Predictors of seizure outcomes of autoimmune encephalitis: a clinical and morphometric quantitative analysis study

Abstract

Objective:

Autoimmune encephalitis can be followed by treatment-resistant epilepsy. Understanding its predictors and mechanisms are crucial to future studies to improve autoimmune encephalitis outcomes. Our objective was to determine the clinical and imaging predictors of postencephalitic treatment-resistant epilepsy.

Methods:

We performed a retrospective cohort study (2012–2017) of adults with autoimmune encephalitis, both antibody positive and seronegative but clinically definite or probable. We examined clinical and imaging (as defined by morphometric analysis) predictors of seizure freedom at long term follow-up.

Results:

Of 37 subjects with adequate follow-up data (mean 4.3 yrs, SD 2.5), 21 (57%) achieved seizure freedom after a mean time of 1 year (SD 2.3), and one third (13/37, 35%) discontinued ASMs. Presence of mesial temporal hyperintensities on the initial MRI was the only independent predictor of ongoing seizures at last follow-up (OR 27.3, 95%CI 2.48–299.5). Morphometric analysis of follow-up MRI scans (n=20) did not reveal any statistically significant differences in hippocampal, opercular, and total brain volumes between patients with postencephalitic treatment-resistant epilepsy and those without.

Significance:

Postencephalitic treatment-resistant epilepsy is a common complication of autoimmune encephalitis and is more likely to occur in those with mesial temporal hyperintensities on acute MRI. Volume loss in the hippocampal, opercular, and overall brain on follow-up MRI does not predict postencephalitic treatment-resistant epilepsy, so additional factors beyond structural changes may account for its development.

Introduction

Autoimmune encephalitis is increasingly recognized as a cause of acute symptomatic seizures and can be complicated by recurrent unprovoked seizures beyond the encephalitic illness, at which point it is referred to as autoimmune-associated epilepsy.^1–3 Autoimmune-associated epilepsy is typically resistant to both conventional epilepsy therapies (antiseizure medications (ASMs), surgery) and immunotherapy and is therefore a major contributor to the morbidity and mortality among autoimmune encephalitis survivors.¹ The majority of the literature has examined seizure outcomes in specific antibody-mediated encephalitides (eg cell-surface antibodies^4–7, LGI1 only,⁸ and NMDA-R only⁹), with discordant results on the frequency of persistent seizures after encephalitic illness and its predictors. One study found that the type of antibody target (cell surface vs intracellular) was the primary predictor of seizure outcomes after autoimmune encephalitis,³ but it remains unclear whether this finding is related to an ongoing autoimmune cause of seizures in the follow-up period or structural damage incurred by different immunological mechanisms during the initial encephalitic illness. There is a need to understand the predictors of treatment-resistant epilepsy after encephalitis to elucidate its mechanisms.

Because our aim was to determine broad downstream mechanisms of postencephalitic epilepsy across the spectrum of autoimmune encephalitides, we chose to investigate the predictors of seizure outcomes in a rigorously selected cohort of autoimmune encephalitis with and without neural autoantibodies targeting both cell surface and intracellular antigens. After identifying acute imaging findings as an independent predictor of postencephalitic epilepsy, we then performed morphometric analysis of follow-up brain MRI scans to determine whether structural changes follow encephalitis and are the primary determinant of postencephalitic epilepsy.

Methods

Patient population.

We performed a search through our electronic health records for adult patients who had received ICD-10 codes of encephalitis, along with medical history or diagnoses consistent with encephalitis, seen at NYU Langone Medical Center between 01/01/2012 and 12/31/2017 (Supplementary information). We then manually searched all charts (n=990) to determine whether subjects met either of the following categories:

1. Neural autoantibody positive autoimmune encephalitis as defined by the following:

   1. The following neural autoantibodies were included: high titer GAD65 (>20nmol/L or >200 IU/mL in serum), Hu, Ri, amphiphysin, Ma2, GABA-B, GABA-A, LGI1, CASPR2, NMDA-R, GFAP, or MOG.
      AND
   2. A diagnosis of possible autoimmune encephalitis according to the Graus¹⁰ autoimmune encephalitis criteria

