Eight-year follow-up study of brain atrophy in patients with MS

Clinical trial (1990 through 1994).

Clinical and MRI data for the first 2 years were collected during the MS Collaborative Research Group (MSCRG) phase III trial of IFNβ-1a (AVONEX, Biogen, Cambridge, MA). Three hundred one patients with RRMS were enrolled at four sites (Buffalo, NY; Cleveland, OH; Portland, OR; and Washington, DC). To meet the inclusion criteria, patients had to score between 1.0 and 3.5 on the Kurtzke Expanded Disability Status Score (EDSS)19 and had to have had at least two relapses in the prior 3 years. Details of the trial design have been reported elsewhere.20 Clinical measures of disability included (among others) the EDSS, a timed 25-foot walk, the Nine-Hole Peg test, and the 3-second Paced Serial Addition Test. The MSFC21 was computed as described.17 MRI examinations included a dual-echo, proton density, T2-weighted spin-echo series (repetition time [TR] = 2000 ms, echo time [TE] = 30 and 90 ms) and T1-weighted spin-echo series (TR = 600 ms, TE = 20 ms) acquired before and after injection of contrast. Images were acquired with 192 × 256 matrix size, 240 mm field of view, and 5-mm-thick contiguous slices on 1.5-T scanners. MRI lesion measurements included T2-hyperintense lesion volume, contrast-enhancing lesion number and volume, and T1-hypointense lesion volume.22 The number of new and enlarging T2 lesions at year 2 and the cumulative number of contrast-enhancing lesions over 2 years were also determined. Lesion measurements were performed as part of the original trial at the University of Colorado Health Sciences Center in Denver. Brain parenchymal fraction (BPF) was calculated blindly with automated segmentation software developed at The Cleveland Clinic Foundation, as previously described.7

Follow-up study (1999).

The follow-up study protocol was reviewed and approved by the Institutional Review Board (IRB) of each of the four participating sites. Patients who were enrolled for at least 2 years in the phase III study were eligible for this study. This requirement was included to ensure baseline, year 1, and year 2 data for the MSFC for each subject in the follow-up study. All patients who agreed to participate signed an IRB-approved informed consent form. Follow-up study visits consisted of a neurologic examination (EDSS, relapse history, current disease category, and history of disease-modifying drugs), an examination by a trained technician (MSFC, visual contrast letter acuity), a patient quality-of-life questionnaire (Sickness Impact Profile23), and an MRI examination. Patients who were not able to participate in a clinic visit were asked to complete an EDSS self-report,19 disease-modifying therapy questionnaire, and Sickness Impact Profile questionnaire. Family members of patients who had died in the interim were interviewed to record the cause of death and EDSS prior to death.

Image acquisition and analysis.

As in the original clinical trial, a dual-echo, proton density, T2-weighted spin-echo image series was acquired (TR = 2000 ms; TE = 30 and 90 ms) with 256 × 256 matrix size, 240 mm field of view, and 5-mm-thick contiguous slices on 1.5-T scanners at each of the original sites. The images were sent in digital format to the Cleveland Clinic, transferred to a UNIX file server, and converted to an internal image format (BIP, Whitaker Biomedical Imaging Lab, Cleveland Clinic Foundation). All images were visually inspected and compared to images acquired during the phase III trial to verify consistency.

Brain atrophy was assessed with the same method that was used for the retrospective analysis of the phase III trial images.7 In summary, a program was used to automatically measure the brain parenchymal volume (BV) and the total volume within the outer contour of the brain (TCV) using a knowledge-based, three-dimensional segmentation algorithm.24 The BPF was calculated as the ratio BV/TCV to provide a size-normalized measure of whole-brain atrophy. Previous validation studies showed that the accuracy is approximately 99% for brain volume measurement and the scan–rescan variability is approximately 0.2% for BPF.7 To process dual-echo image sets, the complete analysis pipeline included the following steps in one program: 1) correct intensity nonuniformities; 2) normalize the gray levels; 3) perform optimal linear combination of the early and late echo images to null the signal from cerebral spinal fluid; 4) filter out noise if necessary; 5) perform the segmentation; and 6) calculate volumes accounting for partial volume effects. Following segmentation, the image results were visually verified using custom display software. In cases where the segmentation failed with the default program settings, the process was repeated with adjustments to parameters that control the filtering operations. BPF analysis was fully automated and performed blinded to clinical disability scores and other clinical information. Lesion analysis was not included in the follow-up study because it was considered to be outside the scope of the study (which was primarily to validate the MSFC and, secondarily, to study the predictive value of atrophy measurements for disability).

Analysis subgroups.

