Using Gadolinium-enhanced Magnetic Resonance Imaging Lesions to Monitor Disease Activity in Multiple Sclerosis Henry F. McFarland, MD,"Joseph A. Frank, MD,+ Paul S. Albert, PhD,t Mary E. Smith, MD," Roland Martin, MD,* Jonathan 0. Harris, MD," Nicholas Patronas, M D , S Heidi Maloni, RN," and Dale E. McFarlin, MD"
The highly variable clinical course and the lack of a direct measurement of disease activity have made evaluation of experimental therapies in multiple sclerosis (MS) difficult. Recent studies indicate that clinically silent lesions can be demonstrated by magnetic resonance imaging (MRI) in patients with mild relapsing-remitting MS. Thus, MRI may provide a means for monitoring therapeutic trials in the early phase of MS. We studied 12 patients longitudinally for 12 to 21 months with monthly gadolinium (Gd)-enhanced MRIs. The data have been used to identify the most effective design of a clinical trial using Gd-enhanced lesions as the outcome measure. Frequent (>l/mo) Gd-enhancing lesions were observed in 9 of the 12 patients, indicating that the disease is active even during the early phase of the illness. The frequency of the lesions was not constant; there was marked fluctuation in lesion number from month to month. However, the magnitude of the peak number of lesions and the frequency of the peaks varied among patients. Because of this variability, the most effective use of Gd-enhancing lesions as an outcome measure in a clinical trial was a crossover design with study arms of sufficient duration to allow accurate estimation of lesion frequency. Monitoring Gd-enhancing lesions may be an effective tool to assist in the assessment of experimental therapies in early MS. McFarland HF, Frank JA, Albert PS, Smith ME, Martin R, Harris JO, Patronas N, Maloni H, McFarlin DE. Using gadolinium-enhanced magnetic resonance imaging lesions to monitor disease activity in multiple sclerosis. Ann Neurol 1992;32:758-766
The clinical variability in multiple sclerosis (MS) has made design and assessment of clinical trials difficult {If. Although there is increasing interest in treating patients before significant disability occurs, the inability to predict the future course of the disease has made assessment of the risk-benefit relationship difficult. Some patients will continue to have mild disease throughout the course of the illness. Treatment of these patients early in the course of the disease with potentially toxic treatments may pose greater risk than the disease. Further complicating clinical trials has been the difficulty in measuring disease activity, which is usually assessed indirectly by measuring disability. Magnetic resonance imaging (MRI) is well established as the optimal imaging technique for the diagnosis of MS E2-81, and the presence of disease activity in the absence of clinical changes has been confirmed by recent MRI studies C9, 101. Areas of increased signal on T2-weighted images, which reflect demyelination, inflammation, or edema {ll],have been shown to oc-
cur in the cerebrum in clinically stable patients C9, 101. In addition, use of gadolinium (Gd) with T1-weighted imaging can identify areas of breakdown in the bloodbrain barrier. These areas of enhancement seem to represent the initial stage of lesion development [12-151 and are probably associated with active inflammation {14, 16}. Recent studies, including an initial report on 6 of the patients from our present study, indicate that frequent, new Gd-enhancing lesions occur in patients without corresponding clinical changes I17- 191. These findings indicate that MRI parameters and, in particular, Gd-enhancing lesions, should be helpful in monitoring disease activity in patients with MS and may provide a suitable tool for assessing the effectiveness of clinical trials, particularly in patients with early, mild, relapsing-remitting MS [I, 201. Our previous investigation of patients with MS using Gd-enhanced MRI suggested that the occurrence of enhancing lesions was not constant over time. Because the pattern of lesion occurrence affects the usefulness
From the *Neuroimmunology Branch and the tBiometry Branch, National Institute of Neurological Disease and Stroke, and the $Diagnostic Radiology Department, Clinical Center, National Institutes of Health, Bethesda, MD; and the Department of Radiology, Georgetown University Medical Center, Washington, DC.
Received Jan 8, 1992, and in revised form Apr 20 and Jun 24. Accepred for publication Jun 28, 1992.
758
Address correspondence to Dr McF&,,,d, Health, Building 10, R~~~ 5 ~ 1 6~, ~
National Institutes of ~MDh 20892, ~ ~ d ~
,
of MRI in monitoring clinical trials, the initial study has been expanded into an extended longitudinal examination of 10 patients with relapsing-remitting MS and 2 patients with chronic-progressive MS studied for 12 to 21 months. The data derived from this study have been used to evaluate the optimal design of a clinical trial using Gd-enhanced lesions as the primary outcome measure. Materials and Methods Patient Selection This study was reviewed and approved by the Institute Clinical Research Subpanel and informed consent was obtained from each patient prior to beginning the study. Twelve patients with clinically definite MS were selected for this study. Disability was classified according to the Expanded Disability Status Scale (EDSS) [21). Ten patients had relapsingremitting MS and 9 of these had mild disability (EDSS 5 3.5). The tenth patient with relapsing-remitting disease had an EDSS score of 6.5 and required bilateral assistance to walk due to corticospinal tract involvement, but her other clinical findings were mild. Two patients had chronicprogressive disease; 1 started with a relapsing-remitting course, whereas the other (Patient 12) had a chronicprogressive course since the onset of his illness.
Patient Evaluation Patients were imaged and examined monthly. Examination consisted of a standardized neurological examination. Patients were rated using the functional systems scale and EDSS. Exacerbations were defined using a modification of the Schumacker criteria E22) and consisted of any new symptom or worsening of previous symptoms associated with significant changes in signs lasting longer than 24 hours.
MRI Scans Imaging was performed monthly for 12 to 21 months in the 12 patients. Reproducible head positioning from month to month was assured by placing one vitamin E capsule in the external ear canal and taping a second to the lateral canthus of the eye. The canthal-meatal line between these capsules delineated the plane of each scan. A scout slice was performed at the start of each study to establish consistent head positioning. MRI scans were performed on a General Electric Signa 1.5T unit (General Electric, Milwaukee, WI) using a T2-weighted spin echo pulse sequence with an echo time (TE) of 80 msec, a repetition time (TR) of 2,000 msec, 2 excitations, a 128 x 256 acquisition matrix, and 5 mm contiguous interleaved slices. T1-weighted spin-echo (TR 600/ TE 20) images of the brain were performed before and approximately 5 minutes after Gd (Magnevist, 0.1 mmollkg; Berlex Laboratories, Cedar Knolls, NJ) was administered, using 2 excitations, a 192 x 256 acquisition matrix, and 5 mm contiguous slices. All studies were performed with a field of view of 24 cm.
Image Evaluation MRIs were evaluated by at least 2 neurologists, and questionable lesions were reviewed with a neuroradiologist. T 1-
weighted, Gd-enhanced images were visually compared to images taken prior to administration of contrast. Most enhancing lesions were easily recognized as areas of increased signal intensity in the white matter. Questionable small areas of enhancement near the cortical surface were excluded from the analysis. Lesions were numbered sequentially, and new enhancing lesions were defined as lesions that had not enhanced the previous month. Reenhancement of lesions was noted in patients studied for more than 12 months. Because reenhancing lesions may reflect areas of renewed breakdown of the blood-brain barrier, they have been considered new lesions.
Statistical Anahsis Inspection of the frequency of lesions indicated that the number of lesions varied from month to month in the same patient. Because there was a suggestion of a cyclical trend in lesion frequency, new and total enhanced lesion data were examined by fitting the data to Poisson regression models [23] with sinusoidal trends. The best fitting sinusoidal curve with a frequency between 2 and 14 months was obtained and compared using a chi-square test to the data fitted to a model employing a constant mean. Lesions occurring randomly would have a best fit with the model employing a constant mean. Such a comparison is particularly sensitive for detecting a pattern of fluctuations between high and low frequency. This analysis was done using Poisson regression in conjunction with an analytic technique {23] originally proposed for another application {24}. A p value of 0.01 was chosen as the cutoff for statistical significance to account for multiple comparisons inherent in fitting these models separately to each of the 10 patients. The data from the 10 patients with relapsing-remitting MS were used to calculate sample sizes for various trial designs. The sample sizes required to detect a 50% reduction in lesion frequency due to treatment (alpha = 0.05, power = 0.8; two-tailed test) were computed for various parallel and crossover study designs using variance estimates obtained with repeated sampling techniques (the bootstrap analysis) [25, 26). Sample sizes were computed using standard size calculations {27].
