Clinical Neurophysiology 123 (2012) 406–410

Contents lists available at ScienceDirect

Clinical Neurophysiology
journal homepage: www.elsevier.com/locate/clinph

Combined evoked potentials as markers and predictors of disability in early

multiple sclerosis
Regina Schlaeger a, Marcus D’Souza a, Christian Schindler b,c, Leticia Grize b,c, Ludwig Kappos a, Peter Fuhr a,⇑
a
Department of Neurology, University Hospital Basel, Switzerland
b
Swiss Tropical and Public Health Institute, Socinstrasse 57, P.O. Box, 4002 Basel, Switzerland
c
University of Basel, Petersplatz 1, CH-4003 Basel, Switzerland

See Editorial, pages 221–222

a r t i c l e i n f o h i g h l i g h t s

Article history:  Multimodal evoked potentials can be summarized in a rational number which correlates well with
Available online 22 July 2011 clinical assessment in groups of patients with early MS.
 Prediction of the disease course over 3 years in patient groups is possible with a correlation coefficient
Keywords: of 0.7.
Evoked potentials  Multimodal evoked potentials bear the potential to improve clinical trial design and counselling of indi-
Prognosis
vidual patients.
Multiple sclerosis

a b s t r a c t
Objective: To prospectively assess combined evoked potentials (EP) as markers and predictors of the dis-
ease course of early MS over 3 years.
Methods: Fifty patients in the early phase of relapsing remitting MS prospectively received visual,
somatosensory and motor EP and EDSS assessments at baseline (T1) and at 6 months intervals during
3 years. Spearman rank correlation was used to determine the relationship between z-transformed
EP-latencies (z-EPL) and EDSS. Multivariable linear regression was performed to predict EDSS at year
3 (T7) in function of z-EPLT1. Validity of the models was assessed using group cross-validation.
Results: At each of the seven points in time, EDSS correlated with the sum of z-EPL (0.64 6 rho 6 0.79,
p < 0.001). The change of the sum of z-EPLT7–T1 correlated with the change of EDSST7–T1 (rho = 0.51,
p = 0.001). EDSST7 as predicted by the sum of z-scores of EP latencies or by the number of pathological
EP results at baseline correlated with the observed clinical values after 3 years (rho > 0.70, p < 0.001,
for both measures).
Conclusions: Multimodal EPs correlate well with clinical disability in cross-sectional and longitudinal
comparison in early MS and allow prediction of disease evolution over 3 years.
Significance: EPs seem well suited as markers of the disease course in early MS in clinical trials and bear
potential for supporting decision-finding in individual patients.
Ó 2011 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights
reserved.

1. Introduction However, reliable monitoring and prediction of the disease

While early and reliable diagnosis of multiple sclerosis (MS) is
now possible (Polman et al., 2005), the pace of disability progres-
sion in individual patients still remains highly unpredictable.
⇑ Corresponding author. Address: Department of Neurology, University Hospital
Basel, Petersgraben 4, CH-4031 Basel, Switzerland. Tel.: +41 612654167; fax: +41
612654100.
E-mail addresses: pfuhr@uhbs.ch, Peter.Fuhr@unibas.ch (P. Fuhr).
course is a requirement to individually tailor disease-modifying
strategies as well as to select appropriate candidates for therapeutic
trials. This is of paramount importance at early stages of the
disease, when clinical variability (as expressed by range of EDSS
scores) is still relatively low. Clinical assessments alone – e.g.
documentation of the number and severity of relapses or of the
expanded disability status score (EDSS) – do not allow prediction
of the clinical course (Runmarker et al., 1994; Confavreux et al.,
2000, 2003; Young et al., 2006).

1388-2457/$36.00 Ó 2011 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.clinph.2011.06.021
 R. Schlaeger et al. / Clinical Neurophysiology 123 (2012) 406–410
Conventional magnetic resonance imaging is by far the most
extensively assessed biomarker in MS trials, despite poor correla-
tions with disability progression in MS once the diagnosis is estab-
lished – a well-known phenomenon in multiple sclerosis often
labeled as ‘‘clinico-radiological paradoxon’’ (Barkhof, 2002). More-
over, the predictive value of conventional MRI beyond the time of
confirmation of diagnosis is limited (Daumer et al., 2009; Barkhof
and Filippi, 2009).
