Abstract
Background and Objectives
Pathogenic variants in the neuronal sodium channel α1 subunit gene (SCN1A) are the most frequent monogenic cause of epilepsy. Phenotypes comprise a wide clinical spectrum, including severe childhood epilepsy; Dravet syndrome, characterized by drug-resistant seizures, intellectual disability, and high mortality; and the milder genetic epilepsy with febrile seizures plus (GEFS+), characterized by normal cognition. Early recognition of a child's risk for developing Dravet syndrome vs GEFS+ is key for implementing disease-modifying therapies when available before cognitive impairment emerges. Our objective was to develop and validate a prediction model using clinical and genetic biomarkers for early diagnosis of SCN1A-related epilepsies.
Methods
We performed a retrospective multicenter cohort study comprising data from patients with SCN1A-positive Dravet syndrome and patients with GEFS+ consecutively referred for genetic testing (March 2001–June 2020) including age at seizure onset and a newly developed SCN1A genetic score. A training cohort was used to develop multiple prediction models that were validated using 2 independent blinded cohorts. Primary outcome was the discriminative accuracy of the model predicting Dravet syndrome vs other GEFS+ phenotypes.
Results
A total of 1,018 participants were included. The frequency of Dravet syndrome was 616/743 (83%) in the training cohort, 147/203 (72%) in validation cohort 1, and 60/72 (83%) in validation cohort 2. A high SCN1A genetic score (133.4 [SD 78.5] vs 52.0 [SD 57.5]; p < 0.001) and young age at onset (6.0 [SD 3.0] vs 14.8 [SD 11.8] months; p < 0.001) were each associated with Dravet syndrome vs GEFS+. A combined SCN1A genetic score and seizure onset model separated Dravet syndrome from GEFS+ more effectively (area under the curve [AUC] 0.89 [95% CI 0.86–0.92]) and outperformed all other models (AUC 0.79–0.85; p < 0.001). Model performance was replicated in both validation cohorts 1 (AUC 0.94 [95% CI 0.91–0.97]) and 2 (AUC 0.92 [95% CI 0.82–1.00]).
Discussion
The prediction model allows objective estimation at disease onset whether a child will develop Dravet syndrome vs GEFS+, assisting clinicians with prognostic counseling and decisions on early institution of precision therapies (http://scn1a-prediction-model.broadinstitute.org/).
Classification of Evidence
This study provides Class II evidence that a combined SCN1A genetic score and seizure onset model distinguishes Dravet syndrome from other GEFS+ phenotypes.
Epilepsy affects an estimated 50–65 million individuals worldwide.1 The majority of epilepsies are thought to be genetic in origin due to single gene disorders or complex inheritance.2 Pathogenic variants in the sodium voltage-gated channel alpha subunit 1, SCN1A (OMIM 182389), are the most common monogenic cause of epilepsy, affecting 1 in 12,200 live births.3 Clinical presentation is highly variable and includes the severe infantile-onset Dravet syndrome as well as phenotypes within the mild genetic epilepsy with febrile seizures plus (GEFS+) spectrum.4 Whereas Dravet syndrome leads to a significant developmental and epileptic encephalopathy with difficult-to-treat seizures and severe intellectual disability,5,6 individuals with other GEFS+ phenotypes live independent lives with normal cognition and very mild epilepsy.7 Distinction of these 2 conditions on clinical grounds alone is challenging in the first 2 years of life because the encephalopathy associated with Dravet syndrome is insidious and early development is within normal limits. Genotype–phenotype correlations are not well-established and when a pathogenic SCN1A variant is found, it is not possible for clinicians to accurately predict whether a child will develop Dravet syndrome or other GEFS+ phenotypes.8 Both disorders may present with recurrent, often prolonged febrile seizures in an otherwise apparently normal infant. The full Dravet syndrome phenotype only emerges in the second and third year of life and is associated with high epilepsy mortality in early childhood (15.84/1,000 person-years), due to status epilepticus and sudden unexpected death in epilepsy (SUDEP).5,6,9
Accurate prediction of whether a young child with a pathogenic SCN1A variant will develop the severe epilepsy Dravet syndrome or milder GEFS+ phenotypes is important for counseling, patient management, and treatment planning. Clinicians often miss the opportunity for early intervention as they wait for symptoms such as developmental delay to emerge before making a diagnosis of Dravet syndrome. Treatment strategies have focused on achieving better seizure control with stiripentol, clobazam, and sodium valproate, as well as the use of cannabidiol and fenfluramine.10-13 New gene-specific, disease-modifying therapies have been shown to significantly reduce seizure burden and mortality in Dravet rodent models when given early and the first-in-human trial of gene-based therapy in Dravet syndrome recently began.14 Prompt diagnosis is important to enable timely administration of new treatments in Dravet syndrome and to avoid unnecessary and possibly harmful treatment in other GEFS+ phenotypes.
