Machine learning as a diagnostic decision aid for patients with transient loss of consciousness.

BACKGROUND: Transient loss of consciousness (TLOC) is a common reason for presentation to primary/emergency care; over 90% are because of epilepsy, syncope, or psychogenic non-epileptic seizures (PNES). Misdiagnoses are common, and there are currently no validated decision rules to aid diagnosis and management. We seek to explore the utility of machine-learning techniques to develop a short diagnostic instrument by extracting features with optimal discriminatory values from responses to detailed questionnaires about TLOC manifestations and comorbidities (86 questions to patients, 31 to TLOC witnesses). METHODS: Multi-center retrospective self- and witness-report questionnaire study in secondary care settings. Feature selection was performed by an iterative algorithm based on random forest analysis. Data were randomly divided in a 2:1 ratio into training and validation sets (163:86 for all data; 208:92 for analysis excluding witness reports). RESULTS: Three hundred patients with proven diagnoses (100 each: epilepsy, syncope and PNES) were recruited from epilepsy and syncope services. Two hundred forty-nine completed patient and witness questionnaires: 86 epilepsy (64 female), 84 PNES (61 female), and 79 syncope (59 female). Responses to 36 questions optimally predicted diagnoses. A classifier trained on these features classified 74/86 (86.0% [95% confidence interval 76.9%-92.6%]) of patients correctly in validation (100 [86.7%-100%] syncope, 85.7 [67.3%-96.0%] epilepsy, 75.0 [56.6%-88.5%] PNES). Excluding witness reports, 34 features provided optimal prediction (classifier accuracy of 72/92 [78.3 (68.4%-86.2%)] in validation, 83.8 [68.0%-93.8%] syncope, 81.5 [61.9%-93.7%] epilepsy, 67.9 [47.7%-84.1%] PNES). CONCLUSIONS: A tool based on patient symptoms/comorbidities and witness reports separates well between syncope and other common causes of TLOC. It can help to differentiate epilepsy and PNES. Validated decision rules may improve diagnostic processes and reduce misdiagnosis rates. CLASSIFICATION OF EVIDENCE: This study provides Class III evidence that for patients with TLOC, patient and witness questionnaires discriminate between syncope, epilepsy and PNES.

Value of witness observations in the differential diagnosis of transient loss of consciousness.

OBJECTIVE: This retrospective study explores to what extent additional information from event witnesses provided using the novel 31-item Paroxysmal Event Observer (PEO) Questionnaire improves the differentiation among epilepsy, syncope, and psychogenic nonepileptic seizures (PNES) achievable with information provided by patients alone. METHODS: Patients with transient loss of consciousness caused by proven epilepsy (n = 86), syncope (n = 79), or PNES (n = 84) attending specialist neurology/syncope services in the United Kingdom and event observers provided Paroxysmal Event Profile (PEP), PEO, and personal information (PI) (e.g., sex, age, medical history) data. PEO data were subjected to exploratory factor analysis (EFA) followed by confirmatory factor analysis (CFA). PEO, PEP, and PI data were used separately and in combination to differentiate diagnoses by pairwise and multinomial logistic regressions. Predicted diagnoses were compared with gold standard medical diagnoses. RESULTS: EFA/CFA identified a 4-factor structure of the PEO based on 26/31 questionnaire items with loadings ≥0.4. Observer-reported factors alone differentiated better between syncope and epilepsy than patient-reported factors (accuracy: 96% vs 85%, p = 0.0004). Observer-reported data improved accuracy over differentiation based on patient-reported data alone from 90% to 100% between syncope and epilepsy (p = 0.005), 76% to 83% between epilepsy and PNES (p = 0.006), and 93% to 95% between syncope and PNES (p = 0.098). CONCLUSIONS: Information from observers can make an important contribution to the differentiation of epilepsy from syncope or PNES but adds less to that of syncope from PNES.

Phenytoin-related ataxia in patients with epilepsy: clinical and radiological characteristics.