Evoked potentials (EP) provide quantitative information on
the functional integrity of well-defined pathways of the central
nervous system. In an earlier study, we found significant correla-
tions between a numerically scaled score derived from visual and
motor EP and clinical assessment in cross-sectional and longitudi-
nal comparison, and also between visual and motor EP at baseline
and clinical development over the following 2 years in a relatively
small group of patients at heterogeneous stages of the disease
(Fuhr et al., 2001).
Based on these findings, the aims of the present study were (1)
to reproduce the previous results in a different group of patients in
a relatively early stage of the disease, and (2) to determine whether
including somatosensory evoked potentials increases the predic-
tive strength of multimodal EP for characterization of the disease
course, and (3) to assess whether multimodal EP allow prediction
of the disease course already at an early stage when therapeutical
decisions have the strongest impact.
2. Methods
2.1. Patients and clinical examination
In this prospective study the inclusion criteria comprised either
a diagnosis of definite MS according to McDonald criteria (McDon-
ald et al., 2001), Poser criteria (Poser et al., 1983) or a diagnosis of
clinically isolated syndrome (CIS) with clinical or paraclinical signs
of spatial dissemination but lack of temporal dissemination of the
disease process; disease duration of less than 10 years since first
symptoms, baseline EDSS 0–3.5, age 18–55 years, either no or sta-
ble immunomodulatory therapy for at least 3 months, and written
informed consent. Patients with contraindications for magnetic
evoked potentials, relapses or steroid treatment within 6 weeks
prior to baseline examination, patients suffering from additional
neurological diseases potentially interfering with MS symptoms,
or patients suffering from severe additional other diseases were
excluded. The study was approved by the Local Ethics Committee
and patients provided written informed consent before being
included.
Investigations consisted of Neurostatus, a standardized neuro-
logical examination and assessment of the EDSS (Kurtzke, 1983;
Kappos et al., 2007), visual EPs, motor EPs to upper and lower
Table 1
Characteristics of patients with and without immunomodulatory or immunosuppressive treatment.
Parameter
EDSST1
EDSST7
DEDSST7–T1
Number of relapses prior to study entry
Number of relapses in 2 years prior to study entry
Disease duration (since first symptoms)
Rs-values of z-EPLT1
DRs-values of z-EPLT7–T1
Rs-values of z-EPLT7
Abbreviations: T1, baseline; T7, 3 years follow-up; s-latency value, z-score of latency value (right) + z-score of latency value (left); z-EPL, z-transformed
evoked potential latencies (CMCTUE, CMCTLE, P100, N13–N20, N22–P40 at the respective time points).
a
Immunomodulatory or immunosuppressive treatment between T1 and T7, except steroid pulse therapy.
407
extremities as well as median and tibial nerve somatosensory
EPs at entry (T1), at 6 (T2), 12 (T3), 18 (T4), 24 (T5), 30 (T6) and
36 months (T7). If a patient had a relapse at the assigned study vis-
it, the electrophysiological and clinical examination was postponed
to at least 6 weeks after the first relapse symptoms.
Patients who decided to discontinue the study before T7 were
again invited to our center for a comprehensive examination or
underwent a structured standardized telephone interview for
determination of the EDSS by phone (Lechner-Scott et al., 2003)
at T5 and T7.
The study included 47 participants with relapsing RRMS, and
three participants with CIS. At baseline the participants had a mean
age of 37 years (range 20.2–54.6 years), a mean disease duration
since first manifestation of 5.36 years (SD 2.81). Thirty-two
patients were female. At baseline, 32 patients (64%) were receiving
immunomodulatory treatments (Glatirameracetate, Interferon-b),
one patient (2%) immunosuppressive treatment (Mitoxantrone)
and 17 patients had no treatment (34%), six patients received more
than one treatment between T1 and T7. A description of patients
according to treatment is given in Table 1.
Patients with and without immunomodulatory or immunosup-
pressive treatment did not differ in terms of EDSS and disease
duration at entry. Patients with treatment had more relapses prior
to entry into the study, and more abnormal baseline EP values;
however, these differences were not statistically significant.
