Variation in Treatment and Outcomes of Children With Acute Disseminated Encephalomyelitis

This retrospective cohort study was approved by the institutional review board with a waiver of informed consent. In this study, we used data from the Pediatric Health Information System (PHIS), an administrative database that contains inpatient data from 48 US tertiary care children’s hospitals (Children’s Hospital Association, Lenexa, KS). The electronic medical record (EMR) at 1 hospital was used to establish the diagnostic accuracy of International Classification of Diseases, Ninth Revision (ICD-9) discharge diagnosis codes for ADEM. The PHIS database was then used to describe variation in treatment and outcomes across hospitals for ADEM and to identify factors associated with worse outcomes. The PHIS database includes demographic information and up to 41 ICD-9 diagnosis and 41 ICD-9 procedure codes for each encounter. Daily billing data were collected for each hospitalization, and patients were tracked longitudinally across admissions by using a unique patient identifier. Further clinical information, including clinical documentation or laboratory results, are not included within the database.

A validation cohort from a large freestanding tertiary care pediatric hospital was created to determine the accuracy of the ICD-9 codes. Data from the PHIS database for children admitted to the hospital were extracted with the ICD-9 code 323.61, with no other exclusion criteria at that time. The EMR for these patients was reviewed to validate the clinical diagnosis of ADEM. Specificity was calculated by comparing the number of patients identified by PHIS as having ADEM with those diagnosed by the treating pediatric neurologist as having ADEM at the time of evaluation. Sensitivity was determined by comparing data for eligible patients in PHIS with patients with a diagnosis of ADEM in a prospectively collected database of neurocritical care consultations at the same institution.

On the basis of this validation cohort, a strategy for exclusion was devised for the full cohort to be studied. Admissions in PHIS were included for patients >30 days old with a diagnosis of ADEM (ICD-9 code 323.61) from January 1, 2010, to September 30, 2015. Patients were excluded if any encounter included a diagnosis code for multiple sclerosis (MS), neuromyelitis optica (NMO), or any diagnosis found to be misclassified as ADEM on the basis of the validation cohort described above.

The data extracted included demographics, level of care (ICU versus non-ICU), and treatment with steroids (methylprednisolone, prednisolone, prednisone, and dexamethasone), IVIg, or plasmapheresis. These treatments were defined as standard treatments for the purpose of this study. Additionally, the number of treatment days and the time from admission to initiation of treatment were collected. ICU admission was defined by admission to an ICU at any point during the hospitalization.

The primary outcomes for this study were discharge location (home, home with health care services [HHS], skilled facility, death, or other), readmissions with an ICD-9 diagnosis code for ADEM (ADEM readmissions), and readmissions within 90 days without an ICD-9 diagnosis code for ADEM (non-ADEM readmissions). A favorable outcome was defined as disposition to home, and an unfavorable outcome was defined as disposition to HHS, a skilled facility, death, or other.

Categorical variables were summarized by using frequencies with percentages, and continuous variables were summarized by using medians and interquartile ranges (IQRs). Clinical characteristics and treatments were compared across children who required ICU admission and those who did not in a bivariate manner by using χ² statistics. Generalized linear mixed-effects models for readmission and disposition accounted for patient clustering within hospital with a random intercept and important clinical and demographic factors. All analyses were performed by using SAS version 9.4 (SAS Institute, Inc, Cary, NC). P < .05 was considered statistically significant.

A total of 27 patients were identified in the PHIS database from the validation hospital with an ICD-9 code for ADEM (323.61) before any exclusion criteria was applied; of these, 26 had clinical data available in the EMR. Of those patients, 23 had a clinical diagnosis of ADEM after the EMR review (88% specificity). Three patients (12%) were assigned an ICD-9 code for ADEM but were subsequently identified as having alternative diagnoses on the EMR review. These diagnoses included complex febrile convulsions, a metabolic disorder, and hemophagocytic lymphohistiocytosis. At the time of the last follow-up from admission, ranging from a single hospital follow-up appointment to continuous follow-up for 5 years, the specificity was 65%. During follow-up, 4 patients were subsequently diagnosed with relapsing remitting MS or clinically isolated syndrome. No patients were subsequently identified as having NMO.

