Background: It is necessary to select a range of consistently identified prognostic factors from exploratory studies to include in multivariate models of confirmatory studies. We illustrate a systematic approach to selecting consistently identified prognostic factors using the example of predictors of remission in newly diagnosed epilepsy. Methods: Medline and Embase were searched for reports of cohort studies enrolling at least 100 people with epilepsy within 1 year of diagnosis, and followed up for at least 1 year. We included studies that identified predictors of remission after adjusting for confounders using multivariate regression analysis. To identify consistent predictors a chart was designed to list the variables considered for inclusion in each model and those retained in more than one model from different cohorts were deemed to be consistent. Results: Remission off medication was less likely if there was more than one seizure between 6 and 12 months on medication and if there was comorbid intellectual disability in childhood onset epilepsy. The likelihood of remission on or off medication reduces with mixed seizure types at onset, intellectual disability, symptomatic aetiology, and also with increasing number of seizures before diagnosis or in the first 6 months after diagnosis. Conclusion: A greater number of seizures before diagnosis and early in treatment, intellectual disability, and symptomatic aetiology are consistent predictors of less likelihood of remission. This suggests that early identification, diagnosis of epilepsy, and seizure control should be the primary aim of medical intervention, and that these predictors should be included in future confirmatory studies of prognostic factors of remission in newly diagnosed epilepsy.