Results Characteristics of Gd-enhancing Lesions The clinical characteristics of the patients are shown in Table 1. These patients were studied with monthly MRIs, including Gd-enhanced images, for periods ranging from 12 to 21 months. A total of 197 MRIs were analyzed in this study. With the exception of Patient 12, numerous Gdenhancing lesions were observed in each of the patients (see Table 1). The total number of enhancing lesions was greater than the number of new enhancing lesions because some lesions persisted and were observed on more than one sequential examination (data not shown). In the majority (68%) of new lesions, enhancement was not detected at the next monthly examination, indicating that the duration of the enhancement was less than 2 months. This observation is McFarland et al: MRI Lesions in MS 759
Table 1 , Clinical Parameters of Patients Studied by Serial Gadolinium-enhanced MRI
Patient
Age (yr)
1 2 3 4
44 27 41 28
5 6 7
28 41 38 31 38 28 43
8
7 10 11 12
WR
60
Years from Diagnosis
2 2 5 3 2.5 3.5 16 10 2.5
6 7 17
Average No. Gd-enhancing Lesions
EDSS Course
Months Studied
Start
End
Exacerbations
Total
New
R/R WR R/R R/R R/R R/R R/R-CP R/R WR WR WR CP
17 17 20 19 21 20 14 14 13 12 14 12
1.5 2.0 1.5 1.5 1.5 3.5 5.5 1.5 2.0 6.5 2.0 6.0
1.5 2.0 1.5 2.0 1.5 2.5 6.0 1.5 2.5 6.5 2.0 6.5
0 1 1 4 1 1 0 2 3
3.4 0.5 1.6 5.0 3.3 5.6 1.8 5.6 0.7 8.0 3.2 0.1
2.2 0.2 1.1 4.1 2.4 3.3 1.1 3.5 0.7 5.7 2.1 0.1
1 1 0
= relapsing-remitting; CP = chronic-progressive.
consistent with previous reports { 15, 181. Twentyeight percent of the lesions were seen on 2 concurrent examinations, whereas 5% persisted for 3 months. None of the lesions persisted for more than 4 months. In general, lesions persisting for 2 months or longer were observed in patients with a large number of new enhancing lesions (Patients 5, 6, 8, and 10). Reenhancement of lesions that were enhanced previously was observed in 5 patients. It is likely that reenhancement will be observed more frequently as patients are followed longer. The mean number of Gd-enhancing lesions per month varied among the 12 patients. Although 3 of the patients with relapsing-remitting disease (Patients 2 , 3, and 9) and both of the patients with chronicprogressive disease (Patients 7 and 12) had an average of 2 or fewer new lesions per month, 4 of the patients with relapsing-remitting disease (Patients 4, 6, 8, and 10) had an average of 3 or more new lesions per month.
Pattern of Lesion Occurrence As indicated, inspection of the frequencies of new enhancing lesions indicated that the lesions were not occurring at a constant rate but tended to occur in bursts of increased lesion frequency. Figure 1 illustrates the fluctuating number of Gd-enhancing lesions in Patient 4. By inspection, the pattern of lesion frequency in this patient had a cyclical trend, A suggestion of a similar cyclical pattern in lesion frequency was also observed in the other patients. Consequently, the likelihood that the pattern of lesion occurrence was not random was assessed using Poisson regression models as described in Materials and Methods. In 6 of the 10 patients with relapsing-remitting MS, the total and new lesion data
fit a model employing a sinusoidal mean significantly
better than a model using a constant mean (p < 0.01) (Fig 2). This finding indicates that the mean of the lesion frequency was not constant and that the lesions were not occurring with random frequency. Although not reaching statistical significance, a sinusoidal trend in lesion frequency was also observed in 3 of the remaining 4 patients @ < 0.06, 0.12, 0.12). The Poisson regression model with sinusoidal trends was used only as a means to describe the fluctuating nature of lesion frequency over short durations and to test for nonconstancy in lesion frequency. The model was not intended as a definitive description of lesion occurrence. Using the fitted curves, the period between the peaks in lesion frequency varied considerably and ranged from 2.1 to 12 months. The mean period between peaks in the patients showing a significant fit to a sinusoidal curve was 6.2 and 5.9 months for all 10 patients.
Cowelations Between Clinical Changes and Gd-enhancing Lesions As reported previously, the frequency of enhancing lesions seen on MRI did not necessarily parallel clinical changes. Most notable was Patient 12, who despite continued clinical progression had only I enhancing lesion during 10 months of study. Because the clinical course demonstrated by Patient 12 may represent a unique form of MS (i.e., primary chronic-progressive MS) 118, 28, 297, the findings from this patient were excluded from the subsequent analysis. A dissociation between clinical change and enhancing lesions in the cerebrum was observed in the 11 remaining patients. All had new enhancing lesions, often numerous, that occurred without new symptoms or abnormalities on neurological examination.
760 Annals of Neurology Vol 32 No 6 December 1772
Patient 4 A Total
Lesions
New Lesions
r' T
2
~
4
~
6
~
8
1
10
~
12
1
14
~
16
18
1
~
20
Months
B
C
Fig 1. (A) Fluctuations in number of total and new Gdenhancing lesions in Patient 4. (B) TI -weighted postgadolinium MRI of Patient 4 at month 14. Four representative slices
are shown demonstrating multtiple enhancing lesions. (C) TI weighted postgadolinium MRI at month 15. Four representative slices showing reduction in number of enhancing lesions.
McFarland et al: MRI Lesions in MS 761
1
Patient 4
Patient 3 12 I
-
10
8
p d 0.01
#
I-
N
I-
L
a-
8 W
I-
8
R A n
-
10
W
L
Fnpwncy 4.1 mo.
2.1 mo.
#
N
-
12
I
Froguoncy p 4 0.01
I
4-
0
4
n a
8
2-
2 0
o
2
4
a
a
m
12 Month
14
IO
m
0'
o
2 0 2 2
d 2
'
'
'
I
4
'
I
'
m 1 2 Month
1
'
4
'
'
'
m
u
a
o
2
2
Patient 8
Patient 6
4
Fnquoncy = n.6 mo. #
mo.
p
I1
10 L 8
Froquoncy = 9.2 0.01
R
0
I 0
8
*
4
n
2
o
2
4
o
a
0
m 1 2 1 4 v a i e . 2 0 2 2 Month
o
2
a
I
4
m i z Month
u
m
m
2
0
2
2
Patient 11
Patient 10 14
#
U W
10
Fnquoncy = 12.0 p 0.01
R
12
mo. p 0.01
-
I-
L 8
a-
8
4-
0 n
I
I
0
n
4
8
8
2
2-
0'
o
'
z
' 4
'
I
"
I
m
' iz
' 14
"
m
0 0
"
m z o 2 z
Month
2
4
1
1
m
i z Month
~
m
~
z
o
2
A Fig 2. The observed number of new lesions and the frecpency of lesions derived fmm the best fitting Poisson regression model with stnusoiahl trends. (A) p < 0.01. (B) p > 0.01.
For example, Patient 1, who had a mean of 2.2 new Gd-enhancing lesions per month, was clinically stable throughout the 19-month study period. Each of the remaining 10 patients had one or more exacerbations (see Table 1). The clinical findings associated with these exacerbations could not be explained by the location of the new enhancing lesions in the cerebrum and, in most instances, were consistent with involvement of the spinal cord or the brainstem.
762 Annals of Neurology Vol 32 NO 6 December 1992
2
Patient 1
Patient 2 Froquancy
10 -
-
-
4
8.3 mo.
x
Frequency #
4.1 mo.
3-
N 8 W
L
9 -
8 8
I
0
0
2
4
8
8
l 0 1 2 1 4 1 O l 8 2 0 2 2 Month
Month
Patient 5
Patient 9 4 Frequency. 6.8 mo.
Froqwncy
-
4.0 mo.