2. Autoantibody negative definite limbic encephalitis, according to the Graus¹⁰ autoimmune encephalitis criteria

3. Autoantibody negative but probable autoimmune encephalitis, according to the Graus¹⁰ autoimmune encephalitis criteria

For each of the three above diagnostic categories, we also required patients to either have documented follow-up for at least 1 year after onset of signs and symptoms of autoimmune encephalitis in the electronic medical record including documentation of the outcomes of interest (use of antiseizure medications (ASMs), persistence of seizures). In those who did not have documented follow-up, we included those who consented to participate in a telephone survey targeted at seizure history, ASM use, and relapse history (Supplementary information).

Outcome definition.

We defined seizure freedom or persistence as presence of any seizures (focal aware, focal impaired awareness, or focal to bilateral tonic-clonic) at the last follow-up, defined either through neurology clinical note documentation specifically inquiring about seizure frequency or, if the subjects had not had any follow-up beyond 1 year from symptom onset, per telephone survey. When patients reported a period of at least 6 months of seizure freedom, the date of the last seizure was documented. Additional outcome measures were ongoing ASM use, date of last ASM use when patients were no longer taking ASMs, and any relapse.

Predictor variables.

We performed a chart review to collect clinical data on patient demographics, clinical characteristics, acute EEG findings, acute MRI findings, cerebrospinal fluid (CSF) profile, and immunotherapy type and timing of initiation compared to symptom onset.

Imaging.

MRI acquisition.

Imaging was obtained in study participants as part of each individual’s clinical assessment at the NYU Langone Medical Center in the post-acute period. Clinical imaging was obtained using both 1.5 T scanners (N = 11) and 3T scanners (N = 9). Images were visually reviewed and any scans with excessive motion or other imaging artifacts were excluded from our analyses. Because imaging was obtained clinically, image acquisition parameters were not standardized; however, all participants were imaged on Siemens MRI scanners using T1-weighted whole-brain MPRAGE with slice thickness = 1mm. In-plane voxel size ranged from 0.5 × 0.5 mm to 1 × 1 mm. Image acquisition parameters for all participants are provided in Supplementary Table 1.

MRI processing.

MRI scans were processed using Freesurfer v6.0 with default image processing settings. Volumes of interest included left and right hippocampus, obtained from the standard “aseg” labels obtained as part of the subcortical segmentation routines provided with Freesurfer, and left and right opercular volumes. Because there is no pre-defined operculum label output by Freesurfer, we estimated the volume of the left and right opercular regions by summing the volumes of a number of regions covering the operculum defined using the Destrieux cortical parcellation atlas.¹¹ These specific labels are provided as supplementary material.

Statistical analysis.

Clinical predictors of outcome.

Descriptive statistics were performed for each variable including means, medians, and standard deviations for continuous variables and frequencies for categorical variables. Bivariate analysis was first performed using Wilcoxon rank-sum and Fisher’s exact tests to compare patients with postencephalitic epilepsy to those who achieved seizure freedom. Variables identified on each analysis with P < 0.1 were then subjected to a multivariate logistic regression model to study the probability of seizure freedom. Results with P < 0.05 after multivariate analysis were considered statistically significant.

Imaging predictors of outcome.

We then analyzed morphometric MRI characteristics among the cohorts with versus without epilepsy after encephalitis. The relationship between left and right hippocampal and opercular volume and seizure outcomes (seizure-free or not at follow-up) were investigated using multiple linear models, with the volume of interest as the dependent variable and seizure-free status, age at time of MRI scan, sex, and total brain volume as predictor variables. We carried out power analyses to determine the minimum volume change detectable using our imaging data. Standard T-test-based power analysis methods were used, with standard deviations estimated from each of our volumes of interest, power = 0.8, and alpha = 0.05.

