If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
EDSS scores is moderately correlated with cervical spinal cord atrophy (CSCA).
•
We explored the main source of heterogeneity in the meta-analysis.
•
We provided further evidence for the clinical predictive value of CSCA.
Abstract
Background
Cervical spinal cord atrophy (CSCA), which partly reflects the axonal loss in the spinal cord, is increasingly recognized as a valuable predictor of disease outcome. However, inconsistent results have been reported regarding the correlation of CSCA and clinical disability in multiple sclerosis (MS). The aim of this meta-analysis was to synthesize the available data obtained from 3.0-Tesla (3T) MRI scanners and to explore the relationship between CSCA and scores on the Expanded Disability Status Scale (EDSS).
Methods
We searched PubMed, Embase, and Web of Science for articles published from the database inception to February 1, 2019. The quality of the articles was assessed according to a quality evaluation checklist which was created based on the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. We conducted a meta-analysis of the correlation between EDSS scores and CSCA at 3T MRI in MS.
Results
Twenty-two eligible studies involving 1933 participants were incorporated into our meta-analysis. Our results demonstrated that CSCA was negatively and moderately correlated with EDSS scores (rs = -0.42, 95% CI: -0.51 to -0.32; p < 0.0001). Subgroup analyses revealed a weaker correlation in the group of relapsing-remitting multiple sclerosis (RRMS) and clinically isolated syndrome (CIS) (rs = -0.19, 95% CI: -0.31 to -0.07; p = 0.0029).
Conclusions
The correlation between CSCA and EDSS scores was significant but moderate. We encourage more studies using reliable and consistent methods to explore whether CSCA is suitable as a predictor for MS progression.
Multiple sclerosis (MS) is a chronic immune-mediated disease that affects the neurological function of individuals in their early life. It is crucial to monitor the clinical outcomes and responses to drug treatment at the early stage of MS. Brain atrophy is thought to be closely associated with disease deterioration in MS (
) and has been accepted as an important endpoint for monitoring treatment response in MS-related clinical trials. However, increasing evidence has demonstrated that MRI-based quantification of cervical spinal cord atrophy (CSCA) is moderately to highly correlated with the clinical status of MS (
). 3T scanners are demonstrated to have better sensitivity and reliability than 1.5 T scanners in measuring the brain volume and cerebral deep gray matter volume in MS (
). The relatively poor spatial resolution of low-field-strength MRI hampers high-quality imaging of small structures such as spinal cord, which is probably the reason for the discrepancies in this clinicoradiological correlation across studies. To increase the signal-to-noise ratio and make the results more convincing, a growing number of researchers have adopted more advanced 3T technology in their studies (
). The objectives of this systematic review and meta-analysis were to synthesize the available data obtained with 3T MRI scanners and to present pooled analyses to inform clinical practice.
2. Materials and methods
2.1 Data sources and searches
We performed and reported the results of this meta-analysis based on the Preferred Reporting Items for Systematic Reviews and Meta-analyses guidelines. We searched PubMed, Embase, and Web of Science for articles published from the database inception to February 1, 2019, without language restrictions. The following keywords were used in the literature search: “multiple sclerosis”, “clinically isolated syndrome”, “MS”, “CIS”, “RRMS”, “SPMS”, “PPMS”, “PMS”, “BMS”, “cord atrophy”, “cord area”, “cord cross-sectional area”, “cord volume”, “cord gray matter atrophy”, “cord gray matter area”, “cord white matter atrophy”, “cord white matter area”, “cervical”, “spinal”, and “thoracic”. We additionally scrutinized the reference lists of reviews in the retrieval results to ensure that no relevant article was omitted.
2.2 Selection criteria
Studies were eligible for our meta-analysis if they (1) enrolled people with clinically isolated syndrome (CIS), relapsing-remitting multiple sclerosis (RRMS), secondary progressive multiple sclerosis (SPMS), or primary progressive multiple sclerosis (PPMS); (2) presented 3T MRI-derived cervical spinal cord area or volume in MS patients; (3) investigated the associations between CSCA and EDSS scores; and (4) provided the regression coefficients (β), Pearson correlation coefficient (r), Spearman correlation coefficient (rs), partial correlation coefficient (rp), or semipartial correlation (rsp) in the papers. We excluded (1) meeting abstracts, posters, case reports or reviews; (2) studies in animals or non-MS patients; (3) studies in which the correlation coefficients or regression coefficients were not available; and (4) studies based on the results from 1.0T, 1.5T or 2.0T MRI.
