Development and validation of clinical tools for predicting outcome in newly diagnosed epilepsy

Background Epilepsy is a common condition with a prevalence of 0·5–1% and lifetime incidence of up to 5%.

Antiseizure medication (ASM) is the treatment mainstay, with approximately 70% of patients achieving a period of seizure remission of at least 12 months.

However, patient outcomes are heterogeneous and ASMs fail to control seizures or cause intolerable side effects for many patients. Approximately 30% of patients will become drug-resistant.

Given the heterogeneity, there are numerous patient factors that are associated with important outcomes of seizure remission, treatment failure and drug-resistance.

Identifying an individual patient s risk could facilitate more satisfactory conversations and shared-decision making as well as identifying patients that may require additional monitoring or earlier intervention such as brain surgery.

Prediction models provide clinicians and patients with tools for estimating the risk that an event will occur in the future. Prediction models should be regularly re-assessed to ensure that the best models are being used in clinical practice.

Aims To develop, update and validate prediction models for seizure remission, treatment failure and drug-resistance in patients with newly diagnosed epilepsy.

Objectives Identify and review existing models and consolidate what is already known Update existing models and develop new models if required Externally validate models where possible so that they are ready for implementation into practice Assess the added value of genetic data.

Methods Work package 1 will identify and critically appraise all published prediction models for seizure remission, treatment failure and drug-resistance. The review will be conducted and reported in accordance with current best practice. A study specific data extraction tool will be developed based on the CHARMS checklist and piloted prior to use.

Models will be critically appraised using the PROBAST tool.

Models classed as high applicability will be taken forward into work package 2 and updated using regression coefficient updating and meta-model updating. New prediction models will be developed if we fail to identify suitable existing models.

Models will be externally validated using the most relevant data that we have available from 37 clinical trials (14,789 patients) by evaluating the discrimination and the calibration of the identified prediction model in the cohort of patients being used for external validation. The final validated models will be discussed with patients and presented as nomograms to aid interpretation.

Work package 3 is an exploratory analysis to assess the added value of genetic data.

Previous research has demonstrated that genetic factors are useful for classifying patients according to epilepsy seizure type, and it is believed that quantification of genetic variant burden may also be valuable in guiding epilepsy treatment.

Using genome-wide data from the SANAD trials (n=1200), we will explore the added value of including genetic data in each prediction model by comparing models with and without the genetic with reference to discrimination and calibration measures.

Anticipated impact This study will provide the epilepsy community with comprehensive, up-to-date evidence-informed recommendations for predicting outcomes in patients with newly diagnosed epilepsy. This has the potential to improve the management and care provided to epilepsy patients.