Results

Sample characteristics

Of the 990 patients receiving a diagnostic code consistent with encephalitis at NYU between 01/01/2012 and 12/31/2017, a total of 44 met diagnostic criteria for possible autoimmune encephalitis with the presence of a neural autoantibody, definite limbic encephalitis but no neural autoantibodies, or seronegative probable autoimmune encephalitis. Of these, 37 had either documented neurologic follow-up 1 year from symptom onset or were contacted to administer a telephone survey to ascertain seizure outcomes (Figure 1). Of these, 26 were antibody positive (NMDA-R n=9, GAD65 n=7, LGI1 n=8, Hu n=1, GABA-B n=1), and 11 were antibody negative. General clinical characteristics were similar to that of the published literature on autoimmune encephalitis (Supplementary Table 2), with the majority experiencing acute symptomatic seizures, requiring a high number of antiseizure medications (median 3). About one fourth experienced relapses, which presented in the majority of cases (6/9) with seizures.

Figure 1.

Patient identification. We identified a large cohort of patients who were coded in electronic health record as having an autoimmune encephalitis diagnosis, then performed a chart review to determine which patients met diagnostic criteria for antibody-positive autoimmune encephalitis or, if antibody negative, definite limbic encephalitis or probable autoimmune encephalitis. 37 patients total were identified with adequate follow-up data.

Seizure outcome

Over half of patients (21/37, 57%) achieved seizure freedom for at least 6 months prior to the last follow-up, after a mean time of 1 year from the onset of encephalitis (SD 2.3). One third (13/37, 35%) were able to discontinue antiseizure medications, and therefore, definitively did not have postencephalitic epilepsy, after a mean time of 1.95 years (SD 2.1).

Clinical predictors of postencephalitic epilepsy

We examined clinical predictors of ongoing seizures within 6 months of last follow-up, ie treatment resistant postencephalitic epilepsy. Of note, many patients still taking ASMs without seizures had never had a trial off ASMs, making it difficult to ascertain whether they had epilepsy or not. We found the following predictors to be associated with postencephalitic treatment-resistant epilepsy on bivariate analysis (Table 1): higher number of seizure types, lack of admission to the ICU, MRI showing mesial temporal hyperintensities in the acute phase, presence of interictal epileptiform discharges, antibody negative, GAD65 antibodies (vs NMDA-R being predictive of seizure freedom), higher number of ASMs used, and higher number of relapses. After logistic regression analysis, only mesial temporal hyperintensities on acute MRI were predictive of postencephalitic epilepsy (eg Fig. 2). Because of this finding, we explored next whether structural changes in the hippocampal regions could account for postencephalitic epilepsy.

Table 1.

Clinical predictors of seizure outcomes (bivariate analysis)

Predictor	OR or proportions	P value
Number of seizure types	OR 0.299 (SE 0.587)	0.0396
Admission to the ICU	5/11 (not seizure-free) vs 14/20 (seizure free)	0.02
MRI showing mesial temporal hyperintensity in the acute phase	10/12 (not seizure-free) vs 4/15 (seizure free)	0.0063
Interictal epileptiform discharges	9/16 (not seizure-free) vs 3/21 (seizure free)	0.0124
Antibody positive	8/16 (not seizure-free) vs 18/21 (seizure-free)	0.03
Antibodies		0.0192
GAD65	4/16 (not seizure-free) vs 3/21 (seizure-free)
LGI1	3/16 (not seizure-free) vs 5/16 (seizure-free)
NMDA	0/16 (not seizure-free) vs 9/16 (seizure-free)
Number of ASMs used	OR 0.57 (SE 0.21)	0.009
Number of relapses	OR 0.19 (SE 0.76)	0.03

Figure 2.

Imaging example. This 69-year-old man presented with rapidly progressive encephalopathy and frequent facial motor seizures, requiring escalation of multiple ASMs and anesthetic agents. MRI in the acute phase showed bilateral mesial temporal hyperintensities (panels A and B), and CSF showed lymphocytic pleocytosis. No antibodies were identified. A diagnosis of definite limbic encephalitis was made. He continued experiencing facial motor seizures after treatment and with 4 years of follow-up and was classified as treatment-resistant postencephalitic epilepsy in our cohort.

