Clinical Prediction Models for Treatment Outcomes in Newly-diagnosed Epilepsy

Background Up to 35% of individuals diagnosed with epilepsy proceed to develop pharmacoresistant epilepsy, leading to persistent uncontrolled seizure activity that can directly, or indirectly, significantly degrade an individual’s quality of life. The factors underlying pharmacoresistance are unclear, but it has been hypothesised that repeated ictogenic activity is conducive to the development of a more robust epileptogenic network. To ensure that the most effective treatment choices are made and ictogenic activity is minimised, accurate outcome modelling at the point of diagnosis is key.

Objectives This review therefore aims to identify demographic, clinical, physiological (e.g. EEG), and imaging (e.g. MRI) factors that may be predictive of treatment outcomes in patients with newly diagnosed epilepsy (NDE).

Data sources, study eligibility criteria, participants, and interventions MEDLINE and EMBASE were searched for prediction models of treatment outcomes in patients with newly diagnosed epilepsy and any non-surgical treatment plan.

Study appraisal and synthesis methods Study characteristics were extracted and subjected to assessment of risk of bias (and applicability concerns) using the PROBAST tool. Prognostic factors associated with treatment outcomes are reported.

Results After screening, 48 models were identified in 32 studies, which generally scored low for concerns of applicability, but universally high for susceptibility to bias. Outcomes reported were heterogenous, but fit broadly into four categories: pharmacoresistance, 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 (semiology), epilepsy history, and age at onset. ASM response was often included as a prognostic factor, potentially obscuring factor relationships at baseline.

Conclusions Currently, outcome prediction models for NDE demonstrate a high risk of bias. Model development could be improved with a stronger adherence to recommended TRIPOD practices, and by avoiding including response to treatment as a prognostic factor.

Implications of key findings This review identified semiology, epilepsy history, and age at onset as factors associated with treatment outcome prognosis, suggesting that future prediction model studies should focus on these factors in their models. Furthermore, we outline actionable changes to common practices that are intended to improve the overall quality of prediction model development in NDE.

Key Points

• This paper presents a systematic literature search for treatment outcome prediction models in newly diagnosed epilepsy.

• The risk of bias in the included models were evaluated using the PROBAST framework, finding a universally high risk level.

• The relationship between semiology, epilepsy history, and age at onset with seizure remission should be examined in future prediction model studies.

• To improve the overall quality of prediction model development in NDE, prospective authors are advised to adhere to TRIPOD guidelines, and to avoid including response to treatment as a prognostic variable.