2.3 Data collection
Two authors (X.S. and D.L.) independently identified the titles and abstracts in the retrieval results to obtain potentially eligible studies and exclude those that were obviously irrelevant and duplicated. Then, data were extracted through reviewing the full texts of the articles fulfilling predetermined selection criteria.
The following key items were extracted on a data extraction form: authors, year of publication, countries, number of participants, basic demographics and clinical characteristics of participants, MRI sequences, spinal cord segmentation methods, software tools assistant for quantitative image analysis, and correlation coefficients. Any inconsistency about the selection of studies or data extraction was discussed in an integrative session and ultimately referred to an independent arbiter for consultation.
2.4 Study quality assessment
Another two authors (Z.L. and H.D.) independently assessed the quality of the studies. Since all studies incorporated into this meta-analysis were cross-sectional or cohort study designs, we created a quality evaluation checklist (Supplementary Table S1) based on the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. The maximum possible score on the checklist was “20″. We regarded studies that scored “13–20″ as high quality, those that scored “7–12″ as moderate quality, and those that scored “0–7″ as low quality.
2.5 Statistical analysis
We conducted meta-analyses in R software with the "metafor" package. If values for the Spearman correlation coefficient (rs) were not available in a particular study, the Pearson correlation coefficient (r), multiple regression coefficient (β), or coefficient of determination of simple linear regression values (R2) was used to estimate rs according to the following formulas (1) (
), or (3) . If the correlation coefficient in a study was reported to be not statistically significant, it was converted to zero effect. Bivariate and partial effect sizes were synthesized in one meta-analysis if (1) the synthesis included mainly bivariate effects (a small number of partial effect sizes) and (2) the summary estimates remained largely unchanged compared with the primary estimate if any single partial effect was removed. Otherwise, partial effect sizes and bivariate correlations were separately meta-analyzed since the effects of combining two types of effect sizes are not completely understood (
). Before conducting a meta-analysis, each correlation coefficient (rs) was converted into its corresponding z value by performing Fisher's z transformation (
We applied the Cochran's Q test and the I2 statistic for the assessment of heterogeneity between studies (I2 < 50% was defined as no statistically significant heterogeneity; I2 > 50% was defined as statistically significant heterogeneity). The selection of fixed-effects or random-effects models was depended on the significance of the I2 index. When significant heterogeneity existed, we attempted to analyze the source of the heterogeneity by performing subgroup analyses on the basis of MS subtype, spinal cord segmentation methods, and image processing software. Sensitivity analyses were performed by deleting a single study each time to evaluate the consistency and robustness of the primary results.
We assessed publication bias by Begg's adjusted rank correlation test and Egger's regression asymmetry test. Meanwhile, a funnel plot was visually evaluated for symmetry. We concluded that no significant publication bias existed if the p-value > 0.05.
3. Result
3.1 Identification and description of studies
A total of 1632 articles were available after searching the three databases. A total of 600 articles remained after duplicate citations were excluded, and an additional 438 studies were further removed after screening the titles and abstracts. By browsing the remaining articles in full, 21 articles were found to meet our eligibility criteria. Furthermore, one additional article was obtained by scrutinizing the reference lists of the reviews. Ultimately, 22 articles with a total number of 1933 subjects were incorporated into our meta-analysis (
Single scan quantitative gradient recalled echo MRI for evaluation of tissue damage in lesions and normal appearing gray and white matter in multiple sclerosis.
All studies were published between 2012 and 2019. Individual study sizes ranged from 15 to 239 participants. The mean age of the subjects ranged from 33.9 to 54.9 years, the proportion of females included in the studies ranged from 44.4% to 87.5%, and the mean disease duration ranged from 3.3 to 20.0 years. The correlation coefficient results varied across studies, with a range from 0 to −0.75. The demographics and clinical characteristics and the CSCA quantification techniques are separately summarized in Tables 1 and 2.
Table 1.Demographics and clinical characteristics of studies.
According to the quality evaluation checklist (Supplementary Table S1) based on the STROBE statement, fourteen articles were evaluated to be of high quality, and eight articles were assessed to be of moderate quality. All articles obtained maximal scores on the items of background, key results, limitations, interpretation, and generalizability. No articles described how the study sample size was arrived at (Supplementary Table S2).
