele University, Milan, Italy
Jaume Sastre-Garriga: Department of Neurology/Neuroimmunology, Multiple Sclerosis Centre of Catalonia (CEMCAT), Hospital Universitari Vall d’Hebron, Universitat Autònoma de Barcelona, Barcelona, Spain
Hugo Vrenken: Department of Radiology and Nuclear Medicine, MS Center Amsterdam, Amsterdam, The Netherlands
Tarek Yousry: NMR Research Unit, Institute of Neurology, University College London, London, UK
Claudio Gasperini: Department of Neurology and Psychiatry, University of Rome Sapienza, Rome, Italy.
R Notebook used for statistical analysis
James Cole - March 2019. Built with R version 3.5.2
This is Notebook contains the final brain age analysis of MS patient data and controls from the UCL cohort, the MAGNIMS consortium and the Imperial College London PET study (n=25). The analysis uses brain-predicted age difference (brain-PAD) to look at brain ageing in the context of MS. The brain-PAD values were generated in PRONTO, using an independent healthy (n=2001) training dataset, and the values were corrected for proportional bias using the intercept and slope of the age by brain-predicted age regression in the training dataset.
Initial set up of analysis
Clear workspace, load libraries, set colour palette
Get data from CSV and define longitudinal data.frames
Exclude participants with errors in the database & correct time since diagnosis errors
There were 13 subjects with 38 scans excluded in total. Data entry errors in original spreadsheet; age at baseline not consistent within subject.
Load data for lesion filling analysis
Generate baseline only data.frame and show data frame structure
Basic stats
Total number of subjects, and by group
The total number of subjects included was n = 1354
The total number of MS patients (including CIS) was n = 1204 and healthy controls was n = 150
Number of scans in total and per group
Total number of scans = 3565
Number of people with 2 or more scans
Number of people with 3 or more scans
Generate demographics table using qwarps2
Need to get treatment NAs using table()
Need to get length of follow-up from longitudinal database
Baseline brain-age analysis
Estimated marginal means
Generate EMMs for all MS/CIS and healthy controls. LME adjusting for age, gender, ICV, cohort and scanner status.
Effects of cohort and scanner status on brain-PAD.
The significant interaction motivates nesting in the LME.
Meta-analysis looking at all the separate cohorts with MS/CIS patients and controls
Check which cohorts contain healthy controls and patients.
Create data.frame with summary data appropriate for meta-analysis.
Run meta-analysis using the metafor package, to fit a random-effects meta-analysis using REML.
Forest plot of results
Linear regression analysis restricted to cohort UCL3, includes covariates: age, gender, intracranial volume.
Main result
LME model to predict brain-PAD based on group
Lesion filling
To establish whether using the FSL lesion filling software influences brain-predicted age values. This analysis was conducted only in UCL patients.
Correlation between brain-predicted age from filled and unfilled images: Pearson’s r = 0.994. Median absolute error (MAE) between brain-predicted age from filled and unfilled images = 0.3717 years. Mean difference between brain-predicted age from filled and unfilled images = 0.28 years.
Bland-Altman plot
LME model predicted brain-PAD based on subtype
This analysis excluded controls.
Brain-PAD estimated marginal means for subtypes
Generate EMMs for all MS subtypes. LME adjusting for age, gender, ICV, cohort and scanner status.
Brain-PAD boxplot by MS subtype
Brain-PAD boxplot by MS subtype in cohort UCL3 only
Post-hoc pairwise brain-PAD comparison of subtypes
Brain-PAD by subtype descriptive statistics
Correlates of brain-PAD at baseline
EDSS score, an index of disability
LME accounting for fixed effects of age at baseline, gender, ICV and random effects of Cohort and scanner status.
When predicting brain-PAD in a LME model, the effect of EDSS at baseline beta = 1.74, 95% CI = 1.39, 2.09, p = < 2.22e-16.
Test for interaction between subtype and EDSS on brain-PAD:
Use simple slopes from jtools to extract adjusted slopes for each subtype.
Use interact_plot() from jtools to plot the adjusted slopes per group.
Age at diagnosis
LME accounting for fixed effects of age at baseline, gender, ICV and random effects of Cohort and scanner status. Exclude CIS patients and healthy controls.
When predicting brain-PAD in a LME model, the effect of age at diagnosis at baseline beta = −0.45, 95% CI = −0.55, −0.36, p = < 2.22e-16.
Test for interaction between subtype and age at diagnosis on brain-PAD
Use simple slopes from jtools to extract adjusted slopes for each subtype.
Time since diagnosis
LME accounting for fixed effects of age at baseline, gender, ICV and random effects of Cohort and scanner status. Exclude controls, CIS patients and anyone with a time since diagnosis = 0.
When predicting brain-PAD in a LME model, the effect of time since diagnosis at baseline beta = 0.48, 95% CI = 0.4, 0.57, p = < 2.22e-16.
Test for interaction between subtype and time since diagnosis on brain-PAD
Use simple slopes from jtools to extract adjusted slopes for each subtype.
Plot
EDSS progression survival analysis
Based on Arman Eshaghi’s code used in Eshaghi et al., 2018 Annals of Neurology.
Numbers of EDSS progressors
The number of MS patients with >= 2 EDSS scores was 1143.
Run survival analysis
The hazard ratio for brain-PAD on time-to-disease-progression was HR (95% CI) = 1.027, 1.016, 1.038. That means for every additional +1 year of brain-PAD there is a 1.027% increase in the likelihood of EDSS progression. Extrapolated over 5 years of brain-PAD, there is a 1.141 increase in the likelihood of EDSS progression.
Time-to-EDSS progression Kaplan-Meier plots
Based on a median split of brain-PAD. The median value = 9.68 years.
Longitudinal brain-age analysis
The total number of people with two or more scans was n = 1266.
determine change in brain-PAD from baseline to last follow-up
Mean annualised rates of change in brain-PAD per group
determine change in EDSS from baseline to last follow-up
Mean annualised rates of change in EDSS per group
Correlation between annualised EDSS change and brain-PAD change
Interaction between subtype and EDSS change
Use jtools package to get slopes from the model, per subtype.
Plot
Correlate baseline brain-PAD with the number of follow-up scans completed in the n=104 with >1 scan.
Longitudinal brain-predicted age trajectories
Interaction between group and time
Brain-PAD change EMMs, using annualised difference between baseline and final follow-up
Generate EMMs for all groups and MS subtypes. LME adjusting for age, gender, ICV, cohort and scanner status.
Generate EMMs for HCs and MS/CIS combined. LME adjusting for age, gender, ICV, cohort and scanner status.
Slopes per group
Controls
MS patients
Longitudinal brain-PAD by interval plots
Supplementary analysis
Hierarchical partitioning of brain-PAD
Acknowledgements
JC is funded by a UKRI/MRC Innovation Fellowship. OC, RN, FB and DC acknowledge the National Institute for Health Research University College London Hospitals Biomedical Research Centre. RN acknowledges the National Institute for Health Research Imperial College London Hospitals Biomedical Research Centre. AE received the McDonald Fellowship from Multiple Sclerosis International Federation (MSIF, http://www.msif.org) and ECTRIMS-MAGNIMS fellowship.