The numbers of patients with available MRI measurements were different at each point in time. The numbers varied because some patients did not have MRI scans at the 8-year follow-up, and images from the original trial were omitted from analysis if the electronic data were corrupted or lost in transfer or if the images were not acquired according to protocol—that is, with gaps between the slices or varying pixel sizes. Therefore, two different subgroups of patients were considered to address the specific questions of this study.

For the first question regarding the relation between atrophy and subsequent disability, we used the subgroup of all patients who had at least two analyzable MRI scans during the original trial and an EDSS measurement at the 8-year follow-up. This subgroup was chosen so that the results would not be biased by only including patients who were well enough at follow-up to have an MRI examination. The requirement of at least two MRI scans during the original trial ensured that atrophy rate during the original trial could be estimated. The 2-year atrophy rate was estimated as the percent change in BPF from baseline to year 2 or, for cases with only two consecutive MRI scans available (n = 13), the 2-year atrophy rate was estimated as twice the annual percent change in BPF.

For the second two questions on the progression of atrophy, the subgroup of all patients who had analyzable MRI scans at all four time points was considered. This subgroup was defined to ensure that all patients would have BPF measurements from the original trial and from the 8-year follow-up examination. MRI lesion measurements were also available for the majority of these patients during the original trial.

Statistical analyses.

Correlations were calculated between BPF and disability measurements at each time point using Spearman rank statistics. Patients were categorized into two outcome groups based on whether their EDSS score at 8-year follow-up was less than 6.0 or greater than or equal to 6.0. This particular cutoff value was selected because it is a clinically meaningful and relatively objective milestone: EDSS 6 is the point at which a patient needs assistance to walk. Patients were subclassified into quartiles based on percent BPF change during the 2-year phase III trial. The percent of patients in each quartile who reached EDSS 6 or greater were determined. Mean BPF at each time point was compared between the two EDSS outcome groups using the Mann–Whitney test. Logistic regression analyses were performed to determine which variables during the original trial were significant predictors of disability status at the 8-year follow-up. The following demographic and disease-related variables were considered in all models: age, sex, treatment group, disease duration, number of relapses during the trial, time from baseline to the 8-year follow-up, and total percent time on treatment over the 8-year period. In addition, specific disability and MRI variables were also considered for inclusion in each model. Continuous factors considered for the logistic regression models were plotted using a locally weighted smoothing scatterplot technique to assess linearity on the logit scale. A forward stepwise selection procedure was used with a p value of 0.15 required to enter the model and a p value of 0.05 required to remain in each model.

Spearman correlations were also calculated to determine the relation between MRI lesion measurements from the phase III trial and subsequent atrophy. Multiple regression analysis was performed to determine which parameters from the original trial were significant predictors of the change in BPF from year 2 to the 8-year follow-up. The same demographic and disease-related variables as those used for the logistic regression analyses were considered for inclusion in all models, in addition to model-specific disability and MRI variables. Linearity between each independent variable and the dependent variable was confirmed. A forward stepwise selection procedure was used with p values of 0.15 to enter and 0.05 to remain in the model.

Descriptive statistics.

Follow-up clinical data were obtained on 160 (93%) of the 172 patients who were eligible for the study, including 3 patients who returned for a neurologic examination but refused MRI, 16 patients who completed self-reports for EDSS but did not have neurologic or MRI examinations, and 7 patients who died. EDSS scores prior to the terminal event for these seven patients were assigned by the neurologist after interview with the family and record review. The mean follow-up time was 8.1 years (SD 0.4, range 7.2 to 8.8 years) and disease duration at follow-up was 14.3 years (SD 5.5, range 8.5 to 39.0 years). MRI scans at the 8-year follow-up were acquired on 134 patients. All images were analyzable with the BPF software.

As a whole, the group worsened in terms of both disability and brain atrophy. The mean disability level, measured with either the EDSS or MSFC, and the severity of brain atrophy worsened significantly between baseline and the 8-year follow-up point (table 1). At baseline, the mean EDSS was 2.4 (SD 0.9) and at follow-up it had increased to 4.1 (SD 2.0). Fifty-seven (35%) patients had progressed to EDSS ≥ 6.0 by the 8-year follow-up. Similarly, mean MSFC worsened from 0.0 (SD 0.7) at baseline to −1.1 (SD 2.2) at the 8-year follow-up. The mean BPF decreased 3.1% from 0.83 (SD. 0.02) at baseline to 0.80 (SD 0.03) at the 8-year follow-up.

During the placebo-controlled 2-year trial, 49% of the patients in the follow-up group were taking IFN treatment. Treatment was uncontrolled and variable in the 6-year interim between the end of the phase III trial and the follow-up examination. Patients were taking disease-modifying drugs (IFN, glatiramer acetate, pulsed steroids, methotrexate, cyclophosphamide, or azathioprine) during the interim time from the end of the phase III trial until follow-up. The percent time on treatment during the year 2 to follow-up interval was 54.2% for the original treatment group and 46.5% for the original placebo group.