2.2. Evoked potentials
Motor evoked potentials (MEPs) were recorded from the pre-
innervated abductor digiti minimi and tibialis anterior muscles
bilaterally. Magnetic stimuli were delivered to the hand and leg
areas of the motor cortex with a Magstim 200 device (The Magstim
Company Ltd., Whitland, UK) via a round coil with an inner diam-
eter of 9 cm using maximal output of the stimulator (2.2 T). For
spinal stimulation the rim of the coil was placed over the seventh
cervical and fifth lumbar vertebra. When MEPs were identifiable,
the shortest onset latency out of eight stimulations was deter-
mined and used for calculating the central motor conduction time
(CMCT). CMCT was assessed according to published normal values
(Stoehr, 2005).
Visual evoked potentials (VEPs) were recorded from an active
electrode placed 3 cm above Oz and a reference electrode at Fz.
Pattern reversal stimulation was presented to each eye separately
at a frequency of 0.5 Hz. When VEPs were identifiable the peak la-
tency of the P100 was determined. P100 latency was assessed
according to normal values determined in our laboratory (normal
mean: 100 ms, SD: 5.3 ms).
Median and tibial nerve somatosensory evoked potentials
(SSEPs) were obtained by stimulating the median nerve at the
Patients without therapy Patients with therapya Mann–Whitney U-test
Mean ± SD Mean ± SD p-Value
2 ± 0.64 2.02 ± 0.94 0.82
2.38 ± 1.05 2.65 ± 1.43 0.63
0.38 ± 0.57 0.64 ± 0.95 0.38
3.08 ± 1.56 5.21 ± 3.81 0.07
1.64 ± 1.80 2.13 ± 1.36 0.19
5.36 ± 3.03 5.33 ± 2.84 0.99
14.64 ± 27.63 20.10 ± 28.09 0.44
10.88 ± 24.17 5.56 ± 18.50 0.96
29.03 ± 48.40 23.44 ± 33.86 0.70
408 R. Schlaeger et al. / Clinical Neurophysiology 123 (2012) 406–410
wrist and the posterior tibial nerve at the ankle, respectively: the
stimulus intensity exceeded visual motor threshold by about
3 mA, the stimulus frequency was 3 Hz. For the median nerve
SSEPs active recording electrodes were placed subcutaneously be-
tween cervical vertebra 6 and 7 and 7 cm lateral and 2 cm posterior
to Cz with the reference electrode placed at Fz. For tibial nerve
SSEPs active recording electrodes were placed between the 12th
thoracic and first lumbar vertebra and 3 cm posterior to Cz with
the reference electrodes placed subcutaneously near the anterior
superior iliac spine and Fz, respectively. When cortical components
were identifiable the latency differences N13a–N20 and N22–P40
were calculated. Median nerve SSEPs were assessed according to
published normal values (Stoehr, 2005), tibial SSEPs were evalu-
ated according to height dependent normal values determined in
our laboratory (normal mean/SD: height >170 cm: 17.28 ms/
1.42 ms; height 6170 cm: 16.60 ms/1.25 ms).
In general, values exceeding the normal mean by >2.5 SD were
regarded as pathological.
When the cortical components of MEPs, SSEPs or VEPs were not
identifiable, the longest central conduction time observed in the
whole cohort and within the same EP modality served as a substi-
tute, since such results were considered to be at least as patholog-
ical as the one with the longest measurable latency. This procedure
allowed us to also include data of patients with the most patholog-
ical results.
2.3. Statistical analysis
Details on the definitions of variables and regression models are
given in the Appendix A. EP data were summarized (a) as the num-
ber of pathological results outside the normal range as defined
above, counting VEPs on each side, and SSEPs and MEPs at each
limb separately, and (b) as the sum of the s-values of CMCTUE,
CMCTLE, P100, N13–N20, N22–P40 (Rs-values of z-EPL) with
s-values as the sum for each latency measure, of its left and right
side z-score.
These measures were defined separately for each time point.
z-Scores were computed using the mean and standard deviation
from a normal population of healthy subjects, as referenced in
the previous section, to reflect the extent of damage independently
from group composition. The association of EP data at baseline
(T1), EDSS after 3 years (EDSST7) and change of EDSS (DEDSST7–T1)
was examined using Spearman rank correlation. The relationship
of the disease course and EP-values at entry was further assessed
using multivariable linear regression analysis. Based on the
results of the precursor study, the following potentially predictive
variables were examined for association with EDSST7: s-P100T1,
s-CMCTUE,T1, s-CMCTLE,T1, s-N13–N20T1, s-N22–P40T1 and therapy
(with therapy as a binary variable (no therapy/any kind of
immunomodulatory or immunosuppressive therapy between T1
and T7, except steroid pulse therapy)).
