Initial response to antiepileptic drugs in patients with newly diagnosed epilepsy

Predicting efficacy of antiseizure medication treatment with machine learning algorithms in North Indian population

PURPOSE

This study aimed to develop a classifier using supervised machine learning to effectively assess the impact of clinical, demographical, and biochemical factors in accurately predicting the antiseizure medications (ASMs) treatment response in people with epilepsy (PWE).

METHODS

Data was collected from 786 PWE at the Outpatient Department of Neurology, Institute of Human Behavior and Allied Sciences (IHBAS), New Delhi, India from 2005 to 2015. Patients were followed up at the 2nd, 4th, 8th, and 12th month over the span of 1 year for the drugs being administered and their dosage, the serum drug levels, the frequency of seizure control, drug efficacy, the adverse drug reactions (ADRs), and their compliance to ASMs. Several features, including demographic details, medical history, and auxiliary examinations electroencephalogram (EEG) or Computed Tomography (CT) were chosen to discern between patients with distinct remission outcomes. Remission outcomes were categorized into 'good responder (GR)' and 'poor responder (PR)' based on the number of seizures experienced by the patients over the study duration. Our dataset was utilized to train seven classical machine learning algorithms i.e Extreme Gradient Boost (XGB), K-Nearest Neighbor (KNN), Support Vector Classifier (SVC), Decision Tree (DT), Random Forest (RF), Naïve Bayes (NB) and Logistic Regression (LR) to construct classification models.

RESULTS

Our research findings indicate that 1) among the seven algorithms examined, XGB and SVC demonstrated superior predictive performances of ASM treatment outcomes with an accuracy of 0.66 each and ROC-AUC scores of 0.67 (XGB) and 0.66 (SVC) in distinguishing between PR and GR patients. 2) The most influential factor in discerning PR to GR patients is a family history of seizures (no), education (literate) and multitherapy with Chi-square (χ2) values of 12.1539, 8.7232 and 13.620 respectively and odds ratio (OR) of 2.2671, 0.4467, and 1.9453 each. 3). Furthermore, our surrogate analysis revealed that the null hypothesis for both XGB and SVC was rejected at a 100 % confidence level, underscoring the significance of their predictive performance. These findings underscore the robustness and reliability of XGB and SVC in our predictive modelling framework.

SIGNIFICANCE

Utilizing XG Boost and SVC-based machine learning classifier, we successfully forecasted the likelihood of a patient's response to ASM treatment, categorizing them as either PR or GR, post-completion of standard epilepsy examinations. The classifier's predictions were found to be statistically significant, suggesting their potential utility in improving treatment strategies, particularly in the personalized selection of ASM regimens for individual epilepsy patients.

Clinical prediction models for treatment outcomes in newly diagnosed epilepsy: A systematic review

Up to 35% of individuals diagnosed with epilepsy continue to have seizures despite treatment, commonly referred to as drug-resistant epilepsy. Uncontrolled seizures can directly, or indirectly, negatively impact an individual's quality of life. To inform clinical management and life decisions, it is important to be able to predict the likelihood of seizure control. Those likely to achieve seizure control will be able to return sooner to their usual work and leisure activities and require less follow-up, whereas those with a poor prognosis will need more frequent clinical attendance and earlier consideration of epilepsy surgery. This is a systematic review aimed at identifying demographic, clinical, physiological (e.g., electroencephalographic), and imaging (e.g., magnetic resonance imaging) factors that may be predictive of treatment outcomes in patients with newly diagnosed epilepsy (NDE). MEDLINE and Embase were searched for prediction models of treatment outcomes in patients with NDE. Study characteristics were extracted and subjected to assessment of risk of bias (and applicability concerns) using the PROBAST (Prediction Model Risk of Bias Assessment Tool) tool. Baseline variables associated with treatment outcomes are reported as prognostic factors. After screening, 48 models were identified in 32 studies, which generally scored low for concerns of applicability, but universally scored high for susceptibility to bias. Outcomes reported fit broadly into four categories: drug resistance, short-term treatment response, seizure remission, and mortality. Prognostic factors were also heterogenous, but the predictors that were commonly significantly associated with outcomes were those related to seizure characteristics/types, epilepsy history, and age at onset. Antiseizure medication response was often included as a baseline variable, potentially obscuring other factor relationships at baseline. Currently, outcome prediction models for NDE demonstrate a high risk of bias. Model development could be improved with a stronger adherence to recommended TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) practices. Furthermore, we outline actionable changes to common practices that are intended to improve the overall quality of prediction model development in NDE.

