Validation of a brief screening instrument for the ascertainment of epilepsy

Valid screening questions to identify people with a history of seizures or epilepsy are crucial for epidemiologic studies. They are also essential for identifying affected family members in genetic studies. Screening relies on careful history-taking, but it is not clear what questions should be asked, or how accurately they identify affected individuals. Several previous epidemiologic studies have assessed epilepsy prevalence using a two-stage screening strategy in which an initial broad screen identified possible cases and subsequent, more detailed assessment separated true from false positives (Osuntokun et al., 1982; Haerer et al., 1987; Schoenberg, 1987; Meneghini et al., 1992; Placencia et al., 1992; Nicoletti et al., 1999; Borges et al., 2004; Melcon et al., 2007; Noronha et al., 2007). The initial screens included both items that targeted recognized seizures and items that targeted symptoms possibly reflecting unrecognized seizures [e.g., “Have you ever had attacks in which you lose contact with the surroundings?”(Placencia et al., 1992)]. The use of symptom-based questions is essential for maximizing sensitivity (i.e., the proportion of true cases in the population who screen positive), especially in settings with limited access to medical care, and many of these questionnaires were found to have sensitivity ≥95% (Osuntokun et al., 1982; Meneghini et al., 1992; Placencia et al., 1992; Borges et al., 2004; Melcon et al., 2007); however, these high sensitivities came with a “cost” of false positives, and estimates of specificity (i.e., the proportion of unaffected individuals who did not screen positive) were usually lower (Osuntokun et al., 1982; Meneghini et al., 1992; Placencia et al., 1992; Borges et al., 2004; Melcon et al., 2007). False-positive rates have major implications for the effort and cost involved in a study because they affect positive predictive value (PPV): the proportion of screen-positive individuals subsequently confirmed to be affected. Because epilepsy is relatively uncommon (life-time prevalence 0.5–3% in many studies), even a small false-positive rate will be applied to more than 95% of the individuals sampled, so that the proportion of screen-positive individuals who are truly affected could be quite low. For many of the screening questionnaires used previously, PPV was as low as 20% (reviewed in Placencia et al., 1992). We designed a nine-question screening instrument to identify subjects with epilepsy in a telephone interview. We assessed sensitivity and false-positive rates for the instrument by administering it to individuals with medical record–documented epilepsy or isolated unprovoked seizure and individuals who were seizure-free on medical record review, from a population-based study. We also explored the effect of restricting the screen to different subsets of questions. Finally, we used our estimates of sensitivity and false-positive rates to extrapolate to the anticipated PPV that would be obtained in a population survey with our instrument.