The crucial aspect in deciding the best treatment approach and timing is the infant's odds of developing Dravet syndrome vs other GEFS+ phenotypes. To date, only 2 studies have attempted to predict Dravet syndrome vs GEFS+ based on clinical and genetic data.15,16 These studies showed a moderate association between single outcome predictors such as early seizure onset or truncating variants being linked to a more severe phenotype, but there are no validated actionable prediction models available to guide clinical decision-making.15,16
The challenge of outcome prediction is not unique to genetic epilepsies, and risk prediction models are routinely used to aid decision-making in cardiovascular disease and cancer.17,18
Using a large SCN1A patient cohort, we hypothesized that combining clinical and genetic data will allow us to develop a statistical model for the early prediction of SCN1A-related epilepsy phenotypes.
Methods
Study Design, Participants, and Clinical Assessments
We conducted a multicenter retrospective cohort study to develop and validate a statistical model combining age at seizure onset (febrile or afebrile, whichever occurred first) and the SCN1A genetic score in predicting Dravet syndrome vs other GEFS+ phenotypes. Results are reported using the Enhancing the Quality and Transparency of Health Research (EQUATOR) network Standards for Reporting of Diagnostic Accuracy (STARD) guidelines for diagnostic accuracy studies.19 We developed the clinical–genetic prediction model from a retrospective cohort of 1,018 patients from 7 countries: United Kingdom (n = 276), France (n = 201), Italy (n = 126), Netherlands (n = 109), Denmark (n = 31), Australia (n = 203), and Belgium (n = 72). All cases were identified from consecutive referrals for genetic testing in different centers in the respective countries or for research referral from March 2001 to June 2020. We included patients with Dravet syndrome and patients with GEFS+ carrying pathogenic SCN1A variants from the following sites: The Royal Hospital for Children (Glasgow, UK),4,20 The Hôpital Necker-Enfants Malades (Paris, France),21 The A Meyer Children's Hospital (Florence, Italy),15 The University Medical Center Utrecht and Radboud University Nijmegen Medical Center (the Netherlands),16 The Danish Epilepsy Centre Filadelfia (Dianalund, Denmark),22,23 The University Hospital Antwerp (Belgium),24 The Austin Health and Royal Children's Hospital (Melbourne, Australia), and unpublished cases (eMethods and eTable 1, links.lww.com/WNL/B785).
Phenotypes were classified by experts in the management of Dravet syndrome and GEFS+ according to the following criteria: Dravet syndrome was defined as generalized or hemiclonic seizures frequently triggered by fever and often prolonged, typically followed by other seizure types including myoclonic, focal impaired awareness, and absence seizures, and normal cognitive and psychomotor development prior to seizure onset with subsequent slowing including plateauing or regression of skills in the second year of life. Patients were given a diagnosis of other GEFS+ phenotypes if they had presentations consistent with the febrile seizures plus spectrum, with or without a relevant family history and normal intellect,7 which in the context of this study excludes Dravet syndrome. In most cases diagnoses were made at age >24 months; however, a number of patients with Dravet syndrome were diagnosed at an earlier age if the phenotype was highly suggestive, including the plateauing or regression of skills.
We developed the prediction model using a training cohort, including patients from the United Kingdom, France, Italy, Netherlands, and Denmark (n = 743). We then tested the prediction model in 2 blinded validation cohorts from Australia (validation cohort 1, n = 203) and Belgium (validation cohort 2, n = 72). Because our model is based on age at onset and genetic data, we only included patients who had these data available.
Blinding of Validation Cohorts
Whereas clinical information (Dravet syndrome vs GEFS+) was available to the assessors for the training cohort, data for the 2 validation cohorts were supplied without disclosing the phenotype. Details on whether a patient had Dravet syndrome or other GEFS+ phenotypes was only made available after the prediction analysis had been completed.
Standard Protocol Approvals, Registrations, and Patient Consents
Retrospective review of anonymized clinical referral data and variant findings was approved by the relevant institutional review boards (West of Scotland Research Ethics Committee, reference number 16/WS/0203).
Genetic Analysis and SCN1A Genetic Score
We included missense and protein truncating variants (PTVs). PTVs were composed of premature stop codons, frameshifts leading to stop codons, large deletions, and whole gene deletions. Variants whose effect cannot be predicted based on position, amino acid exchange, or truncation were excluded from the study. This applied to splice variants, in-frame small insertions/deletions, and synonymous variants. Details on molecular analysis for each center are provided in the eMethods (links.lww.com/WNL/B785). For each pathogenic variant, we generated a SCN1A-specific genetic score by combining paralog conservation of the mutated amino acid position25 with the physicochemical properties (Grantham score26) of the observed substitution. Paralog conservation accounts for the degree of amino acid conservation across a single gene family alignment. In the case of the voltage-gated sodium channels gene family, 10 genes were aligned to calculate the paralog score: SCN1A–SCN11A. The score ranges from amino acid positions with −2.06 (least conserved) to 1.23 (most conserved) and is independent of the exchange observed. Paralog conserved sites are particularly enriched for pathogenic variants in voltage-gated sodium channels and high Grantham scores reflect radical amino acid substitutions that are more likely to be deleterious.25,26 The SCN1A genetic score ranged from 0 (similar) to 207 (dissimilar) and is the result of the paralog score observed in the position multiplied by the Grantham score associated with the amino acid exchange. PTVs are assumed deleterious for protein function and were assigned the maximum SCN1A genetic score observed (207). We compared performance of the SCN1A genetic score with established variant interpretation tools such as CADD (Combined Annotation Dependent Depletion)27 and REVEL (Rare Exome Variant Ensemble Learner).28
Statistical Analysis and Prediction Model Development
Our primary research question was as follows: What is the discriminative accuracy of a statistical model combining age at seizure onset and the SCN1A genetic score in predicting Dravet syndrome vs other GEFS+ phenotypes? This study provides Class II evidence relating to this research question.
