Identifying epilepsy surgery referral candidates with natural language processing in an Australian context

OBJECTIVE

Epilepsy surgery is known to be underutilized. Machine learning-natural language processing (ML-NLP) may be able to assist with identifying patients suitable for referral for epilepsy surgery evaluation.

METHODS

Data were collected from two tertiary hospitals for patients seen in neurology outpatients for whom the diagnosis of "epilepsy" was mentioned. Individual case note review was undertaken to characterize the nature of the diagnoses discussed in these notes, and whether those with epilepsy fulfilled prespecified criteria for epilepsy surgery workup (namely focal drug refractory epilepsy without contraindications). ML-NLP algorithms were then developed using fivefold cross-validation on the first free-text clinic note for each patient to identify these criteria.

RESULTS

There were 457 notes included in the study, of which 250 patients had epilepsy. There were 37 (14.8%) individuals who fulfilled the prespecified criteria for epilepsy surgery referral without described contraindications, 32 (12.8%) of whom were not referred for epilepsy surgical evaluation in the given clinic visit. In the prediction of suitability for epilepsy surgery workup using the prespecified criteria, the tested models performed similarly. For example, the random forest model returned an area under the receiver operator characteristic curve of 0.97 (95% confidence interval 0.93-1.0) for this task, sensitivity of 1.0, and specificity of 0.93.

SIGNIFICANCE

This study has shown that there are patients in tertiary hospitals in South Australia who fulfill prespecified criteria for epilepsy surgery evaluation who may not have been referred for such evaluation. ML-NLP may assist with the identification of patients suitable for such referral.

PLAIN LANGUAGE SUMMARY

Epilepsy surgery is a beneficial treatment for selected individuals with drug-resistant epilepsy. However, it is vastly underutilized. One reason for this underutilization is a lack of prompt referral of possible epilepsy surgery candidates to comprehensive epilepsy centers. Natural language processing, coupled with machine learning, may be able to identify possible epilepsy surgery candidates through the analysis of unstructured clinic notes. This study, conducted in two tertiary hospitals in South Australia, demonstrated that there are individuals who fulfill criteria for epilepsy surgery evaluation referral but have not yet been referred. Machine learning-natural language processing demonstrates promising results in assisting with the identification of such suitable candidates in Australia.

Identifying epilepsy surgery candidates with natural language processing: A systematic review

INTRODUCTION

Epilepsy surgery is an underutilised, efficacious management strategy for selected individuals with drug-resistant epilepsy. Natural language processing (NLP) may aid in the identification of patients who are suitable to undergo evaluation for epilepsy surgery. The feasibility of this approach is yet to be determined.

METHOD

In accordance with the PRISMA guidelines, a systematic review of the databases PubMed, EMBASE and Cochrane library was performed. This systematic review was prospectively registered on PROSPERO.

RESULTS

6 studies fulfilled inclusion criteria. The majority of included studies reported on datasets from only a single centre, with one study utilising data from two centres and one study six centres. The most commonly employed algorithms were support vector machines (5/6), with only one study utilising NLP strategies such as random forest models and gradient boosted machines. However, the results are promising, with all studies demonstrating moderate to high levels of performance in the identification of patients who may be suitable to undergo epilepsy surgery evaluation. Furthermore, multiple studies demonstrated that NLP could identify such patients 1-2 years prior to the treating clinicians instigating referral. However, no studies were identified that have evaluated the influence of implementing such algorithms on healthcare systems or patient outcomes.

CONCLUSIONS

NLP is a promising approach to aid in the identification of patients that may be suitable to undergo epilepsy surgery evaluation. Further studies are required examining diverse datasets with additional analytical methodologies. Studies evaluating the impact of implementation of such algorithms would be beneficial.

The efficacy and tolerability of adjunctive brivaracetam for the treatment of adult epilepsy: An Australian multi-center retrospective real-world observational cohort study

OBJECTIVE

Assess the efficacy and tolerability of add-on therapy brivaracetam (BRV) in adult patients with epilepsy in a real-world setting.

METHODS

This multi-center retrospective observational cohort study examined all adult patients who commenced on BRV at 11 Australian epilepsy centers between 2017 and 2020. Primary outcomes were seizure response (≥50% reduction in frequency) and seizure freedom 12 months post BRV commencement, and tolerability. We report three approaches to missing data (complete case analysis, CCA; last observation carried forward, LOCF; and intention to treat, ITT). Secondary outcomes included the durability of early BRV response and continuous seizure freedom from BRV initiation. Subgroup analysis examined patients with focal and generalized epilepsy and patients with refractory (≥4 prior ASMs) and highly refractory (≥7 prior ASMs) epilepsy. Outcomes were also assessed at 'personalized' seizure outcome time points based on baseline seizure frequency.