Excessively increased thalamocortical connectivity and poor initial antiseizure medication response in epilepsy patients

Objectives

The network mechanism underlying the initial response to antiseizure medication in epilepsy has not been revealed yet. Given the central role of the thalamus in the brain network, we conducted a case-control study to investigate the association between thalamic connectivity and medication response.

Methods

We recruited 39 patients with newly diagnosed and medication-naïve epilepsy of genetic or unknown etiology, including 26 with a good response (GR group) and 13 with a poor response (PR group), and 26 matched healthy participants (control group). We measured the gray matter density (GMD) and the amplitude of low-frequency fluctuation (ALFF) of bilateral thalami. We then set each thalamus as the seed region of interest (ROI) to calculate voxel-wise functional connectivity (FC) and assessed ROI-wise effective connectivity (EC) between the thalamus and targeted regions.

Results

We found no significant difference between groups in the GMD or ALFF of bilateral thalami. However, we observed that the FC values of several circuits connecting the left thalamus and the cortical areas, including the bilateral Rolandic operculum, the left insula, the left postcentral gyrus, the left supramarginal gyrus, and the left superior temporal gyrus, differed among groups (False Discovery Rate correction, P < 0.05), with a higher value in the PR group than in the GR group and/or the control group (Bonferroni correction, P < 0.05). Similarly, both the outflow and the inflow EC in each thalamocortical circuit were higher in the PR group than in the GR group and the control group, although these differences did not remain statistically significant after applying the Bonferroni correction (P < 0.05). The FC showed a positive correlation with the corresponding outflow and inflow ECs for each circuit.

Conclusion

Our finding suggested that patients with stronger thalamocortical connectivity, potentially driven by both thalamic outflowing and inflowing information, may be more likely to respond poorly to initial antiseizure medication.

Metabolic network is related to surgical outcome in temporal lobe epilepsy with hippocampal sclerosis: A brain FDG-PET study

BACKGROUND AND PURPOSE

The aim of this study was to investigate differences in metabolic networks based on preoperative fluorodeoxyglucose (FDG)-positron emission tomography (PET) in temporal lobe epilepsy (TLE) with hippocampal sclerosis (HS) between patients with complete seizure-free (SF) and those with noncomplete seizure-free (non-SF) after anterior temporal lobectomy.

METHODS

This study was retrospectively performed at a tertiary hospital. We recruited pathologically confirmed 75 TLE patients with HS who underwent preoperative FDG-PET. All patients underwent a standard anterior temporal lobectomy. The surgical outcome was evaluated at least 12 months after surgery, and we divided the subjects into patients with SF (International League Against Epilepsy [ILAE] class I) and those with non-SF (ILAE class II-VI). We evaluated the metabolic network using graph theoretical analysis based on FDG-PET. We investigated the differences in network measures between the two groups.

RESULTS

Of the 75 TLE patients with HS, 32 patients (42.6%) had SF, whereas 43 patients (57.3%) had non-SF. There were significant differences in global metabolic networks according to surgical outcomes. The patients with SF had a lower assortative coefficient than those with non-SF (-0.020 vs. -0.009, p = .044). We also found widespread regional differences in local metabolic networks according to surgical outcomes.

CONCLUSION

Our study demonstrates significant differences in preoperative metabolic networks based on FDG-PET in TLE patients with HS according to surgical outcomes. This work introduces a metabolic network based on FDG-PET and can be used as a potential tool for predicting surgical outcome in TLE patients with HS.