Model development and validation was performed according to Transparent Reporting for Individual Prognosis or Diagnosis (TRIPOD) guidance of multivariable prediction models.29 We applied a supervised machine learning approach and trained a generalized linear model using the SCN1A genetic score and the age at seizure onset in months (referred to as the Index SCN1A score and Onset model) as predictors of Dravet syndrome and GEFS+ (eMethods, links.lww.com/WNL/B785). The age at seizure onset was identified as the earliest clinical feature that could easily and reliably be assessed in the first year of life when most other clinical signs have not emerged and has been shown to be a valuable prognostic factor in earlier studies.15,16
To compare our model, we constructed 3 additional models: (1) age at seizure onset Onset-only model, (2) CADD and Onset model, and (3) REVEL and Onset model, following the same procedure. We compared our model against a 6-months seizure onset threshold model proposed previously,15 which served as reference standard as it was the only predictive model available prior to our study. For all models tested (including index and reference standard models), we used a 50% cutoff threshold to positively predict a case of Dravet syndrome. Patients with predictions below 50% were assigned a GEFS+ status. We calibrated and compared the models using the receiver operating characteristic curve, calibration curves, and the index of prediction accuracy (IPA).30 Area under the curve (AUC) and IPA 95% CIs were generated with 1,000 bootstrap sets during cross-validation. Sensitivities, specificities, positive predictive values (PPVs), negative predictive values (NPVs), and accuracies alongside their 95% CIs were calculated following established guidelines.31 All patients with ages at seizure onset, genetic variants, and their corresponding genetic score are detailed in eTable 1 (links.lww.com/WNL/B785)
Data Availability
Anonymized data not published within this article will be available from the lead author by email on reasonable request.
Results
Of an original 862 patients, 119 (14%) carried variants whose effect cannot be predicted and were excluded. The training cohort included 743 patients, of whom 616 (83%) had Dravet syndrome and 127 other GEFS+ phenotypes (17%). The frequency of Dravet syndrome in validation cohort 1 was 147/203 (72%) with 56 (28%) patients with GEFS+ and in validation cohort 2 60/72 (83%) with 12 (17%) patients with GEFS+. The training cohort had 447 missense variant (60%) and 296 PTV (40%) carriers, validation cohort 1 had 134 missense variant (66%) and 69 PTV (34%) carriers, and validation cohort 2 had 44 missense variant (61%) and 28 PTV (39%) carriers.
A summary of the study outline is shown in Figure 1. Among the training cohort, a younger age at seizure onset or a higher SCN1A genetic score were each associated with a diagnosis of Dravet syndrome (Figure 2, A and B). Despite the significantly earlier seizure onset in the Dravet syndrome (mean [SD] age, 6.04 [3.0] months) vs GEFS+ group (14.82 [11.8] months; p < 0.001) and the higher SCN1A genetic score in the Dravet syndrome (133.43 [78.53]) vs GEFS+ group (52.90 [57.58]; p < 0.001), there was considerable overlap between the disorders (Figure 2, A and B).
Figure 1. Study Overview.
Study workflow. Genetic data (SCN1A genetic score) and clinical data (age at seizure onset in months) from 743 patients (training cohort) were introduced to a supervised machine learning approach to produce a prediction model. We tested the prediction model with 2 independent blinded validation cohorts (n = 275). GEFS+ = genetic epilepsy with febrile seizures plus.
Figure 2. Training Cohort Data.
(A) Density plot showing the distribution of the age at seizure onset in training cohort patients with Dravet syndrome (purple area) and genetic epilepsy with febrile seizures plus (GEFS+) (gray area). (B) Density plot showing the distribution of the SCN1A genetic score in training cohort patients with Dravet syndrome (purple area) and GEFS+ (gray area). Statistical difference between the observed means was evaluated with the Wilcoxon test.