RESULTS

Baseline and follow-up data were available for 228 patients. The mean age was 41.5 years (IQR 30, 50). Most had focal epilepsy (188/228, 82.5%). Median number of previous ASMs was 4 (2, 7), and concomitant ASMs 2 (2, 3). Twelve-month responder rate was: 46.3% using CCA (95% CI 34.0, 58.9); 39.5% using LOCF (33.1, 46.1); and 15.4% using ITT (10.9, 20.7). Twelve-month seizure freedom was: 23.9% using CCA (14.3, 35.9); 24.6% using LOCF (19.1, 30.7); and 7.9% using ITT (4.7, 12.1). The most frequent adverse effects were sedation or cognitive slowing (33/228, 14.5%), irritability or aggression (16/228, 7.0%), and low mood (14/228, 6.1%). Outcomes were similar using continuous outcome definitions and 'personalized' outcome assessment time points. Early responses were highly durable, with 3-month response maintained at all subsequent time points at 83%, and seizure freedom maintained at 85%. Outcomes were similar in focal (n = 187) and generalizsed (n = 25) subgroups. Outcomes were similar in refractory patients (n = 129), but lower in the highly refractory group (n = 62), however improvement with BRV was still observed with 12-month seizure freedom of 8.3% using CCA (1.0, 27), 6.5% using LOCF (1.8, 15.7); and 3.2% using ITT (0.4, 11.2).

CONCLUSIONS

Meaningful real-world responder and seizure freedom rates can be still observed in a refractory epilepsy population. Brivaracetam response can occur early and appears to be maintained with minimal later relapse. The results should be interpreted with caution given the retrospective nature of the study and the quantities of missing data at later time points.

Efficacy and tolerability of adjuvant perampanel: an Australian multicenter real-world observational study in refractory focal and generalized epilepsy syndromes

PURPOSE

To explore the efficacy and tolerability of adjuvant perampanel (PER) and their associated risk factors in late add-on drug-resistant epilepsy.

METHOD

Retrospective multicenter 'real-world' observational study. Consecutively identified patients commenced on PER, with mixed epilepsy syndromes, from nine Australian epilepsy centers. Primary efficacy endpoints were at least 50% reduction in seizure frequency (responders), seizure freedom, and retention at 6 and 12 months, following a 3-month titration period. Tolerability endpoints were cessation of PER for any reason, cessation of PER due to treatment-emergent adverse events (TEAE), or cessation due to inefficacy. Outcomes were assessed for a-priori risk factors associated with efficacy and tolerability.

RESULTS

Three-hundred and eighty seven adults were identified and followed up for a median of 12.1 months (IQR 7.0-25.2). Focal epilepsy accounted for 79.6% (FE), idiopathic generalized epilepsy (IGE), 10.3% and developmental epileptic encephalopathy (DEE) 10.1%, of the cohort. All patients had drug-resistant epilepsy, 71.6% had never experienced six months of seizure freedom, and the mean number of antiepileptic medications (AEDs) prior to starting PER was six. At 12 months, with missing cases classified as treatment failure, retention was 40.0%, responder 21.7%, and seizure freedom 9.0%, whereas, using last outcome carried forward (LOCF), responder and seizure freedom rates were 41.3% and 14.7%, respectively. Older age of epilepsy onset was associated with a marginal increase in the likelihood of seizure freedom at 12-month maintenance (OR 1.04, 95% CI 1.02, 1.06). Male sex (adjusted OR [aOR] 2.06 95% CI 1.33, 3.19), lower number of prior AEDs (aOR 0.84, 95% CI 0.74, 0.96) and no previous seizure-free period of at least 6-month duration (aOR 2.04 95% CI 1.21, 3.47) were associated with retention. Perampanel combined with a GABA receptor AED was associated with a lower responder rate at 12 months but reduced cessation of PER. The most common TEAEs were neuropsychiatric (18.86%), followed by dizziness (13.70%), and sleepiness (5.68%).

CONCLUSIONS

Adjuvant PER treatment, even in late-add on drug-resistant epilepsy is an effective and well-tolerated treatment for drug-resistant epilepsy.