Predicting the antiepileptic drug response by brain connectivity in newly diagnosed focal epilepsy

OBJECTIVE

Growing evidence has suggested that epilepsy is a disease with alterations in brain connectivity. The aim of this study was to investigate whether the changes in brain connectivity can predict the response to an antiepileptic drug (AED) in patients with a newly diagnosed focal epilepsy of unknown etiology.

METHODS

This observational study was independently performed at two tertiary hospitals (Group A and B). Thirty-eight patients with newly diagnosed focal epilepsy of unknown etiology were enrolled in Group A and 46 patients in Group B. We divided these patients into two groups according to their seizure control after AED treatment: AED good and poor responders. We defined the AED good responders as those in whom had seizure free for at least the last 6 months while AED poor responders who were not. All of the subjects underwent diffusion tensor imaging, and graph theoretical analysis was applied to reveal the brain connectivity. We investigated the difference in the clinical characteristics and network measurements between the two groups.

RESULTS

Of the network measures, the assortativity coefficient in the AED good responders was significantly higher than that in the AED poor responders in both Groups A and B (- 0.0239 vs. - 0.0473, p = 0.0110 in Group A; 0.0173 vs. - 0.0180, p = 0.0024 in Group B). The Kaplan-Meier survival analysis revealed that the time to failure to retain the first AED was significantly longer in the patients with assortative networks (assortativity coefficient > 0) than in those with disassortative networks (assortativity coefficient < 0) in Group B.

CONCLUSION

We demonstrated that the assortativity coefficient differed between patients with newly diagnosed focal epilepsy of unknown etiology according to their AED responses, which suggests that the changes in brain connectivity could be a biomarker for predicting the responses to AED.

Prediction of antiepileptic drug treatment outcomes of patients with newly diagnosed epilepsy by machine learning

OBJECTIVE

The objective of this study was to build a supervised machine learning-based classifier, which can accurately predict the outcomes of antiepileptic drug (AED) treatment of patients with newly diagnosed epilepsy.

METHODS

We collected information from 287 patients with newly diagnosed epilepsy between 2009 and 2017 at the Second Affiliated Hospital of Zhejiang University. Patients were prospectively followed up for at least 3 years. A number of features, including demographic features, medical history, and auxiliary examinations (electroencephalogram [EEG] and magnetic resonance imaging [MRI]) are selected to distinguish patients with different remission outcomes. Seizure outcomes classified as remission and never remission. In addition, remission is further divided into early remission and late remission. Five classical machine learning algorithms, i.e., Decision Tree, Random Forest, Support Vector Machine, XGBoost, and Logistic Regression, are selected and trained by our dataset to get classification models.

RESULTS

Our study shows that 1) compared with the other four algorithms, the XGBoost algorithm based machine learning model achieves the best prediction performance of the AED treatment outcomes between remission and never remission patients with an F1 score of 0.947 and an area under the curve (AUC) value of 0.979; 2) The best discriminative factor for remission and never remission patients is higher number of seizures before treatment (>3); 3) XGBoost-based machine learning model also offers the best prediction between early remission and later remission patients, with an F1 score of 0.836 and an AUC value of 0.918; 4) multiple seizure type has the highest dependence to the categories of early and late remission patients.

SIGNIFICANCES

Our XGBoost-based machine learning classifier accurately predicts the most probable AED treatment outcome of a patient after he/she finishes all the standard examinations for the epilepsy disease. The classifier's prediction result could help disease guide counseling and eventually improve treatment strategies.

Initial Response to Antiepileptic Drugs in Patients with Newly Diagnosed Epilepsy As a Predictor of Long-term Outcome

Objective

To investigate the correlation between initial response to antiepileptic drugs (AEDs) and long-term outcomes after 3 years in patients with newly diagnosed epilepsy.

