The long-term prognosis and predictors of epilepsy: a retrospective study in 820 patients

In this retrospective cohort study, epileptic patients at the outpatient clinic of Xuanwu Hospital, Capital Medical University, from October 2017 to January 2020, were included. The inclusion criteria were: 1) having an established diagnosis of epilepsy; 2) treated by the same physician in our center for more than 1 year, with regular evaluation from once a month to once a year; 3) with comprehensive medical records from their first visit to our center to the last visit. Information was mainly obtained from the medical records, consisting of demographics, seizure frequency, age of seizure onset, etiology, the number of AEDs at first visit, family history of epilepsy or febrile convulsions in first-, second- or third-degree relatives, history of febrile convulsions, electroencephalography (EEG) and magnetic resonance imaging (MRI) results. As most patients had received multiple EEG examinations, only EEG data at the first visit to our center were used. To understand their recent condition, the patients were contacted by telephone at the end of the study. If the patients received epilepsy surgery after inclusion in this study, the observation period ended at the time of the epilepsy surgery.

Epilepsy was classified into the idiopathic, symptomatic, and cryptogenic types, according to the guidance of ILAE [15, 16]. Idiopathic epilepsy, such as juvenile myoclonic epilepsy (JME), is defined as an epilepsy that has age-related onset, specific clinical and EEG characteristics, and a presumed genetic etiology. Symptomatic epilepsy refers to a group with an acquired or genetic cause, including cerebral trauma (head injury and neurosurgery), cerebral tumor, meningitis/encephalitis, stroke, cerebral vascular malformations (CMV), mesial temporal sclerosis (MTS), cortical dysplasia (CD), perinatal brain injury, unclear encephalomalacia, neurocutaneous syndromes (tuberous sclerosis and Sturge-Weber syndrome), etc. The determination of causes mainly relies on neuroimaging and the medical history. Cryptogenic epilepsy is considered to have an existing yet unknown cause.

Continuous variables are presented as median, interquartile range (IQR), and range. Categorical variables are presented as counts and percentages. For exploratory purposes, chi-square tests were employed for comparisons of categorical data and the Mann-Whitney test for comparisons of nonparametric continuous data. The variables with significant findings on univariate analysis (P <  0.1) would enter the multivariable model. The Cox proportional hazards model was used to investigate the simultaneous effects of prognostic factors to cause remission or relapse. The logistic-regression analysis was used to identify predictors for terminal drug-resistance. P <  0.05 was considered as statistically significant. Kaplan-Meier survival analysis was first used to estimate the cumulative probability of seizure remission, followed later by considering significant prognostic factors. Statistical analysis was performed using the SPSS software for Windows, version 21.

In this population, 503 (61.3%) patients achieved 1-year remission throughout the entire observation period (Table 1). In fact, the majority (84.3% of 503) started their first 1-year remission within 1 year after the first visit. The cumulative probability of the first 1-year remission for the overall cohort was 53.1% (95% confidence intervals [CI] 51.4–54.8%), 63.9% (62.1–65.7%), 72.6% (70.4–74.8%), 77.9% (75.0–80.8%) at 2, 4, 6 and 8 years after the index date, respectively (Fig. 1a). The median time to achieve the first 1-year remission was 1 year (range: 1.0–8.1; IQR: 1.0–1.6 years). In the group followed up for at least 2 years, 49.3% (330/669) of patients entered a remission for at least 2 years. The proportions of patients achieving a 2-year remission at 2, 4, 6 and 8 years after the first arrival was 30.8% (29.1–32.5%), 48.6% (46.6–50.6%), 58.0% (55.5–60.5%) and 64.2% (61.3–67.1%) (Fig. 1b).