Epilepsy and Seizure-Related Hospital Admissions to an Australian Neurology Unit: A Prospective Observational Study

This study provides contemporary data on patients admitted to a Neurology unit with seizures, the majority of whom have a diagnosis of epilepsy. There were 278 hospital presentations with seizure during the 6-month study period, with 60 admissions to the Neurology unit included for analysis. Provoking factors were identified in 40%, with poor medication adherence the commonest precipitant. CT-brain had low diagnostic yield in patients with epilepsy presenting with seizure and should be reserved for those with further indications for imaging. Patients with drug-resistant epilepsy comprised 54% of admissions suggesting management strategies in this cohort can be further optimized.

Potential role of cannabidiol for seizure control in a patient with recurrent glioma

Glioma-related epilepsy significantly impact on patients' quality of life, and can often be difficult to treat. Seizures cause significant morbidity for example neurocognitive deterioration, which may result from seizures themselves or due to adverse effects from antiepileptic drugs. Management of tumour with surgery, radiotherapy and chemotherapy may contribute to seizure control, but tumour related epilepsy is often refractory despite adequate treatment with standard anti-epileptic medications. Given the increasing interest in medicinal cannabis (or cannabidiol or CBD) as an anti-epileptic drug, CBD may help with seizure control in glioma patients with treatment-refractory seizures. Here we present a case of a young lady with recurrent glioma who had refractory seizures despite multiple anti-epileptic agents, who had significant benefit with CBD.

EpiNet as a way of involving more physicians and patients in epilepsy research: Validation study and accreditation process

Objective

EpiNet was established to encourage epilepsy research. EpiNet is used for multicenter cohort studies and investigator‐led trials. Physicians must be accredited to recruit patients into trials. Here, we describe the accreditation process for the EpiNet‐First trials.

Methods

Physicians with an interest in epilepsy were invited to assess 30 case scenarios to determine the following: whether patients have epilepsy; the nature of the seizures (generalized, focal); and the etiology. Information was presented in two steps for 23 cases. The EpiNet steering committee determined that 21 cases had epilepsy. The steering committee determined by consensus which responses were acceptable for each case. We chose a subset of 18 cases to accredit investigators for the EpiNet‐First trials. We initially focused on 12 cases; to be accredited, investigators could not diagnose epilepsy in any case that the steering committee determined did not have epilepsy. If investigators were not accredited after assessing 12 cases, 6 further cases were considered. When assessing the 18 cases, investigators could be accredited if they diagnosed one of six nonepilepsy patients as having possible epilepsy but could make no other false‐positive errors and could make only one error regarding seizure classification.

Results

Between December 2013 and December 2014, 189 physicians assessed the 30 cases. Agreement with the steering committee regarding the diagnosis at step 1 ranged from 47% to 100%, and improved when information regarding tests was provided at step 2. One hundred five of the 189 physicians (55%) were accredited for the EpiNet‐First trials. The kappa value for diagnosis of epilepsy across all 30 cases for accredited physicians was 0.70.

Significance

We have established criteria for accrediting physicians using EpiNet. New investigators can be accredited by assessing 18 case scenarios. We encourage physicians with an interest in epilepsy to become EpiNet‐accredited and to participate in these investigator‐led clinical trials.

Predicting Prolonged Dysphagia in Acute Stroke: The Royal Adelaide Prognostic Index for Dysphagic Stroke (RAPIDS)

Dysphagia is common after stroke and is associated with increased morbidity and mortality. Predicting those who are likely to have significant prolonged dysphagia is not possible at present. This study was undertaken to validate the Royal Adelaide Prognostic Index for Dysphagic Stroke (RAPIDS) in the prediction of prolonged dysphagia following acute stroke using clinical and radiographic features. A prospective study of unselected, consecutive admissions to the Royal Adelaide Hospital acute stroke unit was undertaken. Clinical and radiographic features applicable to the RAPIDS test were calculated and the sensitivity, specificity, and likelihood ratio for predicting prolonged dysphagia were calculated with 95% confidence intervals (CI). Of 104 subjects admitted with acute stroke, 55 (53%) had dysphagia and 20 (19%) had dysphagia requiring nonoral feeding/hydration for 14 days or more or died while dysphagic prior to 14 days. The RAPIDS test had sensitivity of 90% (95% CI = 70-97%) and specificity of 92% (95% CI - 84-96%) for predicting this latter group of patients. We conclude that the RAPIDS test can be used early to identify patients likely to have prolonged dysphagia. This test could form a basis for selection of patients for trials of nonoral feeding methods.