Table 2
Longitudinal correlations between evoked potentials and EDSS (n = 45).
Variable
R(s-P100, s-CMCTUE, s-CMCTLE, s-N13–N20, s-N22–P40)T1 versus EDSST7
Number of pathological EP resultsT1 versus EDSST7
DR(s-P100, s-CMCTUE, s-CMCTLE, s-N13–N20, s-N22–P40)T7–T1 versus DEDSST7–T1
DNumber of pathological EP resultsT7–T1 versus DEDSST7–T1
R(s-P100, s-CMCTUE, s-CMCTLE, s-N13–N20, s-N22–P40)T1 versus DEDSST7–T1
Number of pathological EP resultsT1 versus DEDSST7–T1
EDSST1 versus DEDSST7–T1
Abbreviations: T1, baseline; T7, 3 years follow-up; UE, upper extremities; LE, lower extremities; s-latency value, z-score of latency value
(right) + z-score of latency value (left).
a
Spearman rank correlation.
As the variable therapy did not show a significant effect, it was
omitted from the model resulting in the Model 1.
In analogy to the model published earlier (Fuhr et al., 2001) a
second model (Model 2) was calculated – containing only VEP
and MEP data – to assess whether including SSEP information (as
in Model 1) improves prediction of EDSST7.
These linear models may predict values for the EDSST7 higher
than the possible maximum. Therefore a non-linear model using
a logistic function and providing predictions within the range of
possible values of the EDSS was also considered (Model 1a).
As an alternative to the model with the sum of z-scores of
evoked potentials we also considered a model with the number
of pathological results (Model 3).
Mean squared error, R2 and Akaike criteria (AIC) for the models
were calculated and compared. The models were assessed using
group cross-validation.
Statistical analysis was done with SPSS version 14.0 (Chicago,
IL: SPSS Inc. 2005) and SAS version 9.2 (Cary, N.C.: SAS Institute
2002).
3. Results
3.1. Cross-sectional comparison
At each of the seven points in time, EDSS values correlated with
the sum of z-EPL (0.64 6 rhos 6 0.79, p < 0.001) as well as with the
number of pathological EP results (0.62 6 rhos 6 0.73, p < 0.001).
3.2. Longitudinal comparison
The change in the sum of z-EPL over 3 years correlated with the
change in EDSS over this period (rhos = 0.51, p = 0.001). The change
of the number of pathological results over 3 years showed a trend
towards a positive correlation with the change in EDSS over this
period (rhos = 0.34, p = 0.035). The sum of z-EPL at baseline corre-
lated with EDSS at year 3 (rhos = 0.70, p < 0.001) and with change
in EDSS over 3 years (rhos = 0.35, p = 0.02), whereas EDSS at base-
line did not correlate with change in EDSS over 3 years (Table 2).
3.3. Predictive value of EP at baseline
EDSST7 as predicted by Model 1: 1.8557 + 0.0482  s-P100T1 +
0.0452  (s-CMCTUE,T1 + s-CMCTLE,T1) + 0.0256  (s-N13–N20T1 + s-
N22–P40T1) correlated with the observed values (rhos = 0.70,
p < 0.0001) (Table 3). Group cross-validation resulted in R2 = 0.54
for Model 1 (including SSEP) and R2 = 0.47 for Model 2 (VEP and
MEP only), which was thus significantly inferior to Model 1
(p = 0.017). Consistently, the Akaike criterion yielded a slightly
lower value for Model 1 with AIC (Model 1) = 140.5 versus AIC
(Model 2) = 142.1.
Correlation coefficienta p-Value
0.70 <0.001
0.71 <0.001
0.51 0.001
0.34 0.035
0.35 0.017
0.35 0.016
0.22 0.144
 R. Schlaeger et al. / Clinical Neurophysiology 123 (2012) 406–410 409

Table 3 comparison. Already in the early phase of RRMS, they allow to pre-
Models for the prediction of EDSS after 3 years as a function of evoked potentials at dict, to some extent, the clinical course over 3 years.
baseline.
Combined z-transformed motor, visual and somatosensory EP
Parameter b-Estimate Standard error p-Value results at baseline correlate with both measures of clinical out-
Model 1: come, the change in EDSS over 3 years and the EDSS after 3 years.