Neuroimaging predictors of postencephalitic epilepsy

Among the 37 subjects with outcome data, 20 subjects had follow-up brain MRIs, with T1 acquisition adequate for morphometric analysis, performed in the recovery phase or after their encephalitic illness. We explored volume as defined by T1 morphometric analysis, specifically in the hippocampus and the perisylvian region. We chose these regions because our findings indicate that acute changes in the mesial temporal regions predicted postencephalitic treatment-resistant epilepsy and added the perisylvian regions because of their implication in postencephalitic epilepsy by previous electroclinical studies, imaging studies, and intracranial EEG recordings. There were no statistically significant differences in hippocampal, opercular, and total brain volumes between the seizure-free and not seizure-free groups (Table 2).

Table 2.

Imaging characteristics of postencephalitic epilepsy versus seizure-free patients

Structure	Seizure-Free (mm3, mean ± SD)	Not Seizure-Free (mm3, mean ± SD)	p-value	Minimum detectable change (mm3)
Left Hippocampus	4196 ± 1396	3741 ± 1066	0.32	1473
Right Hippocampus	3947 ± 1009	3522 ± 691	0.268	1017
Left Operculum	22763 ± 4736	22926 ± 3154	0.81	4721
Right Operculum	22227 ± 4142	21758 ± 2941	0.86	4237
Intracranial Volume	1428647 ± 264205	1423757 ± 136631	0.94	239862

Discussion

In this study, we report seizure outcomes in a cohort of antibody positive (NMDA-R, LGI1, GAD65, Hu, GABA-B) and negative autoimmune encephalitis and limbic encephalitis. Using rigorous outcome ascertainment, we found over half of patients achieved seizure freedom, and one third successfully discontinued ASMs. On a multivariate analysis of clinical predictor variables, we found that the presence of mesial temporal hyperintensities on brain MRI during encephalitis was the only independent predictor of ongoing seizures and, therefore, investigated whether structural changes on follow-up MRI could account for this propensity for ongoing seizures. Using morphometric analysis of follow-up brain MRIs, we did not identify any significant volumetric differences between those who developed epilepsy in the chronic phase and those who did not.

When compared to the existing literature,^{3, 4, 6} epilepsy at last follow-up was noted at a higher rate in our cohort. Discordant outcomes have been reported in the literature, which may be related to outcome ascertainment (chart review versus direct interview of patients who may dismiss focal aware seizures unless specifically questioned), length of follow-up (some patients may receive immunotherapy in a delayed manner and experience a longer time to response), and immunotherapy approaches (use of prompt and aggressive immunotherapy may differ according to centers and may have changed in recent years). We included patients seen between 2012–2017, which may have led to worse outcomes, since one sixth of patients did not receive immunotherapy, and the median number of immunotherapies used was 2, despite half requiring ICU stay. It is also possible that some of our patients were categorized as having epilepsy at last follow-up but were still experiencing acute symptomatic seizures related to an active autoimmune encephalitis. There is no current operational definition of autoimmune associated epilepsy versus acute symptomatic seizures related to autoimmune encephalitis, and no specific timeline has been recommended to make that distinction.¹ In the absence of such an operational definition, we arbitrarily chose to include those with follow-up greater than 1 year and measured the outcome of interest after this timepoint, although the majority of patients had much longer follow-up time (mean 4.3 years).