Using the training cohort, we generated 4 different models to discriminate between Dravet syndrome and other GEFS+ phenotypes. With an AUC of 0.89 (95% CI 0.86–0.92) and an IPA of 38.7%, the clinical–genetic SCN1A score and Onset model outperformed the prediction model based solely on the age at seizure onset (Onset-only: AUC 0.84 [95% CI 0.80–0.88]; IPA = 33.6%; p < 0.001; Figures 3 and 4). The SCN1A score and Onset model equally outperformed models based on 2 additional pathogenicity scores, namely CADD (CADD and Onset: AUC 0.85 [95% CI 0.82–0.89]; IPA = 31.2%) and REVEL (REVEL and Onset: AUC 0.84 [95% CI 0.80–0.88]; IPA = 31.6%). In addition, our SCN1A score and Onset model outperformed the 6-months seizure onset threshold model proposed previously15 (Figures 3 and 4; AUC 0.79). Dominance analysis showed that age at seizure onset was 2.06 times more important than the SCN1A genetic score to the overall model (eFigure 1 and eTable 2, links.lww.com/WNL/B785). Model performance was similar when focusing only on index cases (eTable 3, links.lww.com/WNL/B785). Next, we tested the performance of the SCN1A score and Onset model in 2 independent blinded validation cohorts of SCN1A epilepsy. Model performance achieved an AUC of 0.94 (95% CI 0.91–0.97) in validation cohort 1 and an AUC of 0.92 (95% CI 0.82–1.00) in validation cohort 2.
Figure 3. Training Cohort Model Performance: ROC Curve Analysis.
Receiver operating characteristic (ROC) curves showing the relationship between the observed sensitivity and specificity for different models using genetic scores and seizure age at onset: SCN1A score and Onset (blue line, n = 743), onset only (orange line, n = 743), CADD (Combined Annotation Dependent Depletion) score and Onset (green line), and REVEL (Rare Exome Variant Ensemble Learner) score and Onset (purple line). Because CADD and REVEL scores are not available for all variants contained in the training cohort, the CADD and Onset and REVEL and Onset models were built with a subset of 651 and 438 training cohort patients, respectively (eTable 1, links.lww.com/WNL/B785). The 6-months seizure onset threshold model (gray line) proposed previously15 is shown for comparison. Area under the curve (AUC) values and 95% CIs are shown at the bottom right corner of the plot.
Figure 4. Calibration Curves per Model.
Training cohort model performance. Individual calibration curves showing the relationship between the predicted risk and the observed frequency for each of the tested models. Index of prediction accuracy (IPA) is shown below each model. Color code: SCN1A score and Onset (blue line), Onset-only (orange line), CADD (Combined Annotation Dependent Depletion) score and Onset (green line), and REVEL (Rare Exome Variant Ensemble Learner) score and Onset (purple line).
In the model evaluation, patients with higher probability values are predicted to have Dravet syndrome and patients with lower values are predicted to have other GEFS+ phenotypes (Figures 5 and 6). Table 1 illustrates the model performance detailing PPVs and NPVs as well as sensitivities and specificities observed at different thresholds for both validation cohorts individually and combined (n = 275; Dravet syndrome = 207, GEFS = 68).
Figure 5. Validation Cohort 1 and 2 Prediction Results.
Patients with probability values above 50% were predicted to have Dravet syndrome and patients with values below 50% were predicted to have genetic epilepsy with febrile seizures plus (GEFS+). (A, B) Predicted values across validation cohorts 1 and 2 are shown, respectively. Each bar corresponds to a patient. The height of each bar represents the probability of that patient developing Dravet syndrome. Patients with true Dravet syndrome are shown in purple; patients with true GEFS+ are shown in gray. Dotted horizontal line denotes a 50% threshold with values above 50% predicting Dravet syndrome and values below 50% predicting GEFS+. Area under the curve (AUC) and index of prediction accuracy (IPA) 95% CIs are given.
Figure 6. Validation Cohort 1 and 2 Phenotype Distribution.
Phenotype distribution with density of prediction performed on validation cohorts 1 and 2, respectively. Patients with true Dravet syndrome and patients with genetic epilepsy with febrile seizures plus (GEFS+) accumulate across their corresponding model predictions (horizontal axis). Dotted vertical line denotes a 50% threshold with values above 50% predicting Dravet syndrome and values below 50% predicting GEFS+.
Table 1.
Model Performance Measures According to Different Thresholds
To explore potential performance confounders due to patient country ascertainment, we combined and randomly split the entire cohort (n = 1,018) into an additional training cohort with 70% of patients (n = 713) and a single validation cohort with 30% of patients (n = 305). In keeping with our previous results, the SCN1A score and Onset model yielded an AUC of 90.4 (95% CI 87.5–93.2) in the random training cohort and an AUC of 91.5 (95% CI 87.1–95.9) in the random validation cohort (eFigure 2, links.lww.com/WNL/B785).
We developed the prediction model into an online tool designed to evaluate any missense or PTV found in a given patient with an SCN1A pathogenic variant combined with the age at seizure onset. The SCN1A epilepsy prediction model will calculate a patient's probability (%) of developing Dravet syndrome vs other GEFS+ phenotypes in a user-friendly platform that is available online at no cost (eFigure 3, links.lww.com/WNL/B785).
