Role of MRI in the differentiation of ADEM from MS in children

Acute demyelination of the CNS may be a transient illness or may represent the first clinical attack of multiple sclerosis (MS). The clinical features of acute demyelination may localize to a single site or may invoke polyfocal CNS involvement. When polyfocal demyelination is accompanied by encephalopathy, the clinical diagnosis of acute disseminated encephalomyelitis (ADEM) is applied.¹ MRI features of ADEM typically include widespread, bilateral, and asymmetric involvement of supratentorial and infratentorial white matter, deep gray nuclei, and spinal cord. Although ADEM is classically considered to be a monophasic disorder, at least 18% of all children ultimately diagnosed with MS will experience a first demyelinating event clinically indistinguishable from typical ADEM.² Descriptive studies in which the MRI appearance of patients with ADEM are compared with MRI scans obtained during the first attack of MS have determined that the MRI features of these two clinical scenarios cannot be reliably distinguished.^3,4

We apply comprehensive quantitative and qualitative analytical methodologies to MRI images obtained at the time of a first demyelinating event in children with monophasic ADEM¹ and children with clinically definite MS.⁵ We propose MRI criteria that may aid clinicians in distinguishing the first attack of MS from ADEM in children.

METHODS

RESULTS

Demographic features of the ADEM and MS groups are compared in table 1. As anticipated, children with ADEM were younger than children presenting with the first attack of MS (p = 0.016). There were more males in the ADEM group (female:male ratio 0.53:1) relative to the MS group (female:male ratio 1.2:1). The female:male ratio in the MS group is lower than the 2:1 or 3:1 ratio reported in other series, in which female preponderance is particularly notable in patients with adolescent-onset disease.⁹ The fact that 11 (39%) of the MS group in the present study experienced their first attack of MS before age 10 years likely accounts for the reduced female:male ratio.

The mean T2 lesion counts for each location and size category are summarized in table 2. Only the periventricular lesion counts differed between the ADEM and MS group (Cohen d = 0.9, p = 0.0036).

The qualitative variables are summarized in table 3. Although every attempt was made to ensure that a standardized definition was applied, “diffuse bilateral lesions” remained a subjective variable. Prospective application of this criterion by other investigators is required to determine whether subjectivity will limit the validity of this component of the criteria. Figures e-1 and e-2 (on the Neurology® Web site at www.neurology.org) illustrate the MRI appearance of the qualitative variables typical of ADEM and MS in children.

Spinal cord imaging was performed in four children with ADEM (all had only one spinal lesion, extending more than three vertebral segments in length in three children) and six children with MS (three had only one lesion, two had two discrete lesions, and one had three spinal lesions; only one patient had lesions extending beyond three spinal cord segments).

Based on the results of Student t tests and Fisher exact tests, the following categories were given the opportunity to enter the forward stepwise conditional logistic regression analysis: mean periventricular white matter lesion count, presence of lesions perpendicular to the long axis of the corpus callosum, sole presence of discrete lesions, bilateral diffuse lesion distribution, and presence of black holes. The final regression equation for predicting membership to the MS patient group included 1) periventricular white matter lesions (odds ratio [OR] = 1.99, 95% CI 1.21–3.29, p = 0.007), 2) presence of black holes (OR = 15.92, 95% CI 1.32–192.51, p = 0.029), and 3) diffuse bilateral lesion distribution (OR = 0.62, 95% CI 0.003–1.30, p = 0.073). The model correctly classified 88.6% of the study population (sensitivity = 90.0%, specificity = 87.5%, positive predictive value [PPV] = 91.3%, negative predictive value [NPV] = 85.7%; χ² = 31.75, df = 3, p < 0.0001).

After identification of the three categories (periventricular lesions, black holes, and the absence of diffuse bilateral lesions), the model was further refined by determining cutoff values for the mean periventricular lesion count. Maximum classification accuracy was achieved for the following combination of variables: any two of 1) ≥2 periventricular lesions, 2) the presence of black holes, and 3) lesion distribution pattern that was not diffuse bilateral. Based on these criteria, the accuracy for classifying a patient as MS at first attack, as opposed to ADEM, was found to be 83% (sensitivity = 81%, specificity = 95%, PPV = 95%, NPV = 79%). These results are summarized in the figure.