Intercept a 1.8557 0.1902 <0.001 Irrespectively of whether z-transformed values or number of path-
s-P100T1 0.0482 0.0261 0.072 ological results were used, multimodal EP data at entry explained
s-CMCTUE,T1 + s-CMCTLE,T1 0.0452 0.0162 0.008
s-N13–N20T1 + s-N22–P40T1 0.0256 0.0104 0.019
54% of the variance of EDSS at the end of the 3 years observation
period in our patients with early MS. These results confirm findings
Model 2:
Intercept a 1.8710 0.2012 <0.001
of a preceding study with an observation period of 2 years con-
s-P100T1 0.0509 0.0276 0.073 ducted in patients with more advanced disease and with predic-
s-CMCTUE,T1 + s-CMCTLE,T1 0.0672 0.0143 <0.001 tions based on VEPs and MEPs only. Adding SSEPs to the model
Model 3: significantly improves the predictability of disability over 3 years,
Intercept a 1.4866 0.1928 <0.001
while addition of data on electrophysiological change over the first
Number of pathological resultsT1 0.3584 0.0466 <0.001
year of observation does not. Therefore, inclusion of an additional
Abbreviations: T1, baseline; T7, 3 years follow-up; UE, upper extremities; LE, lower EP modality is more important for assessment of prognosis than
extremities; s-latency value, z-score of latency value (right) + z-score of latency
repetitive electrophysiological evaluation within the first year.
value (left).
The sum of z-scores and the number of pathological EP-results
appear to be equivalent predictors of clinical outcome after 3 years.
In contrast, for the purpose of longitudinal comparison between
clinical course and results of multimodal evoked potentials, the
method working with z-values may be more adequate, due to the
fact that the disease may change latencies in a single patient with-
in the range of normal or pathological values without changing the
number of pathological results.
Our results are in concordance with a prospective study by Jung et
al. (2008) in which 37 patients were investigated at a relatively early
stage of RRMS over 2 years. However, in this study EP results were
transformed to an ordinal scale that also comprised qualitative
assessments of potentials, in spite of their high interrater variability
(Comi et al., 1999) and of downgrading numerical information to an
ordinal level (Emerson, 1998). The latter procedure might be the rea-
son for the absence of predictive power of EP in the study of O’Connor
rhoSpearman =0.70 et al. (1998). The present results are also in line with two retrospec-
p=<0.0001 tive studies. In one of them (Kallmann et al., 2006) EP data generated
within 2 years after diagnosis (not first manifestation) was shown to
be predictive of the clinical course over 5 years, in the other study
(Leocani et al., 2006), a group of patients with relapsing-remitting,
secondary or primary progressive course were investigated at a
more advanced stage of the disease. While retrospective design
Fig. 1. Observed and predicted EDSS at year 3 using the linear Model 1. may include uncertainty whether some of the data might have been
obtained during or in temporal proximity to relapses and conse-
EDSST7 as predicted by (Model 3): 1.487 + 0.358  (number of
quently renders the interpretation more difficult, these studies from
pathological resultsT1) correlated with the observed EDSST7 values
different laboratories and with different patients further support the
(rhos = 0.71, p < 0.0001). Group cross-validation resulted in R2 =
validity of EPs as prognostic markers in patients with MS.
0.54.
A substantial part of our patients received immunomodulatory or
The factor therapy did not reach statistical significance neither
immunosuppressive treatment during the study period, which
as a single predictive factor nor in combination with EP. The factor
potentially influences the predictive relation between EPs and dis-
therapy was not included in the final model since it would have in-
ability. This potentially confounding factor is hard to avoid. We tried
creased AIC. Moreover, duration of therapy and clinical outcome
to account for this by including ‘‘therapy’’ as a binary factor into the
were not correlated. Adding data of electrophysiological change
predictive model. In contrast to EPs, treatment by far failed to reach
over the first year did not improve prediction of outcome after
statistical significance. The lack of a significant correlation between
3 years.
therapy duration and clinical outcome is in accordance with the con-
Fig. 1 shows the relationship between observed and predicted
clusion that the factor ‘‘therapy’’ did not influence the present results
EDSST7 using the linear Model 1. As the EDSS is bounded by 10, a
substantially. Patients with treatment, however, tended to have
non-linear model was also calculated. The correlation between
experienced more relapses prior to entry, which simply reflects cur-
the prediction given by the linear model (Model 1) and the non-lin-
rent clinical criteria for patient selection regarding therapeutic
ear model (Model 1a) was rhos = 0.99 (Spearman correlation coef-
interventions. They had slightly higher z-transformed EP values at
ficient). Group cross-validation of the non-linear model resulted in
baseline, followed by a slightly lesser increase during follow-up –
R2 = 0.52. This close agreement between the two models justifies
however these differences were not statistically significant. This
the use of the simpler linear model.
study was not designed to assess the effect of therapy.