Prior studies have aimed to identify predictors of postencephalitic epilepsy, since the rational targeting of a group at higher risk of long-term complications (eg postencephalitic epilepsy) for more aggressive therapy would inform the design of future interventional studies and help counsel patients and their families in the acute phase. Prior studies have identified EEG abnormalities,^{5, 6} delay in administration of immunotherapy,⁶ younger age at disease onset,⁸ female sex,⁸ and intracellular targets³ as predictors of ongoing seizures at last follow-up. The finding that mesial temporal hyperintensities are an independent predictor of postencephalitic epilepsy is novel and raises the possibility that more severe involvement of the mesial temporal structures as measured by MRI FLAIR changes may predict structural changes, which could then act as a trigger for recurrent unprovoked seizures in the long term. Prior studies have shown that hippocampal changes can occur in the chronic phase of LGI1-antibody mediated encephalitis¹² and that hippocampal atrophy correlates with poor cognitive outcomes. To our surprise, there was no significant difference in hippocampal volumes in those continuing to have seizures compared to those who were seizure-free. While the lack of differences between the two groups could have been due to a lack of power (see Table 3 for minimum detectable change), this raises the possibility that structural changes in the hippocampi may be necessary, but not sufficient, for the development of postencephalitic epilepsy. Additional factors, such as ongoing neuroinflammation even after the initial encephalitic illness, may differentiate the postencephalitic group from the one recovering from encephalitis without developing epilepsy. Future studies should investigate biomarkers of inflammation beyond the encephalitic phase of illness and their role in determining outcomes, as has already been done in preliminary studies of cell-surface antibody mediated encephalitis.¹³ In addition, other imaging modalities, such as PET imaging (through FDG or other radioligands tailored to imaging of neuroinflammation, e.g. TSPO-PET) may be helpful in future studies to investigate the predictors of transgression of encephalitis into epilepsy.

We also explored the role of the perisylvian (insular-opercular) region in our morphometric analysis, given the prior evidence for the involvement of this region in electroclinical,^{14, 15} imaging,¹⁶ MEG,¹⁷ and intracranial EEG studies¹⁸ of autoimmune encephalitis and postencephalitic epilepsy. No differences were found, which could have been related to statistical power but also could be related to an immune, not structural, injury of the perisylvian regions in postencephalitic epilepsy.

Limitations

Our small sample size likely impacted our ability to evaluate the effect of weak clinical predictors or small morphometric differences on seizure outcomes. To counteract our small sample size, we aimed to minimize variability in our cohort through rigorous inclusion criteria, as the presence of a pathogenic antibody or strict clinical diagnostic criteria in the antibody negative cohort were necessary for inclusion. We also required a minimum of 1 year follow-up, given that acute symptomatic seizures may take months to resolve after diagnosis and initiation of immunotherapy.⁴ While most of the existing literature has been antibody specific, we chose to include a heterogeneous group of autoimmune encephalitides, spanning various antibody types (cell surface and intracellular) and including definitive but antibody negative groups, in order to investigate broad mechanisms and predictors of postencephalitic epilepsy across the spectrum of disease. We included antibody types, including GAD65 and NMDA-R antibodies, in the logistic regression, and MRI findings of mesial temporal changes remained significantly predictive of outcome. Nevertheless, the heterogeneity of the cohort may have impacted our ability to evaluate antibody-specific effects on specific brain regions and their impact on outcome.

Conclusions

Our study demonstrates that postencephalitic treatment-resistant epilepsy is a significant complication of autoimmune encephalitis, and while mesial temporal FLAIR hyperintensities can predict postencephalitic epilepsy, there are no morphometric differences in the hippocampal, perisylvian, and total brain volumes between those who recover from autoimmune encephalitis without epilepsy and those who develop recurrent seizures. The latter finding raises the possibility of additional mechanisms of postencephalitic epilepsy beyond structural injury, including autoimmunity, which may play a role beyond the acute phase.

Highlights.

• Over half of autoimmune encephalitis survivors become seizure-free.

• The presence of mesial temporal FLAIR hyperintensities is associated with postencephalitic epilepsy.

• There are no structural differences between those with and without postencephalitic epilepsy as measured by morphometric analysis of hippocampal, perisylvian, and total brain volumes.

Abbreviations:

ASM

Antiseizure medications

CSF

cerebrospinal fluid

EEG

electroencephalography

FLAIR

flui-attenuated inversion recovery

GAD65

glutamic acid decarboxylase 65

LGI1

leucine-rich glioma inactivated 1

MEG

magnetoencephalography

MRI

magnetic resonance imaging

NMDA-R

N-methyl-D-aspartate receptor