• Download figure
• Open in new tab
• Download powerpoint

Figure Summary of the diagnostic criteria for classifying MS patients at the time of their first demyelinating episode vs ADEM patients

Part A summarizes the combination of terms required to be met to be classified as having first attack of multiple sclerosis (MS). B provides the breakdown of the classification of the population used in the current study when the criteria were applied. C summarizes the calculated accuracy statistics for the model. D outlines the number of MS patients meeting all or part of the stated criteria. *Four of the 28 MS patients did not have T1-weighted imaging available. However, 2 of these patients met criteria for MS despite the inability to determine the presence/absence of black holes. ADEM = acute disseminated encephalomyelitis; PPV = positive predictive value; NPV = negative predictive value.

DISCUSSION

We provide a comprehensive characterization of the presenting MRI features of children with monophasic ADEM, compare these findings with those of children at the time of their first attack of relapsing–remitting MS (RRMS), and propose MRI criteria to aid in distinguishing the onset of pediatric RRMS from ADEM.

As shown in a companion article, we propose modifications to the McDonald criteria for lesion dissemination in space that are more sensitive for the appearance of MS in children. The proposed criteria for pediatric MS differ from the McDonald MRI criteria for MS in adults as follows: two or more of (rather than three criteria) 1) ≥5 T2 rather than ≥9 T2 lesions (2), ≥2 periventricular lesions rather than >3, and (3) ≥1 brainstem rather than ≥1 infratentorial lesion. The criterion of ≥1 juxtacortical lesion was not found to contribute to the pediatric MRI criteria, and gadolinium-enhancing lesions were not included because many children do not receive gadolinium. Spinal lesions were not evaluated because few children had spinal imaging, and thus the ability of spinal lesions to replace infratentorial lesions could not be evaluated. However, as summarized in table 4, the proposed pediatric MS criteria were met not only by 75% of the 28 children with MS at the time of their first attack, but also by 75% of the children with ADEM. These results suggested that although our proposed pediatric MS criteria are valid as a means of distinguishing MS from nondemyelinating acute neurologic disorders in children, they lack the specificity to distinguish MS from monophasic demyelination (ADEM). As shown in table 4, the McDonald MRI criteria for lesion dissemination in space⁶ also do not distinguish pediatric MS from ADEM (30% specificity). The KIDMUS pediatric MS MRI criteria,⁷ although very specific for MS (100%), were relatively insensitive in distinguishing the MS and ADEM populations (29%). Thus, we determined that quantitative and qualitative MRI analyses were required to determine whether MRI was a useful tool to distinguish children with MS and ADEM at onset.

Lesion number and distribution were remarkably similar between children with their first attack of MS and those with ADEM (p = 0.117; tables 2 and 3), and thus total lesion count was not used to formulate our proposed MS vs ADEM criteria. The marked similarity in total lesion numbers suggests that the degree of CNS inflammation is not itself predictive of MS risk. Future studies evaluating total T2 lesion volume are required to further validate this statement. Only the mean number of periventricular lesions was significantly greater in the MS group compared with the ADEM group (7.5 ± 9.9 vs 1.4 ± 2.3, p = 0.004). This supports the importance of periventricular lesion location as a characteristic feature of MS in children, just as it is in adults.⁶

Black holes were scored as a dichotomous variable (presence or absence) and were found to be a powerful distinguishing feature between the MS and ADEM groups (58% of MS patients, 5% of ADEM children, p < 0.01). The mean age of MS patients with black holes (9.8 years) did not differ from the mean age of MS patients without black holes (10.8 years), suggesting that presence of black holes is not age dependent. It is of relevance, however, that only 4 MS patients had nonenhancing black holes—considered a potential hallmark of long-standing regional degeneration.¹⁰ The majority of black holes detected did enhance, indicative of a more acute process. Serial imaging is required to determine whether these regions resolve or remain as chronic lesions. The near absence of black holes, including enhancing black holes, in the ADEM patients suggests that the pathologic processes underlying ADEM may be less injurious. The capacity for resolution of all visible lesions in children with ADEM, as was seen in 14 of the 20 ADEM patients studied (represented in figure e-1B), supports this concept.