PURPOSE: Phenytoin is an effective anticonvulsant for focal epilepsy. Its use can be associated with long-term adverse effects including cerebellar ataxia. Whilst phenytoin is toxic to Purkinje cells in vitro; the clinical and radiological phenotype and mechanism of cerebellar degeneration in vivo remain unclear. We describe the prevalence, clinical and radiological characteristics of phenytoin-related ataxia. METHODS: Patients with epilepsy receiving treatment with phenytoin were recruited from the Epilepsy clinics at Royal Hallamshire Hospital, Sheffield, UK. Neurological examination was performed on all patients after recruitment. Patients were categorised into those with and without ataxia. We determined the severity of ataxia clinically (SARA score) and the pattern of cerebellar involvement by neuroimaging (MRI volumetry and MR spectroscopy). RESULTS: Forty-seven patients were recruited. Median duration of epilepsy was 24 years, median duration of phenytoin treatment was 15 years and current median phenytoin daily dose was 325 mg. Fifty-five percent of patients complained of poor balance. Clinical evidence of ataxia was seen in 40% patients. Gait, stance and heel-shin slide were the predominant features of cerebellar dysfunction. MRI demonstrated structural, volumetric and functional deficits of the cerebellum. Only one patient with ataxia had phenytoin levels above the normal range. CONCLUSIONS: Cerebellar ataxia is present in 40% of patients with epilepsy and chronic exposure to phenytoin. Patients on long-term phenytoin have reduced cerebellar volume even if they have no clinical evidence of ataxia. Evidence of structural deficits on imaging suggests a predilection for vermian involvement.

Value of patient-reported symptoms in the diagnosis of transient loss of consciousness.

OBJECTIVE: Epileptic seizures, syncope, and psychogenic nonepileptic seizures (PNES) account for over 90% of presentations with transient loss of consciousness (TLOC). The patient's history is crucial for the diagnosis, but the diagnostic value of individual semiologic features is limited. This study explores the diagnostic potential of a comprehensive questionnaire focusing on TLOC-associated symptoms. METHODS: A total of 386 patients with proven epilepsy, 308 patients with proven PNES, and 371 patients with proven syncope were approached by post to recruit 100 patients in each diagnostic group. Symptoms were self-reported on an 86-item questionnaire (the Paroxysmal Event Profile [PEP]) using a 5-point Likert scale (always to never). Data were subjected to exploratory factor analysis (EFA) followed by confirmatory factor analysis (CFA). Factors were used to differentiate between diagnoses by pairwise and multinomial regression. RESULTS: Patients with PNES reported more and more frequent TLOC-associated symptoms than those with epilepsy or syncope (p < 0.001). EFA/CFA identified a 5-factor structure based on 74/86 questionnaire items with loadings ≥0.4. Pairwise logistic regression analysis correctly classified 91% of patients with epilepsy vs those with syncope, 94% of those with PNES vs those with syncope, and 77% of those with epilepsy vs those with PNES. Multinomial logistic regression analysis yielded a similar pattern. CONCLUSIONS: Clusters of self-reported TLOC symptoms can be used to direct patients to appropriate investigation and treatment pathways for syncope on the one hand and seizures on the other, although additional information is required for a reliable distinction, especially between epilepsy and PNES.

Differential diagnosis of seizure disorders: a conversation analytic approach.

"Taking the history" remains the most important diagnostic tool in the assessment of people who have lost consciousness. The distinction of epileptic and non-epileptic seizures (NES) is particularly difficult and relevant. Whereas epileptic seizures can usually be controlled with antiepileptic drugs, NES are considered an expression of psychosocial distress and may improve with psychotherapy. The recording of typical seizures with simultaneous video and electroencephalography (EEG) can produce almost complete certainty about the diagnosis but access to video-EEG is limited, the test is very expensive and often video-EEG fails to capture typical seizures. A German research group used conversation analysis (CA) to examine patients' descriptions of seizures to their doctors. They found that certain linguistic and interactional features clustered together and that these clusters were usually concordant with particular medical diagnoses. This study was undertaken to establish whether the observations made in German-speaking patients could be replicated in English speakers presenting to a less specialised epilepsy service. The findings presented here are based on transcripts of interviews with 11 patients admitted to a neurology ward in England because their consultant felt unable to make a clear diagnosis clinically. Transcripts were only analysed if the diagnosis of epilepsy or NES had been proven with video-EEG. The medical diagnosis was only disclosed to the linguist once a linguistic hypothesis of the diagnosis had been formulated to ensure that the linguist's decision would not be influenced by factors not contained in the 30-min-interview between doctor and patient. The linguist predicted the correct diagnosis in all cases.