Methods

This prospective study included 204 patients with newly diagnosed epilepsy, who were followed-up for at least 36 months. The long-term seizure freedom at 36 months (36MSF) was evaluated in patients with seizure freedom 6 months (6MSF) or 12 months (12MSF) after initial treatment vs those with no seizure freedom after the initial 6 months (6MNSF) or 12 months (12MNSF). Univariate analysis and a multiple logistic regression model were used to analyze the association of potential confounding variables with the initial response to AEDs.

Results

The number of patients with 36MSF was significantly higher for patients that had 6MSF (94/131, 71.8%) than those that had 6MNSF [16/73, 21.9%; χ² = 46.862, p < 0.0001, odd ratio (OR) = 9.051]. The number of patients with 36MSF was significantly higher in patients that had 12MSF (94/118 79.7%) than those that had 12MNSF (19/86, 22.1%; χ² = 66.720, p < 0.0001, OR = 13.811). The numbers of patients that had 36MSF were not significantly different between patients that experienced 6MSF and 12MSF or between patients that had 6MNSF and 12MNSF. Abnormalities observed in magnetic resonance imaging or computed tomography and the number of seizures before treatment correlated with poor initial 6-month response to AEDs.

Significance

The initial 6-month response to AEDs is a valuable predictor of long-term response in patients with newly diagnosed epilepsy. The number of seizures before treatment and brain-imaging abnormalities are two prognostic predictors of initial 6-month seizure freedom.

Brainstem dysfunction in patients with late-onset Lennox-Gastaut syndrome: Voxel-based morphometry and tract-based spatial statistics study

Background:

There have been a few reports of patients who developed Lennox–Gastaut syndrome (LGS) in the second decades of their life.

Objectives:

The aim of this study was to investigate electroclinical presentation in patients with late-onset LGS. In addition, we evaluated structural abnormalities of the brain, which may give some clue about the common pathogenic pathway in LGS.

Materials and Methods:

We enrolled the patients with late-onset LGS. We collected electroclinical characteristics of the patients and evaluated structural abnormalities using voxel-based morphometry (VBM) and tract-based spatial statistics (TBSS) analysis.

Results:

The three subjects were diagnosed with late-onset LGS. The patients have no mental retardation and normal background activities on electroencephalography (EEG), and they had generalized paroxysmal fast activities on EEG, especially during sleep. The TBSS analysis revealed that fractional anisotropy values in the patients were significantly reduced in the white matter of brainstem compared with normal controls. However, VBM analysis did not show any significant difference between the patients and normal controls.

Conclusions:

Patients with late-onset LGS have different clinical and EEG characteristics from those with early-onset LGS. In addition, we demonstrated that brainstem dysfunction might contribute to the pathogenesis of late-onset LGS.

Pharmacotherapy in pediatric epilepsy: from trial and error to rational drug and dose selection - a long way to go

INTRODUCTION

Whereas ongoing efforts in epilepsy research focus on the underlying disease processes, the lack of a physiologically based rationale for drug and dose selection contributes to inadequate treatment response in children. In fact, limited information on the interindividual variation in pharmacokinetics and pharmacodynamics of anti-epileptic drugs (AEDs) in children drive prescription practice, which relies primarily on dose regimens according to a mg/kg basis. Such practice has evolved despite advancements in pediatric pharmacology showing that growth and maturation processes do not correlate linearly with changes in body size.

AREAS COVERED

In this review we aim to provide 1) a comprehensive overview of the sources of variability in the response to AEDs, 2) insight into novel methodologies to characterise such variation and 3) recommendations for treatment personalisation.

EXPERT OPINION

The use of pharmacokinetic-pharmacodynamic principles in clinical practice is hindered by the lack of biomarkers and by practical constraints in the evaluation of polytherapy. The identification of biomarkers and their validation as tools for drug development and therapeutics will require some time. Meanwhile, one should not miss the opportunity to integrate the available pharmacokinetic data with modeling and simulation concepts to prevent further delays in the development of personalised treatments for pediatric patients.