Discussion
In this large, multicenter cohort study, we found that a clinical–genetic prediction model, combining the age at seizure onset with a newly developed SCN1A genetic score, allows an objective early estimation as to whether a child will develop Dravet syndrome vs other GEFS+ phenotypes. We were able to show that our prediction model outperformed any previous or alternative models and represents a validated clinical tool to aid early differentiation between Dravet syndrome and GEFS+.15,16
In the absence of internationally validated expert-based guidelines for the prediction of Dravet syndrome vs other GEFS+ phenotypes in SCN1A-positive patients, judgments about diagnosis and prognosis are challenging, particularly for nonexpert clinicians. Consider the example of a 9-month-old infant presenting with recurrent febrile seizures and a pathogenic SCN1A variant. In this case, a previous recommendation15 would predict that the risk of Dravet syndrome is moderate (51%), based on the age at onset alone. Yet additional information of a high SCN1A genetic score might increase that risk to >90%, whereas a low genetic score might reduce that risk to <10%. Consideration of the age at onset alone will not allow a confident distinction between Dravet syndrome and GEFS+ and the clinician is likely to wait until signs of developmental slowing start emerging in the second or third year of life before making a Dravet syndrome diagnosis.6 In the same way, a PTV variant might suggest a diagnosis of Dravet syndrome; however, that probability will decrease the later the age at seizure onset. Whereas model prediction is mainly determined by age at onset, these examples illustrate how both the genetic information as well as the age at onset play an important part in the outcome prediction model.
Most clinicians subjectively use patient and disease characteristics to predict outcome based on personal experience and knowledge.32 Incorrect clinical stratification results in diagnostic delay, and valuable time in the child's early development, together with subtle slowing of development, may have occurred before precision treatment is started. A validated and quantifiable approach allows Dravet syndrome risk prediction much earlier, as soon as the genetic result is available, which could be within weeks of the child having presented with recurrent seizures.20
Early treatment in Dravet syndrome is important. Studies in Scn1a mutant mice illustrate that early-life febrile seizures are associated with impaired cognition and behavior in the long term.33 Similarly, early use of contraindicated medication in the second year of life has been associated with adverse developmental outcomes in Dravet syndrome,16 emphasizing that early diagnosis is essential to establish appropriate treatment as soon as possible.10-13 Gene-specific therapy approaches are emerging as promising treatment options for Dravet syndrome when given early.14 Notably, mortality rates in Dravet syndrome due to status epilepticus and SUDEP are high, particularly affecting very young children in their first 3 years of life, emphasizing the importance of early diagnosis and treatment.9
It is a strength that the prediction model was not only based on a large and well-phenotyped international training cohort using recognized disease criteria but has been independently retested and validated in 2 equally well-characterized blinded validation cohorts, as well as in additional random samples of the entire cohort, confirming the robustness of our findings. Our approach of using clinical and genetic data combined with machine learning techniques allowed us to better predict outcome than using clinical data or widely adapted variant pathogenicity scores such as CADD or REVEL in isolation. The prediction model uses data that are easily accessible to clinicians in any young infant presenting with a pathogenic SCN1A variant. Details can be entered electronically via a free web-based application generating a probability estimate that informs clinical decision-making. These features allow ease of access across health care settings globally, increasing the model's clinical usefulness.
Weighing possible disease outcomes in an individual patient is a complex task and decision curve analysis helps to determine thresholds of sensitivity and specificity. This allows the researcher to identify the most appropriate model performance measures. Depending on the clinical situation and the harm to benefit ratio, recommendations are likely to differ according to the type of treatment considered and the adverse events reported.34 For instance, starting a young child on antiseizure medication with potentially significant side effects has to be weighed against the benefit of possible seizure freedom. If the harm of unnecessary treatment is deemed limited, then a lower model threshold may be acceptable (Table 1). However, in the case of novel interventions, such as gene-specific therapy approaches, different thresholds might apply. Given these complexities, our prediction model is not intended to replace clinical judgment, but to inform and complement clinical decision-making based on objective and quantifiable data.
There are several limitations to this study. Modeling of disease outcomes based on SCN1A variant information will be affected by a number of modifying factors, including the unknown genetic and environmental background of the individual, epigenetics, as well as transcriptional and posttranslational factors that are beyond our modeling capacity. Given that Dravet syndrome and other GEFS+ phenotypes are part of a disease spectrum, borderline presentations will be more difficult to predict, as shown in Figure 5 and Table 1. We acknowledge that our cohorts are biased towards Dravet syndrome cases and larger, more balanced datasets are needed to improve prediction accuracy. Our logistic regression model achieves an excellent to outstanding fitting (AUC 89.1) and the use of more complex modeling strategies might lead to overfitting with little opportunity to increase performance. Future larger cohorts with additional phenotypic data will allow the implementation of more complex models with increased granularity to better predict the complex heterogeneity of SCN1A-related epilepsies and will include types of variants where functional interpretation is more challenging.