3.3 Overall meta-analysis
Only four studies used partial correlations to describe the linear relationship between EDSS scores and CSCA, and the pooled results remained largely unchanged when we deleted these partial effects one by one. Thus, we incorporated the four partial effects into the overall meta-analysis. Meta-analysis with a random effects model showed that EDSS scores were negatively and significantly correlated with CSCA (rs = −0.42, 95% CI: −0.51 to −0.32; p < 0.0001; I2 = 83%). Notably, significant heterogeneity was observed across studies (I2 = 83%, p < 0.01; Fig. 2). According to the results of the Begg's test (P = 0.53) and the Egger's test (P = 0.38), there was no obvious evidence of publication bias and a funnel plot is shown in Fig. 3. We conducted a sensitivity analysis by omitting studies one by one, and we did not find any obviously changes in the pooled estimates compared with the primary results.
Fig. 2Forest plots of the overall result with corresponding 95% CIs for the correlation between EDSS scores and CSCA at 3T MRI in MS patients. Summary estimates were analyzed using a random-effects model. CI: Confidence interval; COR, correlation coefficient.
Twenty studies enrolled participants with a mix of MS subtypes including CIS, RRMS, PPMS, and SPMS, while only two studies recruited exclusively RRMS patients. Only four studies provided additional correlation coefficients based on the different MS subtypes. Studies that contained people with CIS and RRMS were classified in the relapsing multiple sclerosis (RMS) group, and studies consisting of people with CIS, RRMS, SPMS, and PPMS were classified in the Mix group. We did not further define a progressive MS group because a limited number of SPMS and PPMS patients were available for the pooled meta-analysis. The subgroup analysis presented a weaker correlation coefficient in the RMS group (rs = −0.19, 95% CI: −0.31 to −0.07; I2 = 44%) than in the Mix group (rs = −0.44, 95% CI: −0.53 to −0.34; I2 = 72%). Notable heterogeneity existed in the Mix group (I2 = 72%, p < 0.01; Fig. 4).
Fig. 4Forest plots of the subgroup analysis result with corresponding 95% CIs for the correlation between EDSS scores and CSCA based on different MS subtypes. Summary estimates were analyzed using a random-effects model. Chen ea al-1, Chen ea al-2, Oh et al.−1, Bernistas et al.−1, Lundell et al.−1, and Lundell et al.−2 represented data from the progressive multiple sclerosis group. Chen ea al-3, Oh et al.−2, Bernistas et al.−2, and Lundell et al.−3 represented data from the relapsing multiple sclerosis group. CI: Confidence interval; COR, correlation coefficient; Mix, Mix subgroup; RMS, relapsing multiple sclerosis subgroup.
Half of the studies (n = 11) adopted JIM software of different versions for the quantitative image analysis; the other software tools used in more than one study were JIST (n = 2), Spinal Cord Toolbox (n = 2) and NeuroQLab (n = 3). We divided the studies into the JIM software group and the “other software” group according to the software tools used for quantitative image analysis. The results of the pooled correlation coefficients in the JIM software group tended to be significant (rs = −0.46, 95% CI: −0.53 to −0.38; I2 = 40%). Additionally, the aggregated effect of the “other software” group showed a negative correlation (rs = −0.43, 95% CI: −0.57 to −0.27; I2 = 90%), and obvious heterogeneity was observed (I2 = 90%, p < 0.01; Fig. 5).
Fig. 5Forest plots of the subgroup analysis result with corresponding 95% CIs for the correlation between EDSS scores and CSCA based on software tools assistant for quantitative image analysis. Summary estimates were analyzed using a random-effects model. CI: Confidence interval; COR, correlation coefficient; JIM, JIM software subgroup; Other, “other software” subgroup.
There was a wide variation in the spinal segmentation methods used across studies. The methods adopted in more than two studies were the Losseff method (
). We separated the studies into the Losseff method group, the Horsfield method group and the “other methods” group according to spinal cord segmentation method. Significant heterogeneity existed in the Losseff group (rs = −0.53, 95% CI: −0.80 to −0.09; I2 = 96%) and the “other methods” group (rs = −0.40, 95% CI: −0.52 to −0.26; I2 = 77%) compared with the Horsfield method group (rs = −0.45, 95% CI: −0.54 to −0.36; I2 = 51%; Fig. 6).
Fig. 6Forest plots of the subgroup analysis result with corresponding 95% CIs for the correlation between EDSS scores and CSCA based on different spinal cord segmentation methods. Summary estimates were analyzed using a random-effects model. CI: Confidence interval; COR, correlation coefficient; Horsfield, Horsfield method subgroup; Losseff, Losseff method subgroup; Other, “other methods” subgroup; “ * ”, studies using two spinal cord segmentation methods in one research.