Univariate analysis revealed significant differences in seizure frequency, number of AEDs at first visit, EEG results, MRI and type of epilepsy between patients achieving a 1-year remission and those who did not (Table 1). Multivariate Cox proportional hazard analysis showed that the seizure frequency, the number of AEDs at the first visit, EEG results and the type of epilepsy were independent predictive factors for seizure recurrence (Table 2). Patients with symptomatic or cryptogenic epilepsy were more likely to achieve a remission than patients with idiopathic epilepsy (symptomatic vs idiopathic: P <  0.001; cryptogenic vs idiopathic: P = 0.009). There was also a significant difference in the probability of remission between symptomatic epilepsy and cryptogenic epilepsy who continued to have seizures (P <  0.001). Using survival analysis, longitudinal seizure remission curves were drawn for type of epilepsy (Fig. 2). Patients with high seizure frequency (< 1/month) were 1.48 times more likely to achieve a remission. The patients with normal EEG at the first arrival had higher tendency to achieve a remission than these with epileptiform discharges (P <  0.001). In addition, the patients prescribed with AEDs before the first arrival to our center were less likely to achieve a remission than the newly diagnosed epilepsy patients (1 AED vs 0 AED, P = 0.037; ≥ 2 AEDs vs 0 AED, P = 0.001). Moreover, similar associations were found between 2-year remission and seizure frequency, number of AEDs at first arrival, type of epilepsy and EEG results (Table 2).

Of the 503 patients who achieved a remission, 184 (36.6%) experienced a relapse, including 58 relapses due to external reversible causes and 126 without any known reversible triggers. The external reversible triggers included failure to take medicine occasionally (19 patients), discontinuation of AED treatment (16 patients), reduction of the AED dose (4 patients), severe sleep deprivation (7 patients), fever (4 patients), alcohol (3 patients), emotional change (4 patients), and fatigue (1 patient). Of the 184 patients who experienced a relapse, 81 had a second 1-year remission, and 58 of them had second remission lasting till the end of the study. The second relapse occurred in 23 patients, and the third relapse in 2 patients.

We next investigated predictive risk factors for seizure relapse after achieving long-term (> 1 year) seizure remission. In the univariable analysis, only seizure frequency correlated with relapse (55.4% vs 47.0%, P = 0.069), while other variables, including sex, age of seizure onset, seizure frequency, family history of epilepsy or febrile convulsion, history of febrile convulsion, type of epilepsy, EEG and MRI results did not show a significant association with relapse (P > 0.1) (Table 3). When seizure frequency was included in the Cox proportional hazards model, there was no significant association between seizure frequency and relapse (P = 0.143).

At the end of the study, 322 of 820 patients (39.3%) met the criteria of drug resistance. The prevalence of DRE was 54.4% (160 of 294) in symptomatic epilepsy, 33.7% (135 of 401) in cryptomatic epilepsy, and 21.6% (27 of 125) in idiopathic epilepsy. In particular, 16.1% (52 of 322) of patients with DRE had experienced a remission.

In the univariable analysis, the following variables were associated with DRE: seizure frequency and the number of AEDs at first visit, EEG results, MRI and the type of epilepsy (Table 4). Multivariable analysis showed that the seizure frequency and the number of AEDs at first visit, EEG results, and the type of epilepsy remained significantly associated with the likelihood of DRE at the end of the study (Table 5). Patients with symptomatic epilepsy had the highest probability of DRE, followed sequentially by those with cryptomatic epilepsy, and those with idiopathic epilepsy (symptomatic vs idiopathic: OR = 4.70 (2.74–8.07), P <  0.001; cryptomatic vs idiopathic: OR = 2.26 (1.34–3.82), P = 0.002; symptomatic vs cryptogenic: OR = 2.08 (1.44–3.00), P <  0.001).

In addition, we analyzed the possibility of terminal DRE among patients with symptomatic epilepsy of different causes, and found that the patients with etiology of meningitis/encephalitis (OR: 2.99, 95% CI: 1.32 - 6.77, P = 0.007) were more likely to develop into DRE than patients with other etiologies (Table 6).

At the end of observation time, 40 subjects (4.9%) had been off-medication, 393 (47.9%) were receiving monotherapy, 308 (37.6%) were receiving 2 AEDs, 76 (9.3%) were treated with three drugs, and 3 (0.4%) were taking four. For the patients who were not taking drugs at the end of the study, 80.0% (32 of 40) of the patients were in a remission, including 21 with idiopathic epilepsy (17 with BECTS, 2 with CAE, 1 with JME, 1 with Panayiotopoulos syndrome), 1 with symptomatic epilepsy and 10 with cryptomatic epilepsy.