#
N a
W
L
R
a
L
2
a a I
8
i
0
0
n
n
8
8
2
4
8
8
I0 12 Month
I
,
14
I8
,
1
,
I 20
22
0
0
2
4
0
B
8
10 12 Month
14
M
2 0 2 2
Fig 2 (continued) Trial Design Using Gd-enhancing Lesions as an Outcome Measure Although it is reasonable to consider using Gd-enhancing lesions as an outcome measurement in therapeutic
trials, the fluctuations in lesion frequency noted herein indicate that there are potential difficulties with this approach. To explore the usefulness of the approach, we assessed the longitudinal MRI data obtained in this study using a statistical technique termed bootstrap analysis, which involves repeated random resampling of the data. Using this technique, we calculated the required sample sizes for 2 trial designs: a parallel groups design and a crossover design. First, the sample sizes required to detect a 50% reduction in lesion frequency in a therapeutic trial using a parallel groups design was calculated (Table 2). The number of new lesions, in contrast to the number of total lesions, was chosen as the most appropriate measure because an effective treatment would be more likely to stop new lesion development than to shorten the duration of lesions. The sample size estimates for a parallel groups design were very high and decreased only slightly when the number of MRIs for each individual was increased. The large sam-
Table 2. Estimated Sample Sizar Requiredfor ra Clinical Trial in M S Using a ParalLel Groups Design and New Gd-enhancing Lesions as the Outcome Measure No. Monthly MRIs per Subject
Sample Size"
No. MRIs
165 114 102 94 90 90
330 342 408 470 540 630
"Sample size required for each of the two groups (a treatment and a placebo group) to detect a 50% reduction in lesionslmonth with a power of 0.8 and an alpha = 0.05 (two-tail test).
ple sizes were due largely to the variability among patients. Because of the large sample sizes required for a parallel groups design, a crossover study design was next considered. The bootstrap analysis was used to compute the sample sizes required to detect a 50% reduction in the number of enhancing lesions due to treatMcFarland e t al: MRI Lesions in MS
763
F i g 3. Crossover designs using Gd-enhanced MRI lesions as the primay outcome variable.
out period was included in the design. Because an additional MRI would be needed after the washout period to identify new lesions, the number of MRIs was greater than in the open crossover design. The third crossover design (see Fig 3C) dlows for the selection of patients with a particular esion frequency before randomization. This trial design incorporates elements of both parallel groups and crossover designs because the outcome analysis would compare the change in lesion frequency in the treatment group with that of the placebo group. Consequently, the sample size would be greater than twice that for the open crossover design (see Table 3) but less than that required for the parallel groups design. Because the fluctuations in lesion frequency were usually 3 to 7 months in duration, an initial assessment period of less than 4 months was unlikely to provide an accurate estimate of the lesion frequency. Thus, sample sizes for shorter periods were not calculated.
ment for 3 crossover design variations (Fig 3; Table 3). The first 2 crossover designs are similar because both compare the frequency of lesions in the treatment versus the nontreatment arm. In the open crossover design (see Fig 3A), the sample sizes are based on the assumption that patients with relapsing-remitting disease would be entered into the trial regardless of lesion frequency. This design could be easily modified, however, to select for treatment only patients with a predetermined lesion frequency. Disadvantages of the open crossover design include lack of randomization, difficulty in assessing the MRIs in a blinded manner, and a potential bias in the treatment effect because treatment always follows the nontreatment arm. In the second crossover design (see Fig 3B), patients would be randomized at entry into treatment and placebo groups with subsequent crossover. This approach eliminates the disadvantages noted with the open crossover design. To avoid a carryover effect of the experimental treatment, a 3-month wash-
Discussion Although there has been increasing interest in using MRI as an outcome measure in assessing efficacy of experimental treatments in patients with MS, the appropriate design of trials incorporating MRI measurements of disease activity has only begun to be considered (1, 20). Various MRI parameters, including new lesions as demonstrated on T2-weighted images, changes in the amount (area or volume) of diseased tissue identified by increased T2 signal or lesions that enhance following administration of Gd could be used as an outcome measure. Because breakdown in the blood-brain barrier as reflected by Gd enhancement has been shown to usually represent the initial stage in lesion development (151 and because Gd-enhancing lesions are easily identified, Gd-enhancing lesions represent a reasonable choice for monitoring disease activity. Before any MRI parameters can be used to monitor MS, however, the natural history of that parameter must be established in patients without treatment. The
DESIGNS FOR CROSS-OVER STUDY a)
Finding
Treatment
Treatment
b)
Treatment
Crossover
-\Washout/-
Placebo
Placebo
Randomization
c)
Baseline Finding Select ion
Treatment
I 1
Placebo
Table 3. Estimated Sample Sizes Required for a Clinical Trial Using Crossover Designs and New Gd-enhancing Lmions as the Outcome Measure
No. MRIs per Patient per Study Arm
Design A
Design B
Sample Size"
No. MRIs
Sample Size
No. MRIsb
55 27 22 18 17 12
165 135 154 162 187 156
50 34 28 20
200 204 224 200 204 168
17 12
Design C Sample Size'
No. MRIs
...
... ...
47 45 31
42 3 495 403
aPatients not preselected for lesion frequency. bAn additional MRI is required at the beginning of each arm ro identify new lesions. 'Patients with 2 or more new enhancing lesions/monrh. Sample size and number of MRIs does not include patients screened and excluded.
764 Annals of Neurology Vol 32 No 6 December 1992
present longitudinal study of Gd-enhanced MRI lesions was undertaken to define the natural history of these abnormalities, particularly during the early, relapsing-remitting phase and, based on this longitudinal data, to establish how Gd-enhancing lesions could be best used as an effective outcome measure in clinical treatment trials. The findings described herein confirm previous MRI studies 19, lo}, which demonstrated that MS can be a progressive disease even during the early, relapsingremitting phase and that focal disruptions in the bloodbrain barrier are readily observed during periods of clinical remission {I7- 191. Importantly, the longitudinal study of the patients in this study showed that the frequency of Gd-enhancing lesions is not constant. In all patients, the number of Gd-enhancing lesions fluctuated; in some patients a striking variation in lesion frequency was noted to occur, with a suggestion of regularity. The lesion frequency did not appear to occur randomly because a better fit of the data was demonstrated employing a statistical model using a sinusoidal mean function than with one using a constant mean function. We believe it is unlikely that the fluctuations in lesion frequency in individual patients will continue to occur with a constant periodicity and do not suggest that the Poisson regression model with sinusoidal trends provides a definitive description of lesion occurrence over the duration of disease. With extended follow-up, the pattern of lesion frequency in individual patients may change and bursts of lesions may occur either more or less frequently. However, the cyclical trend in lesion frequency represents an important consideration in the design of clinical trials using Gd-enhancing lesions as an outcome measure. Short periods of study could easily fail to identify the true mean in lesion frequency and could lead to incorrect conclusions. Although additional longitudinal patient data are needed, the present findings indicate that study periods of 6 months or longer are probably needed to obtain a reasonable measure of the frequency of lesions. We examined 2 general approaches for the design of a clinical trial using Gd-enhanced lesions as the outcome measure and calculated the sample sizes and number of MRIs required for the various designs. The first approach examined, the parallel groups design, would require prohibitively large sample sizes and MRI resources. Only a small reduction in the sample size occurred when the number of MRIs per patient was increased, reflecting the large variability among patients. Similar sample sizes were recently reported by other investigators using data obtained from 17 patients with either relapsing-remitting or secondary progressive MS followed for 6 months with monthly MRIs 1207. On the basis of the frequency of Gd-enhancing lesions, 150 patients would be required for a 6-month parallel study.