On review of the local neurocritical care database, 7 patients were identified as having ADEM. On review of the EMR, 6 of these patients had a diagnosis consistent with ADEM. One of these patients was later reclassified as having primary central nervous system angiitis. Of the 6 patients, 5 were identified by the PHIS database (83% sensitivity).

The PHIS cohort consisted of 1269 admissions, of which 152 were excluded. The final cohort consisted of 954 patients and 1117 admissions (Supplemental Fig 2). The mean volume of patients with a diagnosis of ADEM was 4.1 per hospital per year (95% confidence interval [CI]: 3.4–4.9). (Supplemental Table 5). The median age at diagnosis was 6 years (IQR 3–10 years), and 55% of patients were boys.

A total of 34% (382 of 1117) of admissions required ICU-level care. ICU admission was associated with an increased length of stay (LOS) compared with admissions not requiring ICU care (11 days [IQR 6–19] vs 5 days [IQR 4–8]; P < .001). There was no seasonal variation in the incidence of ADEM or in the need for ICU care (Table 1).

A total of 85% (945 of 1117) of admissions for ADEM included billing for at least 1 standard immunosuppressive treatment (ie, steroids, IVIg, or plasmapheresis). Of admissions, 15.4% (172 of 1117) had no standard treatment (Table 2). Admissions included billing for treatment with the following as monotherapy or combination therapy: 80% (893 of 1117) steroids, 22% (224 of 1117) IVIg, and 7% (74 of 1117) plasmapheresis. For admissions with monotherapy, steroids were administered in 93% (664 of 714) of admissions, IVIg was administered in 7% (48 of 714) of admissions, and plasmapheresis was administered in 2 admissions. When steroids were administered, methylprednisolone was the most common (89% [796 of 893] of admissions). Combination therapies were provided in 21% (231 of 1117) of admissions.

ICU admission was associated with increased use of all therapies, alone and in combination (P < .001). First-line treatment differed between hospitalizations with and without ICU admission: IVIg was used in 9% (34 of 382) of ICU admissions versus 7% (49 of 735) of hospitalizations without ICU admission; plasmapheresis was used in 2.6% (10 of 382) vs 0.1% (1 of 735), respectively (P < .001).

The frequency of treatments used significantly differed across hospitals (P < .0001) (Fig 1). Steroid monotherapy was used in a median 62% (IQR 45.5%–73.4%) of hospitalizations depending on the hospital. A median 12% (IQR 8.3%–19.2%) of hospital admissions did not use any standard treatment (Fig 1).

The median time to initiation of standard therapies, regardless of whether the therapy was the initial therapy, was 1 day (IQR 0–2 days) from admission for steroids, 3 days (IQR 1–7 days) for IVIg, and 5 days (IQR 2–8 days) for plasmapheresis. The time to treatment (initial therapy or in combination) was longer for hospitalizations with ICU admission compared with those without ICU admission: 2.0 vs 1.7 days for steroids (P = .001), and 4.8 vs 3.1 days for IVIg (P < .001). When considering only initial therapy, steroids were administered slightly later in ICU admissions compared with non-ICU admissions (0.8 vs 0.6 days; P = .014). When IVIg and plasmapheresis were given as initial therapies, there was no difference based on ICU admission.

For steroids, the median duration of treatment was 5 days for all admissions (IQR 4–6 days). Steroid treatment was prolonged for ICU admissions compared with non-ICU admissions (6.5 vs 4.4 days; P < .001). For IVIg, the median duration of treatment was 3 days for all admissions (IQR 2–4 days). IVIg treatment was prolonged for ICU admissions compared with non-ICU admissions (2.8 vs 2.3 days; P = .004). For plasmapheresis, the median number of days of therapy for all admissions was 5 days (IQR 5–6 days) regardless of the level of care required (Table 3).