In conclusion, multimodal EPs seem well suited as markers of
4. Discussion the disease course in clinical trials in MS as they indicate function-
ally relevant changes already in the early phase of MS, yield
Multimodal, numerically scaled EP-measures correlate well numerical data and produce robust results already in small co-
with clinical disability both in cross-sectional and longitudinal horts. If our results are confirmed by further prospective studies
410 R. Schlaeger et al. / Clinical Neurophysiology 123 (2012) 406–410
with independent and larger sample groups in different laborato-
ries, this will have important implications for the use of EP in indi-
vidual patient counselling regarding therapeutical decisions. By
performing a normalization using z-transformed values instead of
measured latencies, a transfer of this methodological approach to
other laboratories is possible. This is equally true for predictions
using the number pathological EP results.
Integrating information from multimodal EP studies with addi-
tional data, e.g. from imaging, genetics, immunology or demogra-
phy might further improve patient assessment and prognosis,
and ultimately facilitate individually tailored treatment in MS.
Acknowledgements
The authors thank Sylvia Fluri, Monika Kungler, Susanne Tanner
and Beatrice Wessner for technical assistance. The study was spon-
sored by the Swiss Multiple Sclerosis Society and Bayer Schering.
Results of this study were presented in part at the AAN 2010, Tor-
onto, Canada.
Appendix A
In our analysis, we did not distinguish the side of the body on
which measurements were taken and therefore always summed
z-scores of right and left hand measurements.
The variables considered were:
s-CMCTUE,T1 = z-CMCTUE,T1 (left hand side) + z-CMCTUE,T1 (right
hand side),
s-CMCTLE,T1 = z-CMCTLE,T1 (left hand side) + z-CMCTUE,T1 (right
hand side),
s-P100T1 = z-P100T1 (left hand side) + z-P100T1 (right hand side),
s-N13–N20T1 = z-N13–N20T1 (left hand side) + z-N13–N20T1
(right hand side),
s-N22–P40T1 = z-N22–P40T1 (left hand side) + z-N22–P40T1
(right hand side),
The first regression model considered was
EðEDSST7 Þ ¼ a þ b1 x1 þ b2 x2 þ b3 x3 þ b4 x4 ;
where a = intercept term, bi = slope between EDSST7 and the ith pre-
dictor variable and x1 = s-P100T1, x2 = s-CMCTUE,T1 + s-CMCTLE,T1,
x3 = s-N13–N20T1 + s-N22–P40T1, x4 = therapy.
As the variable therapy did not show a significant effect, it was
omitted from all further analyses and the following model was ta-
ken as a basis:
EðEDSST7 Þ ¼ a1 þ b1 x1 þ b2 x2 þ b3 x3 ðModel 1Þ
To assess whether prediction of EDSST7 is improved by including
SSEP information (as in Model 1), a second model was calculated
in analogy to a previously published model (Fuhr et al., 2001):
EðEDSST7 Þ ¼ a2 þ b4 x1 þ b5 x2 ðModel 2Þ
We also computed a non-linear regression model with
EðEDSST7 Þ ¼ 10 expða3 þ b6 x1 þ b7 x2 þ b8 x3 Þ=
½1 þ expða3 þ b6 x1 þ b7 x2 þ b8 x3 Þ ðModel 1aÞ
This model has the advantage that its values cannot exceed the
maximum of 10 of the EDSS-scale.
As an alternative to the model with the sum of z-scores of
evoked potentials we also considered a model with the number
of pathological EP results:
EðEDSST7 Þ ¼ a4 þ b9 x5 ðModel 3Þ
where x5 = number of pathological results at T1.
The Models (1), (1a), (2) and (3) were assessed using group
cross-validation.
In each of 1500 simulation loops, three randomly selected sub-
jects were removed from the data set and prediction equations
were estimated from the data of the remaining subjects. Then
the EDSST7 values of the three omitted subjects were predicted
and the squared prediction errors recorded. A model with a lower
mean squared cross-validation error is considered as superior.