During the observation time, 61.3% of patients with epilepsy achieved a 1-year remission of seizures and 49.3% had a 2-year remission, which clearly showed that the majority of patients could experience a period of seizure freedom. Previous studies [6, 8‐10, 14] have consistently shown that remission is more likely to occur in idiopathic epilepsy and less likely in nonidiopathic epilepsy. However, there are contradictions about the outcomes between cryptomatic epilepsy and symptomatic epilepsy. Some studies have concluded that there is no significant difference in the outcomes between cryptomatic and symptomatic epilepsies [6, 17], while others having not [9, 12]. The present study supports that patients with cryptomatic epilepsy have a better prognosis than these with symptomatic epilepsy. Instead of regarding cryptogenic epilepsy as probably symptomatic, it may be more scientific to consider it as a separate entity with a relatively good prognosis. In addition, we found that patients with AEDs at the first visit had a much lower probability of remission than patients not taking drugs. This is not difficult to explain. Patients who had failed 1 or more AEDs tended to search for better physicians than those who had achieved seizure control. Therefore, this group of patients with a previous treatment had been screened and tended to have a bad prognosis, especially when at least 2 AEDs had failed in them. Moreover, here the abnormal MRI results were associated with remission in univariate analysis, but lost its significance in multivariate analysis. The reason for the above inconsistency may be that the MRI results were highly correlated with the classification of epilepsy.

We examined the probability of seizure relapse in those who had experienced a remission. We found that 184 of 503 (36.6%) patients relapsed during the observation time after achieving a remission. Interestingly, this degree of relapse was consistent with that seen by Schiller et al. [4], who found that 40% of patients who achieved a greater than 1-year remission experienced seizure relapse at 5 years after entering seizure remission. Another study reported that in a cohort of 59 patients with refractory epilepsy who entered a 1-year remission, 34 (57.6%) of them had a relapse [18]. The main difference between this and our cohorts was that Callaghan et al. included only patients with refractory epilepsy, while our study included all patients with epilepsy. All these studies demonstrate that even several years of remission could not guarantee permanent remission. In other words, those who are in a remission may still need many years to get over this disease completely. Physicians must be cautious when discussing prognosis of patients who are in a seizure remission, particularly when planning to decrease dosage or stop medications.

In this study, only a small number of patients experienced a relapse due to medication changes, while a majority (68.5%) of them experienced a relapse without definite causes, suggesting a fluctuating nature of epilepsy course even with medication. This may be explained by two reasons: the development of drug tolerance after prolonged AED exposure, and the progression of underlying epileptogenesis. In the present cohort, no factors were found to be associated with relapse, including the type of epilepsy and the seizure frequency, which was consistent with previous studies [4, 10, 19].

In our cohort, the prevalence of DRE was 39.3%, which is similar to 33% reported by Jose et al. [20]. In both studies, the new definition of ILAE was used. In particular, about one sixth of patients with DRE had experienced a certain period of remission during the observation time, which indicates that drug responsiveness is a dynamic process. Consistent with previous reports, patients with encephalitis/meningitis had the poorest outcomes. Téllez-Zenteno et al. compared causes in patients with DRE and these without DRE. In their study, DRE was found in 71.4% of patients due to cerebral infection, which was higher than other causes [20].

A major strength of our study is the large size of the cohort. Furthermore, all the patients were followed up by the same physician for a long time, with comprehensive medical records, which made it possible to analyze seizure control during the whole observation time. A major limitation is the retrospective cohort study design, which relies on medical records as the major source of information, while other important information may be missed out. Another limitation is the lack of accurate information about treatment efficacy before the first arrival to our center, as some patients had been treated in different hospitals for many years. Therefore, we only examines long-term prognosis from the first arrival to our center.

In conclusion, more than half of patients experienced a remission during the course of epilepsy. However, the remission is not necessarily a persisting process, and the “remitting-relapsing” course may be common. Several factors are related to remission and terminal DRE, such as the type of epilepsy, initial seizure frequency and treatment history. In the group of symptomatic epilepsy, patients with encephalitis/meningitis are significantly more likely to be drug-resistant than those with other causes. These prognostic factors are present early in the course of epilepsy, and can provide reference for making more effective therapies.