Ten studies explored CSCA at C2/C3 vertebral level, and the remaining studies examined other cervical vertebral levels including C1, C2, C2 to 30 mm above, C3/C4, C2/C5, and C1-C7. We divided the studies into the C2/C3 vertebral level group and “other cervical vertebral level” group. Significant heterogeneity was shown in both groups (C2/C3 vertebral level group: rs = −0.43, 95% CI: −0.56 to −0.28; I2 = 85%; “other cervical vertebral level” group: rs = −0.41, 95% CI: −0.54 to −0.27; I2 = 83%; Fig. 7).
Fig. 7Forest plots of the subgroup analysis result with corresponding 95% CIs for the correlation between EDSS scores and CSCA based on different cervical vertebral levels. Summary estimates were analyzed using a random-effects model. CI: Confidence interval; COR, correlation coefficient; C2/C3, C2/C3 cervical vertebral level subgroup; Other, “levels other cervical vertebral level” subgroup.
Our pooled results showed that EDSS scores had a significant association with CSCA measured by 3T MRI in people with MS. The cervical cord is a crucial crossroads between the brain and the thoracolumbar cord. Though the pathogenesis has not yet been elucidated, axonal loss in the spinal cord is mainly considered to contribute to spinal cord atrophy (
) also likely result in this pathological outcome. Axonal loss, neuronal pathology, and demyelination are recognized as the major pathological substrates of permanent functional disability in MS.
An increasing number of studies have shown cervical cord atrophy to be a predictor of clinical disability independent of brain lesion load or atrophy in MS patients (
). In view of the above findings, CSCA may be a promising biomarker that will better monitor the disease progression or treatment response to novel neuroprotective agents. Our analysis provided further evidence for the clinical predictive value of CSCA.
A significant but moderate correlation between CSCA and EDSS scores was observed in our meta-analysis. The moderate correlation may be ascribed to an intrinsic defect of EDSS (
). Although the EDSS is broadly accepted as an essential MS neurological disability scale, it has limitations in evaluating disability in more severely affected MS patients, which is known as the “ceiling effect”. For example, Lundell et al. found a linear correlation between EDSS scores and CSCA reaching a plateau in more severely affected SPMS patients, and in contrast, the Multiple Sclerosis Impairment Scale still shows a good clinicoradiological correlation in those patients (
). In addition, the EDSS is heavily weighted towards ambulatory function and is insensitive to nonambulation clinical deterioration (e.g., sexual and sphincter function) (
). Hence, EDSS scores change little when the patient's walking ability is not severely impaired. For example, in Viola Biberacher's study, the authors ascribed the weak correlation between CSCA and disease severity to the restricted range of EDSS scores covered by their patient group (
). All the inherent shortcomings of EDSS mentioned above could affect the magnitude of the correlation coefficients. In addition, we should consider the inherent limitations of CSCA. A recent post mortem study found a big difference between spinal cord cross-sectional area reduction (reduction by 20%) and axonal loss (reduction by about 60%), suggesting that CSCA significantly underestimates the degree of axonal loss. Gliosis likely counteracts the spinal cord atrophy caused by nerve fiber loss (
). Moreover, a limited number of studies have demonstrated that spinal cord gray matter atrophy is detectable in relapsing and progressive MS. Moreover, upper cervical cord gray matter atrophy may correlate more strongly with neurological functions than brain or cord white matter atrophy does (
). Thus, these findings raise the possibility that CSCA alone cannot accurately capture the clinically relevant changes in the spinal cord. Further investigation of cord gray matter abnormalities seems worthwhile because it may improve clinical relevance.
Significant heterogeneity was shown in our meta-analysis, although Begg's adjusted rank correlation test and Egger's regression asymmetry test showed no obvious evidence of publication bias in the major outcome. To explore the source of heterogeneity, we performed a subgroup analysis based on different disease subtypes. It was suggested that the EDSS scores were negatively related to CSCA in the RMS and Mix groups. It was noteworthy that obvious heterogeneity existed in the Mix group. The reason may be that the Mix group included four different populations, which encompassed the people suffering from CIS, RRMS, SPMS, and PPMS. In addition, there appeared to be a weaker correlation in the RMS group than in the Mix group. A plausible explanation is that people with CIS or RRMS still have adequate cortical adoption reserve, which might promote rehabilitation after spinal cord or brain injury (
Mohammed, H., Hollis, E.R., 2nd, 2018. Cortical reorganization of sensorimotor systems and the role of intracortical circuits after spinal cord injury. 15(3), 588–603.