Available images.

As described in the Methods section, two different subgroups of patients were considered to address the specific questions of this study. There were 138 patients who had at least two analyzable MRI scans during the original trial and an EDSS measurement at the 8-year follow-up (including 5 of the patients who died). This subgroup was considered for the first question regarding the relation between atrophy and subsequent disability. For the second two questions on the progression of atrophy, there were 106 patients who had analyzable MRI scans at all four time points. Baseline demographic and disease-related characteristics of these patient subgroups are given in table 2. There were no substantial differences between the subgroups at baseline for any of the variables measured.

Question 1: Relations between atrophy and disability.

As previously reported,17,18⇓ BPF was correlated with concurrent disability measurements at each of the four time points (baseline, year 1, year 2, and the 8-year follow-up). The concurrent correlations ranged from −0.27 (BPF and EDSS at year 1) to 0.51 (BPF and MSFC at year 1). The correlation between BPF and MSFC was consistently stronger than the magnitude of the correlation between BPF and EDSS at each time point. The change in BPF from year 2 to 8-year follow-up was correlated with change in EDSS in the same period (r = −0.31, p = 0.0005) and with change in MSFC in the same period (r = 0.30, p = 0.001).

The relation between brain atrophy during the original trial and subsequent progression of disability progression was studied in the subgroup of patients (n = 138) with at least two analyzable MRI scans from the original trial and an EDSS measurement at the 8-year follow-up. In this subgroup, the mean EDSS at baseline was 2.3 (SD 0.8) and at the 8-year follow-up it had increased to 4.2 (SD 2.3). Forty-four (32%) of these patients had progressed to EDSS ≥ 6.0 by follow-up. The mean MSFC had decreased from 0.055 (SD 0.65) at baseline to −0.95 (SD 2.1) at the 8-year follow-up. The mean percent change in BPF during the original trial was −1.1% (SD 1.3%) over 2 years.

Brain atrophy during the original phase III trial was related to EDSS, MSFC, and disability status at the 8-year follow-up. Percent change in BPF from baseline to year 2 was correlated with EDSS at the 8-year follow-up (r = −0.27, p = 0.001) and with MSFC at the 8-year follow-up (r = 0.35, p < 0.0001). Quartile analysis showed that the patients with the least amount of atrophy during the trial were less likely to have reached EDSS 6.0 at the 8-year follow-up than those with the most amount of atrophy (figure 1). In the quartile of patients with little or no atrophy from baseline to year 2 (change in BPF +1.3% to −0.26%), only 14.3% (SE 6.0%) had reached EDSS ≥ 6.0 by the 8-year follow-up, whereas in the quartile of patients with the largest amount of atrophy during the 2-year trial (change in BPF −.7 to −6.5%), 55.9% (SE 8.4%) had reached EDSS ≥ 6.0 by the 8-year follow-up. BPF change remained a significant predictor of disability status after adjusting for baseline EDSS score in a logistic regression model. The mean BPF at baseline, year 2, and the 8-year follow-up for each of the two EDSS outcome groups are shown in figure 2. In addition to the significant difference in the rate of atrophy during the original 2-year trial, the mean BPF was also significantly different at each time point for the group who reached EDSS ≥ 6.0 by the 8-year follow-up as compared with those who did not.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Patients (n = 138) were categorized into quartiles based on the amount of atrophy during the phase III trial, i.e., change in brain parenchymal fraction (BPF) from baseline to year 2. Bars represent the percent of patients in each quartile who had reached Expanded Disability Status Scale (EDSS) score ≥6.0 by the follow-up examination. Error bars indicate the SEs. There were a total of 44 patients of 138 who reached EDSS ≥ 6.0.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Mean brain parenchymal fraction (BPF) over the course of 8 years for all patients who had BPF measurements at all time points (diamonds, n = 106), for the subgroup of these patients who had reached EDSS ≥ 6.0 by the follow-up examination (squares, n = 32) and for the subgroup who had not reached EDSS ≥ 6.0 by the follow-up examination (circles, n = 74). Error bars indicate ±SEM. Differences in mean BPF between the outcome groups were significant at each point in time. The rate of BPF change during the original 2-year trial was also significantly different between the outcome groups.

Logistic regression analysis was performed to determine which variables from the original trial were independently related to disability status at the 8-year follow-up. Since correlations were stronger between year 2 values and follow-up than between baseline values and follow-up (data not shown), only the year 2 values were considered. The results are summarized in table 3. The following disability and MRI variables were considered for inclusion in the first model (in addition to the demographic and disease-related variables): EDSS at year 2, MSFC at year 2, T2 lesion volume at year 2, T1 lesion volume at year 2, number of enhancing lesions at year 2, and BPF at year 2. A forward stepwise selection procedure was used. None of the demographic or MRI variables at year 2 was an independent predictor of disability status at the 8-year follow-up. Only EDSS and MSFC at year 2 were significant predictors of reaching EDSS ≥ 6 at the 8-year follow-up.