The accuracy of a mutation-based prediction model is likely to be negatively influenced by specific genetic factors such as postzygotic mosaicism, which is seen in 7.5% of de novo pathogenic SCN1A variants.35 In the same way, truncating SCN1A variants that are normally predicted to be deleterious for channel function might escape nonsense-mediated mRNA decay if occurring in the terminal portion of the gene.36 As we did not observe any PTVs associated with Dravet syndrome beyond amino acid position 1930, the tool informs the user that our model does not provide a reliable prediction in such cases. These rare examples illustrate that in a minority of cases, truncating variants might not always be deleterious—an exception to the rule, which is difficult to model. Recently, very early onset cases of developmental and epileptic encephalopathy with movement disorder have been described that are not Dravet syndrome. Our tool alerts the user to consider such a phenotype for any patient presenting at less than 4 months of age.37,38 Lastly, there may be additional predictors of disease outcome not included here, such as the mode of inheritance (de novo vs familial), which might contribute to the predictive power of the model, as inherited cases are often associated with milder phenotypes. However, these data are often not available, particularly in health care settings where this screening incurs a direct cost to the patient.
The challenge of clinical decision-making is not limited to SCN1A-related epilepsies. Our approach of developing a clinical decision-support algorithm is generalizable and can be applied to many genetic disorders where genetic and clinical data are available.
Our findings suggest that routinely accessible biomarkers such as age at seizure onset combined with an SCN1A genetic score can be used to predict Dravet syndrome. Although the model can be employed at the time of diagnosis, expert clinical assessment will allow further delineation of the phenotype over time. The prediction model represents an important step towards evidence-based clinical outcome prediction, assisting clinicians with prognostic counseling and decisions on early institution of precision therapies.
Acknowledgment
The authors thank the Dravet Syndrome Foundation and Dravet Syndrome UK for their support.
Glossary
- AUC
area under the curve
- CADD
Combined Annotation Dependent Depletion
- GEFS+
genetic epilepsy with febrile seizures plus
- IPA
index of prediction accuracy
- NPV
negative predictive value
- PPV
positive predictive value
- PTV
protein truncating variant
- REVEL
Rare Exome Variant Ensemble Learner
- SUDEP
sudden unexpected death in epilepsy
Appendix. Authors
Footnotes
Class of Evidence: NPub.org/coe
Study Funding
D.L. was supported by funds from the Dravet Syndrome Foundation (grant 272016), BMBF (Treat-ION grant 01GM1907), and NIH NINDS (Channelopathy-Associated Epilepsy Research Center, 5-U54-NS108874). A.B. and S.M.Z. received a grant from Dravet Syndrome UK for the Glasgow SCN1A database (grant 16GLW00). The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Disclosure
A. Brunklaus has received honoraria for presenting at educational events, advisory boards, and consultancy work for Biocodex, Encoded Therapeutics, GW Pharma, Nutricia, Stoke Therapeutics, and Zogenix. E. Pérez-Palma has received honoraria for consultancy work for the Friends of Faces foundation and a grant from Agencia Nacional de Investigación y Desarrollo of Chile (ANID, Grant PAI77200124) and the FamilieSCN2A foundation (2020 Action Potential Grant). I. Ghanty reports no disclosures relevant to the manuscript. J. Xinge holds grants from NovoNordisk, Inc., Merck, Inc., Novartis, and Boehringer Ingelheim Pharmaceuticals, Inc., outside the submitted work. E. Brilstra reports no disclosures relevant to the manuscript. B. Ceulemans has received research funding from Brabant and Zogenix, served as a consultant for Brabant and Zogenix (Patent ZX008), and with the KU Leuven University/Antwerp University Hospital may benefit financially from a royalty arrangement that is related to this research if Zogenix is successful in marketing its product, fenfluramine. N. Chemaly has received honoraria from Nutricia and Eisai for presentations at symposia. I. de Lange and C. Depienne report no disclosures relevant to the manuscript. R. Guerrini has acted as an investigator for studies with Zogenix, Biocodex, Biomarin, UCB, Angelini, and Eisai Inc.; has been a speaker and on advisory boards for Zogenix, Biocodex, Novartis, Biomarin, GW Pharma, and Biocodex; serves/has served on the editorial boards of Epilepsia, Progress in Epileptic Disorders, Neuropediatrics, Journal of Child Neurology, Seizure, BMC Medical Genetics, Topics in Epilepsy, and Neurology®; and receives/has received research support from the Italian Ministry of Health, The European Community, The Tuscany Region, the Mariani Foundation, The Pisa Foundation, The Fund of Epilepsy, The GKT Special Trustees, The Italian Federation for Epilepsy, and The Italian Association for Epilepsy. D. Mei and R.S. Møller report no disclosures relevant to the manuscript. R. Nabbout received research funding from the European Union (Seventh Framework Programme), EJP-RD (European Joint Program for Rare Diseases), Shire, Zogenix, GW Pharma, and Eisai and consultation and lecturer fees from Eisai, Zogenix, Takeda, GW Pharma, Advicenne, Biocodex, Nutricia, Supernus, Biogen, and Novartis. B.M.R. and A.L. Schneider report no disclosures relevant to the manuscript. I.E. Scheffer has served on scientific advisory boards for UCB, Eisai, GlaxoSmithKline, BioMarin, Nutricia, Rogcon, Chiesi, Encoded Therapeutics, and Xenon Pharmaceuticals; has received speaker honoraria from GlaxoSmithKline, UCB, BioMarin, Biocodex and Eisai; has received funding for travel from UCB, Biocodex, GlaxoSmithKline, Biomarin, and Eisai; has served as an investigator for Zogenix, Zynerba, Ultragenyx, GW Pharma, UCB, Eisai, Anavex Life Sciences, Ovid Therapeutics, Epigenyx, Encoded Therapeutics, and Marinus; has consulted for Zynerba Pharmaceuticals, Atheneum Partners, Ovid Therapeutics, Care Beyond Diagnosis, Epilepsy Consortium, and UCB; may accrue future revenue on pending patent WO61/010176 (filed 2008): Therapeutic Compound; has a patent for SCN1A testing held by Bionomics Inc. and licensed to various diagnostic companies; and has a patent molecular diagnostic/theranostic target for benign familial infantile epilepsy (PRRT2) 2011904493 and 2012900190 and PCT/AU2012/001321 (TECH ID:2012-009). A. Schoonjans and J.D. Symonds report no disclosures relevant to the manuscript. S. Weckhuysen received speaker and consultancy fees from UCB, Xenon, Zogenix, Lundbeck, and Biocodex. M.W. Kattan is a consultant for Glaxo Smith Kline and Exosome Diagnostics. S.M.Z. has received honoraria for presenting at educational events, advisory boards, and consultancy work for GW Pharma, Zogenix, Biocodex, Encoded Therapeutics, Stoke Therapeutics, and Nutricia. D. Lal has received honoraria for advisory board work for Encoded Therapeutics. Go to Neurology.org/N for full disclosures.