). Adaptive changes in the cortical central nervous system allow the maintenance of normal function in the presence of irreversible axonal or neural loss (
). In people with progressive multiple sclerosis, the accumulation of tissue damage seems to exhaust this compensatory mechanism, preventing it from counteracting the functional impact of irreversible structural tissue damage (
). However, there were too few studies in the RMS group, and the subgroup analysis result needs to be interpreted with caution. In addition, combining CIS and RRMS may bias the subgroup result because CIS and RRMS have some differences in clinical and radiological presentations (
). Future studies are encouraged to explore the separate associations in different subtypes of MS patients.
The heterogeneity was lower in the subgroup that applied the Horsfield method for cord segmentation than in the one that used various other segmentation methods. However, the heterogeneity was still significant in the Losseff subgroup. We speculate that the presence of too few studies using the Losseff method may have biased the subgroup results. Another plausible reason may be that the interobserver and intraobserver variability of measurement tend to be lower for the Horsfield method than for the Losseff method (
). The lower variability in the Horsfield method may have contributed to the low heterogeneity in the Horsfield method subgroup. As one might expect, lower heterogeneity was observed in the JIM software subgroup than in the “other software” subgroup, which used a variety of software tools for quantitative image analysis. Moreover, significant heterogeneity was shown in both the C2/C3 vertebral level group and the “other cervical vertebral level” group. Therefore, differences in vertebral levels used in the included studies may not have affected the heterogeneity of the meta-analysis results. In addition, to eliminate the effect of swelling on CSCA measurement, only 13 studies in our meta-analysis provided the definite exclusion criteria of the participants including no relapses and no use of corticosteroids or disease-modifying medications prior to MRI examination. Swelling of the spinal cord may be a confound to the measurement of spinal cord areas, thus resulting in heterogeneity of our meta-analysis results. Collectively, the observations mentioned above highlight the significance of adopting reliable and consistent segmentation methods and software tools as technical heterogeneity may present a challenge to comparability among clinical trials using CSCA as a clinical endpoint (
Validation of mean upper cervical cord area (MUCCA) measurement techniques in multiple sclerosis (MS): high reproducibility and robustness to lesions, but large software and scanner effects.
Several limitations should be presented when interpreting the outcome in this meta-analysis. First, this meta-analysis is mainly based on cross-sectional data, which do not allow us to infer relatively direct relationships between CSCA and neurological functions from the results. Nevertheless, the strong correlation between CSCA changes and worsening of EDSS scores has been evidenced recently in a large cohort of people with relapse-onset MS at 6 years of follow-up (
). Hence, we encourage more longitudinal studies to determine whether CSCA assessments qualify as a valuable predictor for MS progression. Second, there was obvious technical variation in the included articles, ranging from the MRI metric for CSCA quantification, the spinal cord segmentation method, and the image processing software. These technical heterogeneities limit the use of aggregated results as a standard reference for future trials. Third, to yield more accurate results of the correlation between CSCA and EDSS scores, we excluded studies providing 1.0T, 1.5T and 2.0T MRI-based measures of CSCA, which might have biased the results. Fourth, the subgroup analysis needs to be interpreted with caution because the results are weakened by the insufficient number of studies. Finally, the overall population included in the meta-analysis was small, and more large scale research is needed to increase the quality and credibility of the studies.
5. Conclusion
The correlation between CSCA and EDSS scores was significant but moderate. We encourage more studies using reliable and consistent methods to explore whether CSCA is suitable as a predictor for MS progression.
Declaration of Competing Interest
No conflict of interest exists in the submission of this manuscript, and the manuscript is approved by all authors for publication. All the authors listed have approved the manuscript that is enclosed.
Acknowledgments
The study was funded by the National Key R&D Program of China, Precision Medicine Program, Cohort study on nervous system diseases (2017YFC0907700), the National Science Foundation of China (grant number 81571633), Beijing Health System Clinicians Training Plan(grant number 20143054).
Mohammed, H., Hollis, E.R., 2nd, 2018. Cortical reorganization of sensorimotor systems and the role of intracortical circuits after spinal cord injury. 15(3), 588–603.
Validation of mean upper cervical cord area (MUCCA) measurement techniques in multiple sclerosis (MS): high reproducibility and robustness to lesions, but large software and scanner effects.
Single scan quantitative gradient recalled echo MRI for evaluation of tissue damage in lesions and normal appearing gray and white matter in multiple sclerosis.
30406-7&id=gr1.jpg){kind=link}
30406-7&id=gr2.jpg){kind=link}
30406-7&id=gr3.jpg){kind=link}
30406-7&id=gr4.jpg){kind=link}
30406-7&id=gr5.jpg){kind=link}
30406-7&id=gr6.jpg){kind=link}
30406-7&id=gr7.jpg){kind=link}