For the 15% (172 of 1117) of admissions without any billing for a standard ADEM therapy, the most common additional diagnoses were an unspecified infection, convulsions, constipation, lack of coordination, and altered mental status (Supplemental Table 6). Of therapies given in >5% of these admissions, methotrexate was administered in 8% (14 of 172) and was the only immunomodulating therapy identified. Indications for methotrexate were unclear from the provided diagnostic codes.

A total of 85% (949 of 1117) of admissions had a favorable outcome (discharge from the hospital to home). The remaining 15% (168 of 1117) of admissions had an unfavorable outcome (discharge to other than home), including 16 patients who died (Table 1, Supplemental Table 7). Each day of ICU care was associated with a 10% decrease in the odds of a favorable outcome (adjusted odds ratio [aOR]: 0.90; 95% CI: 0.88–0.93; P < .001). Admissions with an ICU LOS ≥7 days had an odds ratio of 0.1 for a favorable outcome (95% CI: 0.1–0.1; P < .001). There was no association with treatment choice and outcome after adjusting for length of ICU stay.

There were 117 children who had >1 ADEM hospitalization. Of those who received a standard treatment during their index encounter, 12.7% (105 of 824) were readmitted for ADEM. Of those who did not receive a standard treatment during their index encounter, 9.2% (12 of 130) were readmitted for ADEM (P = .256). Having a previous ICU admission was also not associated with the occurrence of an ADEM readmission (P = .143).

A total of 16% (181 of 1101) of admissions (excluding those resulting in mortalities) were readmitted within 90 days of hospital discharge for a diagnosis other than ADEM. The 5 most common diagnoses for readmission were infections (23%; 42 of 181), encephalitis or encephalomyelitis (5.5%; 10 of 181), optic neuritis (4.4%; 8 of 181) or demyelinating disease unspecified (3.9%; 7 of 181), and convulsions (2.7%; 5 of 181). ICU LOS was the only factor associated with an increase in the adjusted odds of being readmitted with a non-ADEM diagnosis (aOR: 1.03; 95% CI: 1.01–1.05; P = .006). Admissions with an ICU LOS ≥7 days had an odds ratio of 1.9 (95% CI: 1.1–3.1; P = .017) for a non-ADEM readmission (Table 4, Supplemental Table 8).

Our results reflect the characteristics, treatments, and outcomes of the largest published cohort of hospital admissions for children with ADEM. The principal findings are as follows: (1) immunotherapy treatment regimens varied significantly across centers, (2) immunotherapy choice was not associated with readmission (ADEM or non-ADEM) or discharge location after adjusting for ICU-level care, and (3) 34% of hospitalizations for ADEM in children included ICU admission, which was associated with increased immunotherapy, additional health services at discharge, and readmission for diagnoses other than ADEM. These findings have implications for the treatment of ADEM as well as for ICU care and follow-up for patients with ADEM.

Our data revealed significant variability between centers in the use of standard treatments of ADEM. Additionally, we did not find an association between the use of standard treatments and our outcomes for those hospitalized for ADEM. In terms of clinical outcomes for ADEM, a previous study revealed that 89% of patients had a normal neurologic examination at follow-up despite wide variability in treatment.⁴  Another larger multicenter study⁵  that included 122 children and 106 adults revealed that 78% of patients who were initially diagnosed with ADEM had either a monophasic or multiphasic illness (68% and 10%, respectively); 95% of the pediatric patients had a favorable outcome (modified Rankin Scale 0–2). In these studies, treatment regimens were variable, but steroids were the most used therapy. These findings are not surprising for several reasons. First, it is becoming apparent that ADEM is a heterogenous syndrome, and the diagnosis is evolving into distinct phenotypes (eg, myelin oligodendrocyte glycoprotein–associated ADEM).¹⁸  Second, a guideline for the diagnostic evaluation and treatment of ADEM does not exist. Last is the potential that ADEM is a monophasic illness that does not respond to current therapies.¹⁹  These findings suggest that standardization is needed for the diagnosis, treatment, and evaluation of clinically meaningful outcomes for children with a presumed working diagnosis of ADEM.