The SANAD study of effectiveness of carbamazepine, gabapentin, lamotrigine, oxcarbazepine, or topiramate for treatment of partial epilepsy: an unblinded randomised controlled trial.

BACKGROUND: Carbamazepine is widely accepted as a drug of first choice for patients with partial onset seizures. Several newer drugs possess efficacy against these seizure types but previous randomised controlled trials have failed to inform a choice between these drugs. We aimed to assess efficacy with regards to longer-term outcomes, quality of life, and health economic outcomes. METHODS: SANAD was an unblinded randomised controlled trial in hospital-based outpatient clinics in the UK. Arm A recruited 1721 patients for whom carbamazepine was deemed to be standard treatment, and they were randomly assigned to receive carbamazepine, gabapentin, lamotrigine, oxcarbazepine, or topiramate. Primary outcomes were time to treatment failure, and time to 12-months remission, and assessment was by both intention to treat and per protocol. This study is registered as an International Standard Randomised Controlled Trial, number ISRCTN38354748. FINDINGS: For time to treatment failure, lamotrigine was significantly better than carbamazepine (hazard ratio [HR] 0.78 [95% CI 0.63-0.97]), gabapentin (0.65 [0.52-0.80]), and topiramate (0.64 [0.52-0.79]), and had a non-significant advantage compared with oxcarbazepine (1.15 [0.86-1.54]). For time to 12-month remission carbamazepine was significantly better than gabapentin (0.75 [0.63-0.90]), and estimates suggest a non-significant advantage for carbamazepine against lamotrigine (0.91 [0.77-1.09]), topiramate (0.86 [0.72-1.03]), and oxcarbazepine (0.92 [0.73-1.18]). In a per-protocol analysis, at 2 and 4 years the difference (95% CI) in the proportion achieving a 12-month remission (lamotrigine-carbamazepine) is 0 (-8 to 7) and 5 (-3 to 12), suggesting non-inferiority of lamotrigine compared with carbamazepine. INTERPRETATION: Lamotrigine is clinically better than carbamazepine, the standard drug treatment, for time to treatment failure outcomes and is therefore a cost-effective alternative for patients diagnosed with partial onset seizures.

The SANAD study of effectiveness of valproate, lamotrigine, or topiramate for generalised and unclassifiable epilepsy: an unblinded randomised controlled trial.

BACKGROUND: Valproate is widely accepted as a drug of first choice for patients with generalised onset seizures, and its broad spectrum of efficacy means it is recommended for patients with seizures that are difficult to classify. Lamotrigine and topiramate are also thought to possess broad spectrum activity. The SANAD study aimed to compare the longer-term effects of these drugs in patients with generalised onset seizures or seizures that are difficult to classify. METHODS: SANAD was an unblinded randomised controlled trial in hospital-based outpatient clinics in the UK. Arm B of the study recruited 716 patients for whom valproate was considered to be standard treatment. Patients were randomly assigned to valproate, lamotrigine, or topiramate between Jan 12, 1999, and Aug 31, 2004, and follow-up data were obtained up to Jan 13, 2006. Primary outcomes were time to treatment failure, and time to 1-year remission, and analysis was by both intention to treat and per protocol. This study is registered as an International Standard Randomised Controlled Trial, number ISRCTN38354748. FINDINGS: For time to treatment failure, valproate was significantly better than topiramate (hazard ratio 1.57 [95% CI 1.19-2.08]), but there was no significant difference between valproate and lamotrigine (1.25 [0.94-1.68]). For patients with an idiopathic generalised epilepsy, valproate was significantly better than both lamotrigine (1.55 [1.07-2.24] and topiramate (1.89 [1.32-2.70]). For time to 12-month remission valproate was significantly better than lamotrigine overall (0.76 [0.62-0.94]), and for the subgroup with an idiopathic generalised epilepsy 0.68 (0.53-0.89). But there was no significant difference between valproate and topiramate in either the analysis overall or for the subgroup with an idiopathic generalised epilepsy. INTERPRETATION: Valproate is better tolerated than topiramate and more efficacious than lamotrigine, and should remain the drug of first choice for many patients with generalised and unclassified epilepsies. However, because of known potential adverse effects of valproate during pregnancy, the benefits for seizure control in women of childbearing years should be considered.