A Bayesian tool for epilepsy diagnosis in the resource-poor world: development and early validation.

ABSTRACT

PURPOSE: The epilepsy treatment gap in resource-poor countries is so large that existing numbers of doctors are unlikely to be able to close it. Other health workers are likely to be needed but they will need help. The diagnosis of an attack as epileptic or not is an essential step in the management of epilepsy. It should be possible to devise a tool to give the probability of episodes being epileptic based on a Bayesian analysis of the results of history taking. METHOD: We asked about the nature of episodes in patients referred to epilepsy camps in Nepal. Answers were then compared to the final clinical diagnosis of epilepsy and the likelihood ratio (LR) of the episode being epileptic obtained for each answer. The most informative LRs, tested sequentially, formed the basis for a tool which was validated in a different Nepalese population. RESULTS: Data was obtained from 67 patients. The pre-test probability of having epilepsy was 0.76. Answers to 11 questions with the most informative LRs were then combined into a tool. This was tested on 14 different patients. Post-test probability scores in those with epilepsy ranged from 0.88 to 1 and for those with non-epilepsy from 0.07 to 0.42. CONCLUSION: It is possible to devise a tool based on simple clinical information using Bayesian principles. Initial validation suggests that this has the potential to enable health workers to diagnose episodes as epileptic or not. This now needs to be tested in different populations. The tool is easily converted to a mobile phone app.