Diffuse, bilateral lesions (represented in figure e-1A) also emerged as an important differentiating feature between the MS and ADEM groups. However, diffuse bilateral lesions occurred in three of the younger MS patients (ages 2.1, 4.5, and 8.8 years): one with optic neuritis, and two with clinical features at first attack meeting criteria for ADEM.¹ All three have subsequently experienced more than two non-ADEM demyelinating attacks. The MRI features of these three children demonstrate that the MRI appearance may evolve from the diffuse bilateral pattern toward a pattern more consistent with typical pediatric MS, as has been reported.¹¹ Diffuse, bilateral lesions were absent in the remaining 7 MS patients presenting before age 10 years and were not detected in any of the 18 MS patients with onset after age 10 years. Of the 20 children with ADEM, 9 had diffuse bilateral lesions, and all were younger than 10 years (mean age 4.9 years). The remaining 11 ADEM patients without diffuse lesions were older (mean age 9.9 years); 3 were adolescents. Thus, diffuse bilateral lesion formation seems to be a characteristic of young children, and further studies are required to determine whether this MRI feature will remain as a valuable determinant of monophasic ADEM when considered independent of age at onset. Younger children with MS (aged <10 years) were significantly more likely to have large lesions than were older MS patients (p < 0.05). The proclivity for large lesions in younger patients, both those with ADEM and those with MS, has also been noted in previous studies.^12,13 Large lesions, or longitudinally extensive lesions involving the spinal cord, have also been associated with ADEM and with neuromyelitis optica, in contrast to spinal lesions in MS, which tend be small (one to two spinal segments in length). Longitudinal lesions were detected in three of the four ADEM patients and one of six MS patients for whom spinal imaging was performed at presentation. The paucity of spinal imaging precluded inclusion of this information in the present analysis but will be an important aspect of future studies. Whether the capacity for diffuse bilateral and large cerebral lesions has any relationship to the state of myelin maturation or to an age-related capacity for more widespread CNS inflammation remains to be determined.

Important for the present study was our decision to apply recently proposed clinical guidelines for the diagnosis of ADEM and MS.¹ We felt strongly that the role of consensus criteria is to better define a population, and thus we prioritized our work to reflect these criteria. For the MS population, we included only children with clinical evidence of relapsing disease. We did not include children with MRI evidence only for dissemination of disease activity in time, because we did not want to bias our MS cohort to those children with active MRI disease. For the ADEM population, we did not evaluate children with polyfocal demyelination in the absence of encephalopathy, and given that the focus of our work was to evaluate imaging features, we determined inclusion in the ADEM group blinded to the MRI appearance. Future studies of children with clinical and radiographic features of polyfocal demyelination in the absence of encephalopathy are required to determine whether such children follow a monophasic course or whether they develop features consistent with MS.

A retrospective study design was necessary to ensure enrollment of children with sufficient clinical observation to establish their diagnoses of RRMS and monophasic ADEM. It remains to be shown whether the proposed MRI criteria will be predictive of monophasic ADEM when applied to a prospective population of children evaluated at the time of their acute illness. The number of patients studied was small, and the marked similarity in total lesion numbers between the ADEM and MS groups implies that much larger patient numbers are required to detect any differences between these two patients groups, if indeed these distinctions in total lesion number exist. We must also acknowledge that although none of the children included in the monophasic ADEM group had evidence of further clinical or radiographic disease for at least 2 years (mean total observation 4.1 years), it remains possible that a proportion of these children will ultimately experience further attacks leading to a diagnosis of MS. This point is particularly relevant given that the time from first demyelinating attack to second attack is longer in children with an ADEM-like first MS attack.¹⁴ Serial MRI studies are required to validate the expected absence of accrual of clinically silent lesions in children with ADEM—a feature that would be distinct from the expected lesion accrual seen in children with MS. We also cannot comment on the MRI appearance of children with recurrent or multiphasic ADEM. Given that these entities (as defined by recently proposed criteria)¹ are rare,¹⁵ multicenter studies will be required to accrue a sufficient number of such patients for formal MRI analysis.

Our proposed criteria can only be applied to distinguish children with the clinical presentation of ADEM¹ from that of a first MS attack. The MRI features predictive of monophasic disease in children with acute optic neuritis, transverse myelitis, or other clinical manifestations of acute CNS demyelination are likely to be very different. In a study of 36 children with acute optic neuritis, the presence of even a single T2-weighted lesion separate from the optic nerves was associated with a 68% likelihood of MS diagnosis within 2 years.¹⁶ None of the children with optic neuritis and normal brain MRI were diagnosed with MS during the same period of observation. The mere presence or absence of brain lesions thus seems to have significant implications for MS risk in children with optic neuritis. Clearly, however, lesion presence or absence alone would not distinguish ADEM from MS, because all children with ADEM had multiple MRI lesions.

Prospective evaluation of a large cohort of children with acute demyelination is required to determine the predictive utility of our proposed criteria for future MS diagnosis and to evaluate the diagnostic import of serial MRI analyses.

AUTHOR CONTRIBUTIONS

Statistical analysis was performed by D.J.A.C. and D.S.