To minimize the effect of variability among patients, we next examined the use of 3 variations of a crossover design. The first employed an initial series of MRIs to establish the baseline frequency of enhancing lesions in each patient, followed by crossover to a treatment group. Not unexpectedly, the required sample sizes for this crossover design were substantially smaller than for the parallel groups design, and the required sample sizes declined as the number of MRIs increased. The reduction in sample size seen when the number of monthly MRIs increased from 5 to 6 in each arm of the study reflected the frequency of the bursts of lesions, which was, on average, approximately 6 months. Disadvantages of the open crossover design include lack of randomization and difficulty in assessing the MRIs in a blinded fashion. A solution would be to randomize patients at entry and to incorporate a washout period at the time of crossover to avoid a treatment carryover effect. The duration of the washout period could be varied depending on the nature of the treatment. The randomized crossover design would require sample sizes similar to those for the open crossover design. Generally, sample sizes are insensitive to small changes in the washout period. When the carryover effect of a treatment cannot be assessed, the open crossover design would be preferable to design 2. A concern regarding the randomized crossover design is that patients with low frequencies of enhancing lesions ( e g , Patient 2, who had fewer than 1 new lesion per month) would be included; in fact, the sample sizes are calculated on that basis. The relationship between lesion frequency and eventual disability is unknown. However, a recent study demonstrated that patients with benign MS have fewer enhancing lesions than most patients with relapsing-remitting MS 1301. Until the relationship between lesion frequency and eventual disability is better understood, it would seem best to select patients with frequent lesions for therapeutic trials, particularly if the treatment has potential risks. A third possibility (see Fig 3C) represents a blending of crossover and parallel groups design. This design uses an initial evaluation phase to determine lesion frequency and a subsequent randomization to treatment and placebo groups. Unfortunately, a sample size of 31 patients and more than 400 MRIs would be required to conduct this trial. Although this design would seem optimal for treatments other than those with only minimal risk, the required MRI resources are prohibitive unless MRI units dedicated to MS research are available or the design is used in multicenter studies. Consequently, the open crossover design, which allows selection of patients with sufficient lesion frequencies, followed by an open crossover to treatment represents the second choice for testing treatments with potential risks. McFarland et al: MRI Lesions in MS 765
The findings in this and other studies indicate that MRI can provide a means for assessing one aspect of disease activity in patients with early, relapsingremitting MS. The results also indicate that monitoring lesions by MRI can be an effective outcome measure in clinical trials in MS. Evaluation of other MRI parameters, such as the change in the volume of white matter lesions with increased signal intensity on T2-weighted MRI, also may be helpful in monitoring outcome of therapy. The relationship between frequency of MRI lesions as well as other MRI measurements of disease and clinical disability awaits longer follow-up of patients such as those in this study. Consequently, we do not propose that MRI parameters become the sole outcome measures in examining therapeutic effects. We do propose that MRI assessment represents a means for accurately selecting potentially effective treatments that can then be evaluated in more costly phase-I11 clinical trials. This approach will allow the diminishing resources available for clinical studies to be used with the greatest efficiency.
This work was performed in part at the In Vivo Nuclear Magnetic Resonance Research Center of the National Institutes of Health, Bethesda, MD. We thank Ms Susan Inscoe and Ms Jeanette Black for their excellent technical magnetic resonance imaging skills throughout this study. We thank Mrs Irene Naveau for her excellent nursing support during the study. We thank the patients for their cooperation in this study.
References 1. McFarland HF. Clinical trials in multiple sclerosis. In: Porter RJ, Schoenberg BS, eds. Controlled clinical trials in neurological disease. Boston: Kluwer Academic, 1990:321-34 1 2. Young IR, Hall AS, Pallis CA, et al. Nuclear magnetic resonance imaging of the brain in multiple sclerosis. Lancet 1981; 2:1063-1066 3. Lukes SA, Crooks LE, Aminoff MJ, et al. Nuclear magnetic resonance imaging in multiple sclerosis. Ann Neurol 1983; 13392-60 1 4. Gebarski SS, Gabrielsen TO, Gilman S, et al. The initial diagnosis of multiple sclerosis: clinical impact of magnetic resonance imaging. Ann Neurol 1985;17:469-474 5. Jackson JA, Leake DR, Schneiders NJ. Magnetic resonance imaging in multiple sclerosis: results of 32 cases. AJNR 1985; 6:171-176 6. Sheldon JJ, Siddharthan R, Tobias J, et al. MR imaging of multiple sclerosis: comparison with clinical and CT examinations in 74 patients. Am J Roentgen01 1985;145:957-964 7. Ormerod IEC, Miller D H , McDonald WI, et al. The role of NMR imaging in the assessment of multiple sclerosis and isolated neurological lesions: a quantitative study. Brain 1987; 110:1579-1616 8. Paty DW, Oger JJF, Kastrukoff LF, et al. MRI in the diagnosis of MS: a prospective study and comparison of clinical evaluation, evoked potentials, oligoclonal banding and CT. Neurology 1988;38:180-184 9. Isaac C, Li DKB. Genton M, et al. Multiple sclerosis: a serial
766 Annals of Neurology Vol 32
study using MRI in relapsing patients. Neurology 1988;38: 1511-1515 10. Willoughby EW, Grochowski E, Li DKB, et al. Serial magnetic resonance scanning in multiple sclerosis: a second prospective study in relapsing patients. Ann Neurol 1989;25:43-49 11. Stewart WA, Hall LD, Berry K, et al. Magnetic resonance imaging (MRI) in multiple sclerosis (MS); pathological correlation studies in eight cases. Neurology 1986;36(suppl 1):320.13 12. Grossman RI, Gonzalez-Scarano F, Atlas SW, et al. Multiple sclerosis: gadolinium enhancement in MR imaging. Radiology 1986;16 1:72 1-72> 13. Bastianello S, Pozzilli C, Bernardi S, et al. Serial study of gadolinium-DTPA MRI enhancement in multiple sclerosis. Neurology 1990;40:591-595 14. Hawkins CP, Munro PMG, MacKenzie F, et al. Duration and selectivity of blood-brain barrier breakdown in chronic relapsing experimental allergic encephalomyelitis studied by gadoliniumDTPA and protein markers. Brain 1990;113:365-378 15 Kermode AG, Thompson AJ, Tofts P, et al. Breakdown of the blood-brain barrier preceeds symptoms and other MRI signs of new lesions in multiple sclerosis. Brain 1990;113: 1477-1489 16 Katz D, Taubenberger T, Raine C, et al. Gadolinium-enhancing lesions on magnetic resonance imaging: neuropathological findings. Ann Neurol 1990;28:243 (abstract) 17 Grossman RI, Braffrnan BH, Bronson JR, et al. Multiple sclerosis: serial study of gadolinium-enhanced MR imaging. Radiology 1988;169:117-122 18 Miller D H , Rudge P, Johnson G, et al. Serial gadolinium enhanced magnetic resonance imaging in multiple sclerosis. Brain 1988;111:927-939 19. Harris JO, Frank JA, Patronas N , et al. Serial gadoliniumenhanced magnetic resonance imaging scans in patients with early, relapsing multiple sclerosis: implication for clinical trials and natural history. Ann Neurol 1991;29:548-555 20. Miller DH, Barkhof F, Berry I, et al. Magnetic resonance imaging in monitoring the treatment of multiple sclerosis: concerted action guidelines. J Neurol Neurosurg Psychiatry 1991;54:683-688 21. Kurtzke J. Rating neurological impairment in multiple sclerosis: an expanded disability status scale (EDSS). Neurology 1983; 33:1444-1452 22. Schmacher GA, Beebe G, l b l e r RF, et al. Problems of experimental trials of therapy in multiple sclerosis: report by the panal on the evaluation of experimental trials in multiple sclerosis. Ann N Y Acad Sci 1965;112:552-568 23. McCullagh P, Nelder J. Generalized linear models. London: Chapman and Hall, 1989 24. Darby SC, Ellis MJ. A test for synergism between two drugs. Appl Statistics 1976;25:296-299 25. Efron S. The jack knife, the boostrap and other resampling plans. Philadelphia: Society for Industrial and Applied Mathematics, 1982 26. Efron B, Tibshirani R. Statistical data analysis in the computer age. Science 1991;253:390-395 27. Fleiss JL. The design and analysis of clinical experiments. New York John Wiley, 1986. 28. Thompson AJ, Kermode AG, Wicks D, et al. Major differences in the dynamics of primary and secondary progressive multiple sclerosis. Ann Neurol 1991;29:54-62 29. Kappos L, Gold R, Hofmann E, et al. Multiple sclerosis: diagnostic criteria and the role of contrast enhanced MRI. In: Bydder G , Felix R, Bucheler E, eds. Contrast media in MRI. Bussum: Medicom Europe, 1990127-135 30. Thompson AJ, Youl B, Feinstein A, et al. Serial enhanced MRI in early relapsing remitting and benign multiple sclerosis. Neurology 1791;4 l(supp1 1): 169
No 6 December 1992