In addition, we also compared the models based on the cor-
rected Akaike information criterion.
References
Barkhof F. The clinico-radiological paradox in multiple sclerosis revisited. Curr Opin
Neurol 2002;15:239–45.
Barkhof F, Filippi M. MRI – the perfect surrogate marker for multiple sclerosis? Nat
Rev Neurol 2009;5:182–3.
Comi G, Leocani L, Medaglini S, Locatelli T, Martinelli V, Santuccio G, et al.
Measuring evoked responses in multiple sclerosis. Mult Scler 1999;5:263–7.
Confavreux C, Vukusic S, Moreau T, Adeleine P. Relapses and progression of
disability in multiple sclerosis. N Engl J Med 2000;343:1430–8.
Confavreux C, Vukusic S, Adeleine P. Early clinical predictors and progression of
irreversible disability in multiple sclerosis: an amnesic process. Brain
2003;126:770–82.
Daumer M, Neuhaus A, Morrissey S, Hintzen R, Ebers GC. MRI as an outcome in
multiple sclerosis clinical trials. Neurology 2009;72:705–11.
Emerson RG. Evoked potentials in clinical trials for multiple sclerosis. J Clin
Neurophysiol 1998;15:109–16.
Fuhr P, Borggrefe-Chappuis A, Schindler C, Kappos L. Visual and motor evoked
potentials in the course of multiple sclerosis. Brain 2001;124:2162–8.
Jung P, Beyerle A, Ziemann U. Multimodal evoked potentials measure and predict
disability progression in early relapsing-remitting multiple sclerosis. Mult Scler
2008;14:553–6.
Kallmann BA, Fackelmann S, Toyka KV, Rieckmann P, Reiners K. Early abnormalities
of evoked potentials and future disability in patients with multiple sclerosis.
Mult Scler 2006;12:58–65.
Kappos L, Lechner-Scott J, Wu S; 2007. www.neurostatus.net.
Kurtzke JF. Rating neurologic impairment in multiple sclerosis: an expanded
disability status scale (EDSS). Neurology 1983;33:1444–52.
Lechner-Scott J, Kappos L, Hofman M, Polman CH, Ronner H, Montalban X, et al. Can
the expanded disability status scale be assessed by telephone? Mult Scler
2003;9:154–9.
Leocani L, Rovaris M, Boneschi FM, Medaglini S, Rossi P, Martinelli V, et al.
Multimodal evoked potentials to assess the evolution of multiple sclerosis: a
longitudinal study. J Neurol Neurosurg Psychiatry 2006;77(9):1030–5.
McDonald WI, Compston A, Edan G, Goodkin D, Hartung HP, Lublin FD, et al.
Recommended diagnostic criteria for multiple sclerosis: guidelines from the
international panel on the diagnosis of multiple sclerosis. Ann Neurol
2001;50:121–7.
O’Connor P, Marchetti P, Lee L, Perera M. Evoked potential abnormality scores are a
useful measure of disease burden in relapsing-remitting multiple sclerosis. Ann
Neurol 1998;44:404–7.
Polman CH, Reingold SC, Edan G, Filippi M, Hartung HP, Kappos L, et al. Diagnostic
criteria for multiple sclerosis: 2005 revisions to the ‘‘McDonald Criteria’’. Ann
Neurol 2005;58:840–6.
Poser CM, Paty DW, Scheinberg L, McDonald WI, Davis FA, Ebers GC, et al. New
diagnostic criteria for multiple sclerosis: guidelines for research proposals. Ann
Neurol 1983;13:227–31.
Runmarker B, Andersson C, Odén A, Andersen O. Prediction of outcome in multiple
sclerosis based on multivariate models. J Neurol 1994;241:597–604.
Stoehr M. Somatosensible Reizantworten von Nerven, Rückenmark und Gehirn. In:
Stoehr M, Dichgans J, Buettner UW, Hess CW, editors. Evozierte Potentiale. 4th
ed. Heidelberg: Springer Medizin Verlag; 2005, p.63.
Young PJ, Lederer C, Eder K, Daumer M, Neiss A, Polman C, et al. Relapses and
subsequent worsening of disability in relapsing-remitting multiple sclerosis.
Neurology 2006;67:804–8.