Next, we found that one-third of hospitalizations for ADEM required ICU admission. In a single-center study of 86 children with ADEM, nearly half were admitted to the ICU, and 1-sixth required mechanical ventilation.³  In another study of 84 children with ADEM,⁴  43% of children required ICU admission and 16% required mechanical ventilation. We also found that ICU admission for ADEM was associated with an increase in the number of treatments, in the duration of treatment, and in LOS. Importantly, a prolonged ICU admission ≥7 days was associated with increased risk of an unfavorable outcome and readmission for a diagnosis other than ADEM within 90 days. Improved understanding of those factors (eg, specific antibodies, clinical practice management, and follow-up care) that play a role in the risk for ICU admission and hospital readmission is needed because they may impact outcomes. For example, ADEM associated with high anti–myelin oligodendrocyte glycoprotein titers is associated with an increased risk of recurrence and may impact ADEM readmissions.¹⁸  Children with ADEM who require ICU admission may have become medically complex and benefit from a standardized approach to follow-up to identify long-term developmental, physical, and social challenges, which could contribute to readmissions without a diagnosis of ADEM.

This study has several limitations. First is the PHIS database, which is limited to ICD-9 diagnostic codes. We attempted to maximize our sensitivity and specificity for ADEM by validating the ADEM cohort at a tertiary care children’s hospital. Additionally, there were likely encounters coded with less specific diagnoses, which would not have been captured. Our sensitivity analysis was limited to patients admitted to the ICU and included in the neurocritical care database, and as such, the sensitivity of the ICD-9 code for patients without admission to the ICU was not determined. We may have missed ADEM cases and erroneously categorized them as non-ADEM admissions if a symptom-based diagnostic code was used or if the diagnosis of ADEM was unknown or changed during the admission. Treatments were assumed to have been administered on the basis of billing data in PHIS. It is possible in some cases that such treatments were not given, resulting in an overestimate of the number of treatments given. The doses of steroids and IVIg and the volumes exchanged during plasmapheresis could not be determined. Although dosing of IVIg and volumes exchanged for plasmapheresis are relatively standard, the dosing of steroids may have significant variability and cannot be determined from this database. Criteria for ICU admission vary between hospitals, and ICU admission may not truly be indicative of a higher risk for an unfavorable outcome or readmission.

Our outcomes were limited to discharge locations and readmissions because functional outcomes were not included in this administrative and billing data set. The indication for home services and use of rehabilitation facilities likely varies between centers and depends on many factors, such as insurance coverage and parental preferences. In this cohort, the percentage of discharges from the hospital to home is similar to previous data on favorable outcomes, which is supportive of its use as a proxy for minimal disability.^3–5  We found that readmission for ADEM accounted for ∼10% of admissions. Authors of other studies have reported relapse rates of 10% to 31%.⁵  The definition of relapse used in these studies is variable, with some defining all recurrences as relapse and others defining relapse as recurrences <90 days. In our study, we may have underestimated the relapse rate because we included only those who were readmitted to the same hospital and may have missed relapses that could be managed in the outpatient setting, those not coded with an ADEM diagnosis, or admitted to a different hospital. Additionally, patients not specifically diagnosed with ADEM during a readmission, but with either more specific diagnosis codes (optic neuritis) or less specific diagnosis codes (encephalitis or encephalomyelitis and demyelinating disease unspecified) may have represented relapses that were miscategorized as a non-ADEM readmission, although these may represent continued disability and not new symptoms.

Significant variation in the immunosuppressive treatment of ADEM exists across centers, although steroids were the predominant therapy. Admission to the ICU for ADEM was common and associated with increased immunosuppressive therapy, additional health services at discharge, and readmissions for diagnoses other than ADEM. Further multicenter studies with standardized protocols and adaptive trial designs could potentially improve our understanding of the complex interactions between disease severity, treatment choices, risk factors associated with ICU admission, and rare pediatric neurologic conditions, including ADEM.

We thank the Children’s Hospital Association for providing the data used for this study, Elizabeth Alpern for assisting in the creation of the neurocritical care multicenter workgroup, Allison Rusie for maintaining the neurocritical care database at Lurie Children’s Hospital, and the Ruth D. and Ken M. Davee Pediatric Neurocritical Care Program for their support.