The highly variable clinical course and the lack of a direct measurement of disease activity have made evaluation of experimental therapies in multiple sclerosis (MS) difficult. Recent studies indicate that clinically silent lesions can be demonstrated by magnetic resonance imaging (MRI) in patients with mild relapsing-remitting MS. Thus, MRI may provide a means for monitoring therapeutic trials in the early phase of MS. We studied 12 patients longitudinally for 12 to 21 months with monthly gadolinium (Gd)-enhanced MRIs. The data have been used to identify the most effective design of a clinical trial using Gd-enhanced lesions as the outcome measure. Frequent (>l/mo) Gd-enhancing lesions were observed in 9 of the 12 patients, indicating that the disease is active even during the early phase of the illness. The frequency of the lesions was not constant; there was marked fluctuation in lesion number from month to month. However, the magnitude of the peak number of lesions and the frequency of the peaks varied among patients. Because of this variability, the most effective use of Gd-enhancing lesions as an outcome measure in a clinical trial was a crossover design with study arms of sufficient duration to allow accurate estimation of lesion frequency. Monitoring Gd-enhancing lesions may be an effective tool to assist in the assessment of experimental therapies in early MS. McFarland HF, Frank JA, Albert PS, Smith ME, Martin R, Harris JO, Patronas N, Maloni H, McFarlin DE. Using gadolinium-enhanced magnetic resonance imaging lesions to monitor disease activity in multiple sclerosis. Ann Neurol 1992;32:758-766
The clinical variability in multiple sclerosis (MS) has made design and assessment of clinical trials difficult {If. Although there is increasing interest in treating patients before significant disability occurs, the inability to predict the future course of the disease has made assessment of the risk-benefit relationship difficult. Some patients will continue to have mild disease throughout the course of the illness. Treatment of these patients early in the course of the disease with potentially toxic treatments may pose greater risk than the disease. Further complicating clinical trials has been the difficulty in measuring disease activity, which is usually assessed indirectly by measuring disability. Magnetic resonance imaging (MRI) is well established as the optimal imaging technique for the diagnosis of MS E2-81, and the presence of disease activity in the absence of clinical changes has been confirmed by recent MRI studies C9, 101. Areas of increased signal on T2-weighted images, which reflect demyelination, inflammation, or edema {ll],have been shown to oc-
cur in the cerebrum in clinically stable patients C9, 101. In addition, use of gadolinium (Gd) with T1-weighted imaging can identify areas of breakdown in the bloodbrain barrier. These areas of enhancement seem to represent the initial stage of lesion development [12-151 and are probably associated with active inflammation {14, 16}. Recent studies, including an initial report on 6 of the patients from our present study, indicate that frequent, new Gd-enhancing lesions occur in patients without corresponding clinical changes I17- 191. These findings indicate that MRI parameters and, in particular, Gd-enhancing lesions, should be helpful in monitoring disease activity in patients with MS and may provide a suitable tool for assessing the effectiveness of clinical trials, particularly in patients with early, mild, relapsing-remitting MS [I, 201. Our previous investigation of patients with MS using Gd-enhanced MRI suggested that the occurrence of enhancing lesions was not constant over time. Because the pattern of lesion occurrence affects the usefulness
From the *Neuroimmunology Branch and the tBiometry Branch, National Institute of Neurological Disease and Stroke, and the $Diagnostic Radiology Department, Clinical Center, National Institutes of Health, Bethesda, MD; and the Department of Radiology, Georgetown University Medical Center, Washington, DC.
Received Jan 8, 1992, and in revised form Apr 20 and Jun 24. Accepred for publication Jun 28, 1992.
758
Address correspondence to Dr McF&,,,d, Health, Building 10, R~~~ 5 ~ 1 6~, ~
National Institutes of ~MDh 20892, ~ ~ d ~
,
of MRI in monitoring clinical trials, the initial study has been expanded into an extended longitudinal examination of 10 patients with relapsing-remitting MS and 2 patients with chronic-progressive MS studied for 12 to 21 months. The data derived from this study have been used to evaluate the optimal design of a clinical trial using Gd-enhanced lesions as the primary outcome measure. Materials and Methods Patient Selection This study was reviewed and approved by the Institute Clinical Research Subpanel and informed consent was obtained from each patient prior to beginning the study. Twelve patients with clinically definite MS were selected for this study. Disability was classified according to the Expanded Disability Status Scale (EDSS) [21). Ten patients had relapsingremitting MS and 9 of these had mild disability (EDSS 5 3.5). The tenth patient with relapsing-remitting disease had an EDSS score of 6.5 and required bilateral assistance to walk due to corticospinal tract involvement, but her other clinical findings were mild. Two patients had chronicprogressive disease; 1 started with a relapsing-remitting course, whereas the other (Patient 12) had a chronicprogressive course since the onset of his illness.
Patient Evaluation Patients were imaged and examined monthly. Examination consisted of a standardized neurological examination. Patients were rated using the functional systems scale and EDSS. Exacerbations were defined using a modification of the Schumacker criteria E22) and consisted of any new symptom or worsening of previous symptoms associated with significant changes in signs lasting longer than 24 hours.
MRI Scans Imaging was performed monthly for 12 to 21 months in the 12 patients. Reproducible head positioning from month to month was assured by placing one vitamin E capsule in the external ear canal and taping a second to the lateral canthus of the eye. The canthal-meatal line between these capsules delineated the plane of each scan. A scout slice was performed at the start of each study to establish consistent head positioning. MRI scans were performed on a General Electric Signa 1.5T unit (General Electric, Milwaukee, WI) using a T2-weighted spin echo pulse sequence with an echo time (TE) of 80 msec, a repetition time (TR) of 2,000 msec, 2 excitations, a 128 x 256 acquisition matrix, and 5 mm contiguous interleaved slices. T1-weighted spin-echo (TR 600/ TE 20) images of the brain were performed before and approximately 5 minutes after Gd (Magnevist, 0.1 mmollkg; Berlex Laboratories, Cedar Knolls, NJ) was administered, using 2 excitations, a 192 x 256 acquisition matrix, and 5 mm contiguous slices. All studies were performed with a field of view of 24 cm.
Image Evaluation MRIs were evaluated by at least 2 neurologists, and questionable lesions were reviewed with a neuroradiologist. T 1-
weighted, Gd-enhanced images were visually compared to images taken prior to administration of contrast. Most enhancing lesions were easily recognized as areas of increased signal intensity in the white matter. Questionable small areas of enhancement near the cortical surface were excluded from the analysis. Lesions were numbered sequentially, and new enhancing lesions were defined as lesions that had not enhanced the previous month. Reenhancement of lesions was noted in patients studied for more than 12 months. Because reenhancing lesions may reflect areas of renewed breakdown of the blood-brain barrier, they have been considered new lesions.
Statistical Anahsis Inspection of the frequency of lesions indicated that the number of lesions varied from month to month in the same patient. Because there was a suggestion of a cyclical trend in lesion frequency, new and total enhanced lesion data were examined by fitting the data to Poisson regression models [23] with sinusoidal trends. The best fitting sinusoidal curve with a frequency between 2 and 14 months was obtained and compared using a chi-square test to the data fitted to a model employing a constant mean. Lesions occurring randomly would have a best fit with the model employing a constant mean. Such a comparison is particularly sensitive for detecting a pattern of fluctuations between high and low frequency. This analysis was done using Poisson regression in conjunction with an analytic technique {23] originally proposed for another application {24}. A p value of 0.01 was chosen as the cutoff for statistical significance to account for multiple comparisons inherent in fitting these models separately to each of the 10 patients. The data from the 10 patients with relapsing-remitting MS were used to calculate sample sizes for various trial designs. The sample sizes required to detect a 50% reduction in lesion frequency due to treatment (alpha = 0.05, power = 0.8; two-tailed test) were computed for various parallel and crossover study designs using variance estimates obtained with repeated sampling techniques (the bootstrap analysis) [25, 26). Sample sizes were computed using standard size calculations {27].
