Abstracts

Many patients with status epilepticus (SE) have more than one etiology associated with each SE episode. These etiologies are recorded in arbitrary order but are of equal importance. When analyzing data on outcomes such as mortality from SE, including multiple etiologies as covariates poses a challenge. It is often necessary to combine multiple etiologies into a single etiology category. In this study, different ways of assigning etiology by condensing information from multiple etiologies is addressed. Since 1989, NIH Greater Richmond Metropolitan Area Data System on SE has been collecting data. The database, which consists of over 600 patients, is a comprehensive collection of laboratory, clinical and demographic data on every SE case. Several analyses have been performed in the past. The variables studied include mortality, etiology, race, sex, age, seizure duration, seizure type, time to treatment, etc. When more than one etiology was present, patients were assigned a single etiology using one of three different methods. In the first method, one etiology from the multiple etiologies reported was chosen based on consensus from clinicians (called ETCOM). The second and the third definitions were based on data alone. For these a table of mortality by etiology was constructed. Then the etiology that had the highest mortality was assigned to be the etiology of the patient (called ETMAX). For example, if CNS Acute and Withdrawal were observed and the mortality for CNS Acute and Withdrawal in the population were 24% and 6.4%, respectively, then the etiology assigned to that patient, ETMAX, was CNS Acute. Similarly, ETMIN using the lowest mortality was also considered. Three logistic regression models were fitted for the mortality outcome with one of the three definitions of the etiology. The first method (ETCOM) led to 15 categories, which included combined etiologies. The second and the third methods led to 6 categories each. When included in the logistic regression model, in general the three different assignments were found to be comparable. The parameter estimates and the p-values turned out to be similar. Deatails of the analyses will be presented in the poster. Data are being collected constantly in many fields of medical science. Often, in publications of the results from analyses of these data , the nuances applied in the analyses are not explained. To ensure consistency in the approaches in similar fields, it is important to disseminate the new methodology applied even if it is not directly relevant to the subject matter. In this poster, one such issue that relates to defining etiology in SE was addressed. It was found that the consensus based definition of etiology provided similar results to definitions that were purely data driven. It is concluded, to facilitate interpretation it is best to use consensus based definitions. (Supported by NIH P01 NS25630)