Objective: Significant benefit could be realized by developing a clinical decision rule for new-onset seizure victims that would be capable of discriminating between patients having relevant structural lesions visible on computed tomographic (CT) imaging and those who do not. This study sought to determine whether a reliable decision rule could be developed using a limited number of clinical and demographic characteristics.
Methods: Chi-squared recursive partitioning was applied in a secondary analysis of the EMERGEncy ID NET database of new-onset seizure victims. Variables in this database (age, sex, race, ethnicity, seizure type, history of HIV or cysticercosis, and presence or absence of lateralizing neurologic findings or altered mentation) provided the partitioning variables, while CT imaging results provided outcome measures. The study sought to develop a decision rule with 100% sensitivity for detecting any intracranial lesions, and a separate rule with 100% sensitivity for detecting lesions of emergent concern.
Results: A decision rule using age > or = 65 years, lateralizing neurologic findings, altered mentation, high risk or known HIV infection, history of cysticercosis, and Hispanic ethnicity showed a sensitivity of 91.9% [95% confidence interval (95% CI) = 88.8% to 94.9%] in detecting individuals who had any tomographic finding. This rule had a sensitivity of 90.1% (95% CI = 83.4% to 96.7%) in detecting individuals with emergent tomographic findings.
Conclusions: Recursive partitioning failed to produce a decision rule capable of reliably identifying new-onset seizure patients who have important lesions identified on CT. Future attempts to formulate such an instrument may need to include additional variables. In the interim, physicians should use liberal tomographic imaging in evaluating patients who present with new-onset seizures.