Results Characteristics of Gd-enhancing Lesions The clinical characteristics of the patients are shown in Table 1. These patients were studied with monthly MRIs, including Gd-enhanced images, for periods ranging from 12 to 21 months. A total of 197 MRIs were analyzed in this study. With the exception of Patient 12, numerous Gdenhancing lesions were observed in each of the patients (see Table 1). The total number of enhancing lesions was greater than the number of new enhancing lesions because some lesions persisted and were observed on more than one sequential examination (data not shown). In the majority (68%) of new lesions, enhancement was not detected at the next monthly examination, indicating that the duration of the enhancement was less than 2 months. This observation is McFarland et al: MRI Lesions in MS 759
Table 1 , Clinical Parameters of Patients Studied by Serial Gadolinium-enhanced MRI
Patient
Age (yr)
1 2 3 4
44 27 41 28
5 6 7
28 41 38 31 38 28 43
8
7 10 11 12
WR
60
Years from Diagnosis
2 2 5 3 2.5 3.5 16 10 2.5
6 7 17
Average No. Gd-enhancing Lesions
EDSS Course
Months Studied
Start
End
Exacerbations
Total
New
R/R WR R/R R/R R/R R/R R/R-CP R/R WR WR WR CP
17 17 20 19 21 20 14 14 13 12 14 12
1.5 2.0 1.5 1.5 1.5 3.5 5.5 1.5 2.0 6.5 2.0 6.0
1.5 2.0 1.5 2.0 1.5 2.5 6.0 1.5 2.5 6.5 2.0 6.5
0 1 1 4 1 1 0 2 3
3.4 0.5 1.6 5.0 3.3 5.6 1.8 5.6 0.7 8.0 3.2 0.1
2.2 0.2 1.1 4.1 2.4 3.3 1.1 3.5 0.7 5.7 2.1 0.1
1 1 0
= relapsing-remitting; CP = chronic-progressive.
consistent with previous reports { 15, 181. Twentyeight percent of the lesions were seen on 2 concurrent examinations, whereas 5% persisted for 3 months. None of the lesions persisted for more than 4 months. In general, lesions persisting for 2 months or longer were observed in patients with a large number of new enhancing lesions (Patients 5, 6, 8, and 10). Reenhancement of lesions that were enhanced previously was observed in 5 patients. It is likely that reenhancement will be observed more frequently as patients are followed longer. The mean number of Gd-enhancing lesions per month varied among the 12 patients. Although 3 of the patients with relapsing-remitting disease (Patients 2 , 3, and 9) and both of the patients with chronicprogressive disease (Patients 7 and 12) had an average of 2 or fewer new lesions per month, 4 of the patients with relapsing-remitting disease (Patients 4, 6, 8, and 10) had an average of 3 or more new lesions per month.
Pattern of Lesion Occurrence As indicated, inspection of the frequencies of new enhancing lesions indicated that the lesions were not occurring at a constant rate but tended to occur in bursts of increased lesion frequency. Figure 1 illustrates the fluctuating number of Gd-enhancing lesions in Patient 4. By inspection, the pattern of lesion frequency in this patient had a cyclical trend, A suggestion of a similar cyclical pattern in lesion frequency was also observed in the other patients. Consequently, the likelihood that the pattern of lesion occurrence was not random was assessed using Poisson regression models as described in Materials and Methods. In 6 of the 10 patients with relapsing-remitting MS, the total and new lesion data
fit a model employing a sinusoidal mean significantly
better than a model using a constant mean (p < 0.01) (Fig 2). This finding indicates that the mean of the lesion frequency was not constant and that the lesions were not occurring with random frequency. Although not reaching statistical significance, a sinusoidal trend in lesion frequency was also observed in 3 of the remaining 4 patients @ < 0.06, 0.12, 0.12). The Poisson regression model with sinusoidal trends was used only as a means to describe the fluctuating nature of lesion frequency over short durations and to test for nonconstancy in lesion frequency. The model was not intended as a definitive description of lesion occurrence. Using the fitted curves, the period between the peaks in lesion frequency varied considerably and ranged from 2.1 to 12 months. The mean period between peaks in the patients showing a significant fit to a sinusoidal curve was 6.2 and 5.9 months for all 10 patients.
Cowelations Between Clinical Changes and Gd-enhancing Lesions As reported previously, the frequency of enhancing lesions seen on MRI did not necessarily parallel clinical changes. Most notable was Patient 12, who despite continued clinical progression had only I enhancing lesion during 10 months of study. Because the clinical course demonstrated by Patient 12 may represent a unique form of MS (i.e., primary chronic-progressive MS) 118, 28, 297, the findings from this patient were excluded from the subsequent analysis. A dissociation between clinical change and enhancing lesions in the cerebrum was observed in the 11 remaining patients. All had new enhancing lesions, often numerous, that occurred without new symptoms or abnormalities on neurological examination.
760 Annals of Neurology Vol 32 No 6 December 1772
Patient 4 A Total
Lesions
New Lesions
r' T
2
~
4
~
6
~
8
1
10
~
12
1
14
~
16
18
1
~
20
Months
B
C
Fig 1. (A) Fluctuations in number of total and new Gdenhancing lesions in Patient 4. (B) TI -weighted postgadolinium MRI of Patient 4 at month 14. Four representative slices
are shown demonstrating multtiple enhancing lesions. (C) TI weighted postgadolinium MRI at month 15. Four representative slices showing reduction in number of enhancing lesions.
McFarland et al: MRI Lesions in MS 761
1
Patient 4
Patient 3 12 I
-
10
8
p d 0.01
#
I-
N
I-
L
a-
8 W
I-
8
R A n
-
10
W
L
Fnpwncy 4.1 mo.
2.1 mo.
#
N
-
12
I
Froguoncy p 4 0.01
I
4-
0
4
n a
8
2-
2 0
o
2
4
a
a
m
12 Month
14
IO
m
0'
o
2 0 2 2
d 2
'
'
'
I
4
'
I
'
m 1 2 Month
1
'
4
'
'
'
m
u
a
o
2
2
Patient 8
Patient 6
4
Fnquoncy = n.6 mo. #
mo.
p
I1
10 L 8
Froquoncy = 9.2 0.01
R
0
I 0
8
*
4
n
2
o
2
4
o
a
0
m 1 2 1 4 v a i e . 2 0 2 2 Month
o
2
a
I
4
m i z Month
u
m
m
2
0
2
2
Patient 11
Patient 10 14
#
U W
10
Fnquoncy = 12.0 p 0.01
R
12
mo. p 0.01
-
I-
L 8
a-
8
4-
0 n
I
I
0
n
4
8
8
2
2-
0'
o
'
z
' 4
'
I
"
I
m
' iz
' 14
"
m
0 0
"
m z o 2 z
Month
2
4
1
1
m
i z Month
~
m
~
z
o
2
A Fig 2. The observed number of new lesions and the frecpency of lesions derived fmm the best fitting Poisson regression model with stnusoiahl trends. (A) p < 0.01. (B) p > 0.01.
For example, Patient 1, who had a mean of 2.2 new Gd-enhancing lesions per month, was clinically stable throughout the 19-month study period. Each of the remaining 10 patients had one or more exacerbations (see Table 1). The clinical findings associated with these exacerbations could not be explained by the location of the new enhancing lesions in the cerebrum and, in most instances, were consistent with involvement of the spinal cord or the brainstem.
762 Annals of Neurology Vol 32 NO 6 December 1992
2
Patient 1
Patient 2 Froquancy
10 -
-
-
4
8.3 mo.
x
Frequency #
4.1 mo.
3-
N 8 W
L
9 -
8 8
I
0
0
2
4
8
8
l 0 1 2 1 4 1 O l 8 2 0 2 2 Month
Month
Patient 5
Patient 9 4 Frequency. 6.8 mo.
Froqwncy
-
4.0 mo.