A second model consisting of changes during the original 2-year trial was also tested. In addition to the demographic and disease-related variables, the following disability and MRI variables were considered for inclusion in this model: EDSS change from baseline to year 2, MSFC change from baseline to year 2, number of new and enlarging T2 lesions at year 2, change in T1 lesion volume from baseline to year 2, cumulative number of enhancing lesions from baseline to year 2, and BPF change from baseline to year 2. MSFC change and percent BPF change from baseline to year 2 were the only significant predictors of reaching EDSS ≥ 6 at the 8-year follow-up.

Finally, both models were rerun without EDSS and MSFC in order to determine significant MRI predictors. For the first model (with year 2 values as independent variables), BPF at year 2 and number of relapses during the trial were the only significant predictors of reaching EDSS ≥ 6 at the 8-year follow-up. For the second model (with baseline to year 2 changes as independent variables), percent change in BPF from baseline to year 2 and number of relapses during the trial were the only significant predictors of reaching EDSS ≥ 6 at the 8-year follow-up.

Questions 2 and 3: Progression of atrophy over 8 years.

The progression of brain atrophy over the course of 8 years was studied in the subgroup of patients (n = 106) with analyzable MRI scans at all four time points (baseline, year 1, and year 2 of the original trial and the 8-year follow-up). The severity of brain atrophy worsened significantly between baseline and the 8-year follow-up. The mean percent change in BPF over the 6 years from the end of the phase III trial to the 8-year follow-up was −2.0% (SD 1.9%). Patients who were originally in the placebo group of the phase III trial (n = 52) had a mean BPF change of −3.0% from baseline to follow-up, whereas patients who were originally in the treated group (n = 54) had a mean BPF change of −2.5% from baseline to follow-up. This difference was not significant (p = 0.18). BPF and BPF percent change were not related to the total time on disease-modifying therapies from baseline to the 8-year follow-up.

In general, baseline BPF, follow-up BPF, and percent change in BPF were not correlated with age, sex, or disease duration. The exceptions were that baseline BPF was weakly correlated to age (r = −0.18, p = 0.039) and disease duration (r = −0.29, p < 0.001), but these correlations were not observed at the 8-year follow-up. There was a weak correlation between the percent change in BPF from baseline to 8-year follow-up and the number of relapses during the 2-year controlled study (r = −0.26, p = 0.007). There was a strong correlation between baseline BPF and 8-year follow-up BPF (r = 0.71, p < 0.0001).

BPF was correlated to lesion measurements from the phase III trial (table 4). The volumes of T2 lesions and T1-hypointense lesions were significantly correlated with BPF concurrently and at subsequent time points, with Spearman r values ranging from −0.35 to −0.55. The number and volume of enhancing lesions were only very weakly associated with BPF, if at all. However, in terms of predicting the rate of atrophy over the long term, expressed as the change in BPF from year 2 to the 8-year follow-up, the number of enhancing lesions at year 2 was correlated with subsequent BPF change (r = −0.31, p = 0.002). Other lesion measures, especially at year 2, were also correlated with long-term atrophy rate, including the number of new and enlarging T2 lesions (r = −0.36, p < 0.0001), the volume of T2 lesions (r = −0.35, p = 0.0002), the volume of T1-hypointense lesions (r = −0.36, p = 0.0003), and the cumulative number of enhancing lesions during the trial (r = −0.26, p = 0.008).

Multiple regression analysis was performed to determine which variables at year 2 were independent predictors of subsequent atrophy (i.e., change in BPF from the end of the original trial to the 8-year follow-up). The results are summarized in table 5. The following disability and MRI variables were considered for inclusion in the first model (in addition to the demographic and disease-related variables): EDSS at year 2, MSFC at year 2, T2 lesion volume at year 2, T1 lesion volume at year 2, and number of enhancing lesions at year 2. A forward stepwise selection procedure was used. The only significant predictors of subsequent atrophy were the number of enhancing lesions at year 2 and the volume of T2 lesions at year 2 (R² = 0.29, adjusted R² = 0.27).

A second model including changes during the original trial was also tested. In this model, the following disability and MRI variables were considered for inclusion (in addition to the demographic and disease-related variables): EDSS change from baseline to year 2, MSFC change from baseline to year 2, number of new and enlarging T2 lesions at year 2, change in T1 lesion volume from baseline to year 2, and cumulative number of enhancing lesions from baseline to year 2. The only significant predictors of subsequent atrophy were the number of new and enlarging T2 lesions at year 2 and the MSFC change from baseline to year 2 (R² = 0.29, adjusted R² = 0.28).