Factors predictive of late remission in a cohort of Chinese patients with newly diagnosed epilepsy

PURPOSE

Limited data have focused on predictive factors of late remission in patients with newly diagnosed epilepsy. We are aiming to identify prognostic predictors of late remission in a prospective cohort of Chinese patients.

METHODS

Patients with newly diagnosed epilepsy were included from 2009 to September 2012 at a tertiary hospital, with follow-up of at least two years. Early remission was defined by seizure free either immediately or within six months of treatment initiation, late remission was defined by seizure free achieved after more than six months. All analyses were performed with SPSS 13.0 software.

RESULTS

A total of 223 patients were included, and followed for an average of 43 months. 115 patients (51.6%) achieved early remission and 39 patients (17.5%) achieved late remission. Multivariable logistic regression analysis demonstrated more than 3 seizures prior to treatment (OR=3.12, 95% CI 1.39-7.04, p=0.006) and multiple seizure types (OR=2.49, 95% CI 1.02-6.11, p=0.046) may predict late remission. However, nonadherence was not significantly associated with late remission.

CONCLUSION

Patients with a high frequency of seizures prior to treatment or multiple seizure types may achieve late remission. Particular consideration should be given to these patients.

Serum levels of zinc and copper in epileptic children during long-term therapy with anticonvulsants

Objective:

To evaluate the serum levels of zinc and copper in epileptic children during the long-term treatment of anticonvulsant drugs and correlate this with healthy subjects.

Methods:

A hospital-based group matched case-control study was conducted in the Department of Pediatrics, Faculty of Medicine, Zagazig University, Zagazig, Egypt between November 2013 and October 2014. Ninety patients aged 7.1±3.6 years were diagnosed with epilepsy by a neurologist. The control group was selected from healthy individuals and matched to the case group. Serum zinc and copper were measured by the calorimetric method using a colorimetric method kit.

Results:

The mean zinc level was 60.1±22.6 ug/dl in the cases, and 102.1±18 ug/dl in the controls (p<0.001). The mean copper level was 180.1±32.4 ug/dl in cases compared with 114.5±18.5 ug/dl in controls (p<0.001).

Conclusion:

Serum zinc levels in epileptic children under drug treatment are lower compared with healthy children. Also, serum copper levels in these patients are significantly higher than in healthy people. No significant difference in the levels of serum copper and zinc was observed in using one drug or multiple drugs in the treatment of epileptic patients.

Brain morphology in juvenile myoclonic epilepsy and absence seizures

OBJECTIVE

We evaluated the differences in brain morphology among patients with juvenile myoclonic epilepsy according to the occurrence of absence seizures.

MATERIALS AND METHODS

Twenty-one juvenile myoclonic epilepsy patients with (n = 6) and without (n = 15) absence seizures were enrolled. We analyzed whole-brain T1-weighted magnetic resonance imaging using FreeSurfer 5.1. Measures of cortical morphology, such as thickness, surface area, volume, and curvature, and the volumes of subcortical structures, the cerebellum, and cerebrum, were compared between the groups. Moreover, we quantified correlations between clinical variables and each measures of abnormal brain morphology.

RESULTS

Compared to normal controls, patients without absence seizures demonstrated thinning of the cortical thickness in the right hemisphere, including the post-central, lingual, orbitofrontal, and lateral occipital cortex. Compared to normal controls, patients with absence seizures had more widespread thinning of the cortical thickness, including the right post-central, lingual, orbitofrontal, and lateral occipital cortexes as well as the right inferior temporal cortex. Additionally, the volume of cerebellar white matter in patients without absence seizures was significantly smaller than that in normal controls. Patients with absence seizures had a much smaller cerebellar white matter volume than normal controls or patients without absence seizures. Moreover, there was significantly positive correlation between the age of seizure onset and the volume of cerebellar white matter in patients with juvenile myoclonic epilepsy.

CONCLUSIONS

We demonstrated that there were significant brain morphology differences in patients with juvenile myoclonic epilepsy according to the presence of absence seizures. These findings support the hypothesis that juvenile myoclonic epilepsy may be a heterogeneous syndrome.