#
N a
W
L
R
a
L
2
a a I
8
i
0
0
n
n
8
8
2
4
8
8
I0 12 Month
I
,
14
I8
,
1
,
I 20
22
0
0
2
4
0
B
8
10 12 Month
14
M
2 0 2 2
Fig 2 (continued) Trial Design Using Gd-enhancing Lesions as an Outcome Measure Although it is reasonable to consider using Gd-enhancing lesions as an outcome measurement in therapeutic
trials, the fluctuations in lesion frequency noted herein indicate that there are potential difficulties with this approach. To explore the usefulness of the approach, we assessed the longitudinal MRI data obtained in this study using a statistical technique termed bootstrap analysis, which involves repeated random resampling of the data. Using this technique, we calculated the required sample sizes for 2 trial designs: a parallel groups design and a crossover design. First, the sample sizes required to detect a 50% reduction in lesion frequency in a therapeutic trial using a parallel groups design was calculated (Table 2). The number of new lesions, in contrast to the number of total lesions, was chosen as the most appropriate measure because an effective treatment would be more likely to stop new lesion development than to shorten the duration of lesions. The sample size estimates for a parallel groups design were very high and decreased only slightly when the number of MRIs for each individual was increased. The large sam-
Table 2. Estimated Sample Sizar Requiredfor ra Clinical Trial in M S Using a ParalLel Groups Design and New Gd-enhancing Lesions as the Outcome Measure No. Monthly MRIs per Subject
Sample Size"
No. MRIs
165 114 102 94 90 90
330 342 408 470 540 630
"Sample size required for each of the two groups (a treatment and a placebo group) to detect a 50% reduction in lesionslmonth with a power of 0.8 and an alpha = 0.05 (two-tail test).
ple sizes were due largely to the variability among patients. Because of the large sample sizes required for a parallel groups design, a crossover study design was next considered. The bootstrap analysis was used to compute the sample sizes required to detect a 50% reduction in the number of enhancing lesions due to treatMcFarland e t al: MRI Lesions in MS
763
F i g 3. Crossover designs using Gd-enhanced MRI lesions as the primay outcome variable.
out period was included in the design. Because an additional MRI would be needed after the washout period to identify new lesions, the number of MRIs was greater than in the open crossover design. The third crossover design (see Fig 3C) dlows for the selection of patients with a particular esion frequency before randomization. This trial design incorporates elements of both parallel groups and crossover designs because the outcome analysis would compare the change in lesion frequency in the treatment group with that of the placebo group. Consequently, the sample size would be greater than twice that for the open crossover design (see Table 3) but less than that required for the parallel groups design. Because the fluctuations in lesion frequency were usually 3 to 7 months in duration, an initial assessment period of less than 4 months was unlikely to provide an accurate estimate of the lesion frequency. Thus, sample sizes for shorter periods were not calculated.
ment for 3 crossover design variations (Fig 3; Table 3). The first 2 crossover designs are similar because both compare the frequency of lesions in the treatment versus the nontreatment arm. In the open crossover design (see Fig 3A), the sample sizes are based on the assumption that patients with relapsing-remitting disease would be entered into the trial regardless of lesion frequency. This design could be easily modified, however, to select for treatment only patients with a predetermined lesion frequency. Disadvantages of the open crossover design include lack of randomization, difficulty in assessing the MRIs in a blinded manner, and a potential bias in the treatment effect because treatment always follows the nontreatment arm. In the second crossover design (see Fig 3B), patients would be randomized at entry into treatment and placebo groups with subsequent crossover. This approach eliminates the disadvantages noted with the open crossover design. To avoid a carryover effect of the experimental treatment, a 3-month wash-
Discussion Although there has been increasing interest in using MRI as an outcome measure in assessing efficacy of experimental treatments in patients with MS, the appropriate design of trials incorporating MRI measurements of disease activity has only begun to be considered (1, 20). Various MRI parameters, including new lesions as demonstrated on T2-weighted images, changes in the amount (area or volume) of diseased tissue identified by increased T2 signal or lesions that enhance following administration of Gd could be used as an outcome measure. Because breakdown in the blood-brain barrier as reflected by Gd enhancement has been shown to usually represent the initial stage in lesion development (151 and because Gd-enhancing lesions are easily identified, Gd-enhancing lesions represent a reasonable choice for monitoring disease activity. Before any MRI parameters can be used to monitor MS, however, the natural history of that parameter must be established in patients without treatment. The
DESIGNS FOR CROSS-OVER STUDY a)
Finding
Treatment
Treatment
b)
Treatment
Crossover
-\Washout/-
Placebo
Placebo
Randomization
c)
Baseline Finding Select ion
Treatment
I 1
Placebo
Table 3. Estimated Sample Sizes Required for a Clinical Trial Using Crossover Designs and New Gd-enhancing Lmions as the Outcome Measure
No. MRIs per Patient per Study Arm
Design A
Design B
Sample Size"
No. MRIs
Sample Size
No. MRIsb
55 27 22 18 17 12
165 135 154 162 187 156
50 34 28 20
200 204 224 200 204 168
17 12
Design C Sample Size'
No. MRIs
...
... ...
47 45 31
42 3 495 403
aPatients not preselected for lesion frequency. bAn additional MRI is required at the beginning of each arm ro identify new lesions. 'Patients with 2 or more new enhancing lesions/monrh. Sample size and number of MRIs does not include patients screened and excluded.
764 Annals of Neurology Vol 32 No 6 December 1992
present longitudinal study of Gd-enhanced MRI lesions was undertaken to define the natural history of these abnormalities, particularly during the early, relapsing-remitting phase and, based on this longitudinal data, to establish how Gd-enhancing lesions could be best used as an effective outcome measure in clinical treatment trials. The findings described herein confirm previous MRI studies 19, lo}, which demonstrated that MS can be a progressive disease even during the early, relapsingremitting phase and that focal disruptions in the bloodbrain barrier are readily observed during periods of clinical remission {I7- 191. Importantly, the longitudinal study of the patients in this study showed that the frequency of Gd-enhancing lesions is not constant. In all patients, the number of Gd-enhancing lesions fluctuated; in some patients a striking variation in lesion frequency was noted to occur, with a suggestion of regularity. The lesion frequency did not appear to occur randomly because a better fit of the data was demonstrated employing a statistical model using a sinusoidal mean function than with one using a constant mean function. We believe it is unlikely that the fluctuations in lesion frequency in individual patients will continue to occur with a constant periodicity and do not suggest that the Poisson regression model with sinusoidal trends provides a definitive description of lesion occurrence over the duration of disease. With extended follow-up, the pattern of lesion frequency in individual patients may change and bursts of lesions may occur either more or less frequently. However, the cyclical trend in lesion frequency represents an important consideration in the design of clinical trials using Gd-enhancing lesions as an outcome measure. Short periods of study could easily fail to identify the true mean in lesion frequency and could lead to incorrect conclusions. Although additional longitudinal patient data are needed, the present findings indicate that study periods of 6 months or longer are probably needed to obtain a reasonable measure of the frequency of lesions. We examined 2 general approaches for the design of a clinical trial using Gd-enhanced lesions as the outcome measure and calculated the sample sizes and number of MRIs required for the various designs. The first approach examined, the parallel groups design, would require prohibitively large sample sizes and MRI resources. Only a small reduction in the sample size occurred when the number of MRIs per patient was increased, reflecting the large variability among patients. Similar sample sizes were recently reported by other investigators using data obtained from 17 patients with either relapsing-remitting or secondary progressive MS followed for 6 months with monthly MRIs 1207. On the basis of the frequency of Gd-enhancing lesions, 150 patients would be required for a 6-month parallel study.