References
- 1.Ngugi AK, Bottomley C, Kleinschmidt I, Sander JW, Newton CR. Estimation of the burden of active and life-time epilepsy: a meta-analytic approach. Epilepsia. 2010;51(5):883-890. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Thomas RH, Berkovic SF. The hidden genetics of epilepsy: a clinically important new paradigm. Nat Rev Neurol. 2014;10(5):283-292. [DOI] [PubMed] [Google Scholar]
- 3.Symonds JD, Zuberi SM, Stewart K, et al. Incidence and phenotypes of childhood-onset genetic epilepsies: a prospective population-based national cohort. Brain. 2019;142(8):2303-2318. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Zuberi SM, Brunklaus A, Birch R, Reavey E, Duncan J, Forbes GH. Genotype–phenotype associations in SCN1A-related epilepsies. Neurology. 2011;76(7):594-600. [DOI] [PubMed] [Google Scholar]
- 5.Dravet C, Oguni H. Dravet syndrome (severe myoclonic epilepsy in infancy).In: Dulac O, Lassonde M, Sarnat HB, eds. Handbook of Clinical Neurology. Elsevier; 2013:627-633. [DOI] [PubMed] [Google Scholar]
- 6.Brunklaus A, Ellis R, Reavey E, Forbes GH, Zuberi SM. Prognostic, clinical and demographic features in SCN1A mutation-positive Dravet syndrome. Brain. 2012;135(8):2329-2336. [DOI] [PubMed] [Google Scholar]
- 7.Zhang Y-H, Burgess R, Malone JP, et al. Genetic epilepsy with febrile seizures plus: refining the spectrum. Neurology. 2017;89(12):1210-1219. [DOI] [PubMed] [Google Scholar]
- 8.Brunklaus A, Schorge S, Smith AD, et al. SCN1A variants from bench to bedside: improved clinical prediction from functional characterization. Hum Mutat. 2020;41(2):363-374. [DOI] [PubMed] [Google Scholar]
- 9.Cooper MS, Mcintosh A, Crompton DE, et al. Mortality in Dravet syndrome. Epilepsy Res. 2016;128:43-47. [DOI] [PubMed] [Google Scholar]
- 10.Chiron C, Marchand MC, Tran A, et al. Stiripentol in severe myoclonic epilepsy in infancy: a randomised placebo-controlled syndrome-dedicated trial. Lancet. 2000;356(9242):1638-1642. [DOI] [PubMed] [Google Scholar]
- 11.Devinsky O, Cross JH, Laux L, et al. Trial of cannabidiol for drug-resistant seizures in the Dravet syndrome. N Engl J Med. 2017;376(21):2011-2020. [DOI] [PubMed] [Google Scholar]
- 12.Lagae L, Sullivan J, Knupp K, et al. Fenfluramine hydrochloride for the treatment of seizures in Dravet syndrome: a randomised, double-blind, placebo-controlled trial. Lancet. 2019;394(10216):2243-2254. [DOI] [PubMed] [Google Scholar]
- 13.Nabbout R, Mistry A, Zuberi S, et al. Fenfluramine for treatment-resistant seizures in patients with Dravet syndrome receiving stiripentol-inclusive regimens: a randomized clinical trial. JAMA Neurol. 2020;77(3):300-308. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Han Z, Chen C, Christiansen A, et al. Antisense oligonucleotides increase Scn1a expression and reduce seizures and SUDEP incidence in a mouse model of Dravet syndrome. Sci Transl Med. 2020:12:eaaz6100. [DOI] [PubMed] [Google Scholar]
- 15.Cetica V, Chiari S, Mei D, et al. Clinical and genetic factors predicting Dravet syndrome in infants with SCN1A mutations. Neurology. 2017;88(11):1037-1044. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.de Lange IM, Gunning B, Sonsma ACM, et al. Influence of contraindicated medication use on cognitive outcome in Dravet syndrome and age at first afebrile seizure as a clinical predictor in SCN1A-related seizure phenotypes. Epilepsia. 2018;59(6):1154-1165. [DOI] [PubMed] [Google Scholar]
- 17.Zethelius B, Berglund L, Sundström J, et al. Use of multiple biomarkers to improve the prediction of death from cardiovascular causes. N Eng J Med. 2008;358(20):2107-2116. [DOI] [PubMed] [Google Scholar]
- 18.Mallett S, Royston P, Waters R, Dutton S, Altman DG. Reporting performance of prognostic models in cancer: a review. BMC Med. 2010;8:21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Cohen JF, Korevaar DA, Altman DG, et al. STARD2015 guidelines for reporting diagnostic accuracy studies: explanation and elaboration. BMJ Open. 2016;6(11):e012799. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Brunklaus A, Du J, Steckler F, et al. Biological concepts in human sodium channel epilepsies and their relevance in clinical practice. Epilepsia. 2020;61(3):387-399. [DOI] [PubMed] [Google Scholar]