To minimize the effect of variability among patients, we next examined the use of 3 variations of a crossover design. The first employed an initial series of MRIs to establish the baseline frequency of enhancing lesions in each patient, followed by crossover to a treatment group. Not unexpectedly, the required sample sizes for this crossover design were substantially smaller than for the parallel groups design, and the required sample sizes declined as the number of MRIs increased. The reduction in sample size seen when the number of monthly MRIs increased from 5 to 6 in each arm of the study reflected the frequency of the bursts of lesions, which was, on average, approximately 6 months. Disadvantages of the open crossover design include lack of randomization and difficulty in assessing the MRIs in a blinded fashion. A solution would be to randomize patients at entry and to incorporate a washout period at the time of crossover to avoid a treatment carryover effect. The duration of the washout period could be varied depending on the nature of the treatment. The randomized crossover design would require sample sizes similar to those for the open crossover design. Generally, sample sizes are insensitive to small changes in the washout period. When the carryover effect of a treatment cannot be assessed, the open crossover design would be preferable to design 2. A concern regarding the randomized crossover design is that patients with low frequencies of enhancing lesions ( e g , Patient 2, who had fewer than 1 new lesion per month) would be included; in fact, the sample sizes are calculated on that basis. The relationship between lesion frequency and eventual disability is unknown. However, a recent study demonstrated that patients with benign MS have fewer enhancing lesions than most patients with relapsing-remitting MS 1301. Until the relationship between lesion frequency and eventual disability is better understood, it would seem best to select patients with frequent lesions for therapeutic trials, particularly if the treatment has potential risks. A third possibility (see Fig 3C) represents a blending of crossover and parallel groups design. This design uses an initial evaluation phase to determine lesion frequency and a subsequent randomization to treatment and placebo groups. Unfortunately, a sample size of 31 patients and more than 400 MRIs would be required to conduct this trial. Although this design would seem optimal for treatments other than those with only minimal risk, the required MRI resources are prohibitive unless MRI units dedicated to MS research are available or the design is used in multicenter studies. Consequently, the open crossover design, which allows selection of patients with sufficient lesion frequencies, followed by an open crossover to treatment represents the second choice for testing treatments with potential risks. McFarland et al: MRI Lesions in MS 765
The findings in this and other studies indicate that MRI can provide a means for assessing one aspect of disease activity in patients with early, relapsingremitting MS. The results also indicate that monitoring lesions by MRI can be an effective outcome measure in clinical trials in MS. Evaluation of other MRI parameters, such as the change in the volume of white matter lesions with increased signal intensity on T2-weighted MRI, also may be helpful in monitoring outcome of therapy. The relationship between frequency of MRI lesions as well as other MRI measurements of disease and clinical disability awaits longer follow-up of patients such as those in this study. Consequently, we do not propose that MRI parameters become the sole outcome measures in examining therapeutic effects. We do propose that MRI assessment represents a means for accurately selecting potentially effective treatments that can then be evaluated in more costly phase-I11 clinical trials. This approach will allow the diminishing resources available for clinical studies to be used with the greatest efficiency.
This work was performed in part at the In Vivo Nuclear Magnetic Resonance Research Center of the National Institutes of Health, Bethesda, MD. We thank Ms Susan Inscoe and Ms Jeanette Black for their excellent technical magnetic resonance imaging skills throughout this study. We thank Mrs Irene Naveau for her excellent nursing support during the study. We thank the patients for their cooperation in this study.
References 1. McFarland HF. Clinical trials in multiple sclerosis. In: Porter RJ, Schoenberg BS, eds. Controlled clinical trials in neurological disease. Boston: Kluwer Academic, 1990:321-34 1 2. Young IR, Hall AS, Pallis CA, et al. Nuclear magnetic resonance imaging of the brain in multiple sclerosis. Lancet 1981; 2:1063-1066 3. Lukes SA, Crooks LE, Aminoff MJ, et al. Nuclear magnetic resonance imaging in multiple sclerosis. Ann Neurol 1983; 13392-60 1 4. Gebarski SS, Gabrielsen TO, Gilman S, et al. The initial diagnosis of multiple sclerosis: clinical impact of magnetic resonance imaging. Ann Neurol 1985;17:469-474 5. Jackson JA, Leake DR, Schneiders NJ. Magnetic resonance imaging in multiple sclerosis: results of 32 cases. AJNR 1985; 6:171-176 6. Sheldon JJ, Siddharthan R, Tobias J, et al. MR imaging of multiple sclerosis: comparison with clinical and CT examinations in 74 patients. Am J Roentgen01 1985;145:957-964 7. Ormerod IEC, Miller D H , McDonald WI, et al. The role of NMR imaging in the assessment of multiple sclerosis and isolated neurological lesions: a quantitative study. Brain 1987; 110:1579-1616 8. Paty DW, Oger JJF, Kastrukoff LF, et al. MRI in the diagnosis of MS: a prospective study and comparison of clinical evaluation, evoked potentials, oligoclonal banding and CT. Neurology 1988;38:180-184 9. Isaac C, Li DKB. Genton M, et al. Multiple sclerosis: a serial
766 Annals of Neurology Vol 32
study using MRI in relapsing patients. Neurology 1988;38: 1511-1515 10. Willoughby EW, Grochowski E, Li DKB, et al. Serial magnetic resonance scanning in multiple sclerosis: a second prospective study in relapsing patients. Ann Neurol 1989;25:43-49 11. Stewart WA, Hall LD, Berry K, et al. Magnetic resonance imaging (MRI) in multiple sclerosis (MS); pathological correlation studies in eight cases. Neurology 1986;36(suppl 1):320.13 12. Grossman RI, Gonzalez-Scarano F, Atlas SW, et al. Multiple sclerosis: gadolinium enhancement in MR imaging. Radiology 1986;16 1:72 1-72> 13. Bastianello S, Pozzilli C, Bernardi S, et al. Serial study of gadolinium-DTPA MRI enhancement in multiple sclerosis. Neurology 1990;40:591-595 14. Hawkins CP, Munro PMG, MacKenzie F, et al. Duration and selectivity of blood-brain barrier breakdown in chronic relapsing experimental allergic encephalomyelitis studied by gadoliniumDTPA and protein markers. Brain 1990;113:365-378 15 Kermode AG, Thompson AJ, Tofts P, et al. Breakdown of the blood-brain barrier preceeds symptoms and other MRI signs of new lesions in multiple sclerosis. Brain 1990;113: 1477-1489 16 Katz D, Taubenberger T, Raine C, et al. Gadolinium-enhancing lesions on magnetic resonance imaging: neuropathological findings. Ann Neurol 1990;28:243 (abstract) 17 Grossman RI, Braffrnan BH, Bronson JR, et al. Multiple sclerosis: serial study of gadolinium-enhanced MR imaging. Radiology 1988;169:117-122 18 Miller D H , Rudge P, Johnson G, et al. Serial gadolinium enhanced magnetic resonance imaging in multiple sclerosis. Brain 1988;111:927-939 19. Harris JO, Frank JA, Patronas N , et al. Serial gadoliniumenhanced magnetic resonance imaging scans in patients with early, relapsing multiple sclerosis: implication for clinical trials and natural history. Ann Neurol 1991;29:548-555 20. Miller DH, Barkhof F, Berry I, et al. Magnetic resonance imaging in monitoring the treatment of multiple sclerosis: concerted action guidelines. J Neurol Neurosurg Psychiatry 1991;54:683-688 21. Kurtzke J. Rating neurological impairment in multiple sclerosis: an expanded disability status scale (EDSS). Neurology 1983; 33:1444-1452 22. Schmacher GA, Beebe G, l b l e r RF, et al. Problems of experimental trials of therapy in multiple sclerosis: report by the panal on the evaluation of experimental trials in multiple sclerosis. Ann N Y Acad Sci 1965;112:552-568 23. McCullagh P, Nelder J. Generalized linear models. London: Chapman and Hall, 1989 24. Darby SC, Ellis MJ. A test for synergism between two drugs. Appl Statistics 1976;25:296-299 25. Efron S. The jack knife, the boostrap and other resampling plans. Philadelphia: Society for Industrial and Applied Mathematics, 1982 26. Efron B, Tibshirani R. Statistical data analysis in the computer age. Science 1991;253:390-395 27. Fleiss JL. The design and analysis of clinical experiments. New York John Wiley, 1986. 28. Thompson AJ, Kermode AG, Wicks D, et al. Major differences in the dynamics of primary and secondary progressive multiple sclerosis. Ann Neurol 1991;29:54-62 29. Kappos L, Gold R, Hofmann E, et al. Multiple sclerosis: diagnostic criteria and the role of contrast enhanced MRI. In: Bydder G , Felix R, Bucheler E, eds. Contrast media in MRI. Bussum: Medicom Europe, 1990127-135 30. Thompson AJ, Youl B, Feinstein A, et al. Serial enhanced MRI in early relapsing remitting and benign multiple sclerosis. Neurology 1791;4 l(supp1 1): 169
No 6 December 1992