- 21.Depienne C, Trouillard O, Saint-Martin C, et al. Spectrum of SCN1A gene mutations associated with Dravet syndrome: analysis of 333 patients. J Med Genet. 2009;46(3):183-191. [DOI] [PubMed] [Google Scholar]
- 22.Møller RS, Larsen LHG, Johannesen KM, et al. Gene panel testing in epileptic encephalopathies and familial epilepsies. Mol Syndromol. 2016;7(4):210-219. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Johannesen KM, Nikanorova N, Marjanovic D, et al. Utility of genetic testing for therapeutic decision-making in adults with epilepsy. Epilepsia. 2020;61(6):1234-1239. [DOI] [PubMed] [Google Scholar]
- 24.Claes L, Ceulemans B, Audenaert D, et al. De novo SCN1A mutations are a major cause of severe myoclonic epilepsy of infancy. Hum Mutat. 2003;21(6):615-621. [DOI] [PubMed] [Google Scholar]
- 25.Lal D, May P, Perez-Palma E, et al. Gene family information facilitates variant interpretation and identification of disease-associated genes in neurodevelopmental disorders. Genome Med. 2020;12(1):28. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Grantham R. Amino acid difference formula to help explain protein evolution. Science. 1974;185(4154):862-864. [DOI] [PubMed] [Google Scholar]
- 27.Rentzsch P, Witten D, Cooper GM, Shendure J, Kircher M. CADD Predicting the deleteriousness of variants throughout the human genome. Nucleic Acids Res. 2019;47(D1):D886-D894. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Ioannidis NM, Rothstein JH, Pejaver V, et al. REVEL: an ensemble method for predicting the pathogenicity of rare missense variants. Am J Hum Genet. 2016;99(4):877-885. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350:g7594. [DOI] [PubMed] [Google Scholar]
- 30.Kattan MW, Gerds TA. The index of prediction accuracy: an intuitive measure useful for evaluating risk prediction models. Diagn Progn Res. 2018;2:7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Mercaldo ND, Lau KF, Zhou XH. Confidence intervals for predictive values with an emphasis to case–control studies. Stat Med. 2007;26(10):2170-2183. [DOI] [PubMed] [Google Scholar]
- 32.Cervellin G, Borghi L, Lippi G. Do clinicians decide relying primarily on Bayesians principles or on Gestalt perception? Some pearls and pitfalls of Gestalt perception in medicine. Intern Emerg Med. 2014;9(5):513-519. [DOI] [PubMed] [Google Scholar]
- 33.Dutton SBB, Dutt K, Papale LA, Helmers S, Goldin AL, Escayg A. Early-life febrile seizures worsen adult phenotypes in Scn1a mutants. Expl Neurol. 2017;293:159-171. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Steyerberg EW, Vickers AJ, Cook NR, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology. 2010;21(1):128-138. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Langede IM, Koudijs MJ, van’t Slot R, et al. Mosaicism of de novo pathogenic SCN1A variants in epilepsy is a frequent phenomenon that correlates with variable phenotypes. Epilepsia. 2018;59(3):690-703. [DOI] [PubMed] [Google Scholar]
- 36.Khajavi M, Inoue K, Lupski JR. Nonsense-mediated mRNA decay modulates clinical outcome of genetic disease. Eur J Hum Genet. 2006;14(10):1074-1081. [DOI] [PubMed] [Google Scholar]
- 37.Sadleir LG, Mountier EI, Gill D, et al. Not all SCN1A epileptic encephalopathies are Dravet syndrome: early profound Thr226Met phenotype. Neurology. 2017;89(10):1035-1042. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Berecki G, Bryson A, Terhag J, et al. SCN1A gain of function in early infantile encephalopathy. Ann Neurol. 2019;85(4):514-525. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Anonymized data not published within this article will be available from the lead author by email on reasonable request.