Some Epilepsy Patients Sense Impending Seizures

NEW YORK, Jan. 23 -- Some patients with epilepsy appear to know when a seizure is imminent, suggesting that they could take pre-emptive precautionary measures such as avoiding driving, researchers here have found.

Among 134 patients with localization-related epilepsy, those who predicted they would have a seizure the next day had a twofold greater chance for a seizure, reported Sheryl R. Haut, M.D., and colleagues of the Montefiore Medical Center and Albert Einstein College of Medicine in the Bronx.

"Our data indicate that a significant subgroup of patients with epilepsy can predict their seizures, and that for these predictors, the odds of a seizure following a positive prediction is high," the investigators wrote in the Jan. 23 issue of Neurology.

The authors suggested that a significant subset of patients with localization-related epilepsy could be trained to predict when a seizure is likely to occur, by focusing on particular patterns, premonitory symptoms, and precipitating factors.

But whether the study participants were tuned in to their body signals or just got lucky at guessing when a seizure episode was going to strike is still unclear, suggested Brian Litt, M.D.; and Abba Krieger, Ph.D., of the University of Pennsylvania, in an accompanying editorial.

"Straightforward but difficult-to-answer questions abound," they wrote "Did a seizure really occur? Were patients unaware of seizures, a common problem . . . .? Were questionnaires filled out prospectively, or did patients 'cheat' after the fact? Is patient accuracy based upon knowing their regular cycles, or upon changes in brain physiology prior to each seizure? Was prediction performance better than random guessing?"

The investigators recruited 134 adults with localization-related epilepsy into a prospective study using seizure diaries to determine the validity of predictions. Patients were eligible if they were 18 or older and had had at least one seizure within the last year.

Seizure prediction and seizure occurrence were modeled as binary outcomes, and odd ratios, positive predictive value, and negative predictive value were calculated for the group and for individuals, adjusted for the within-person correlations.

In all, 71 patients completed the study requirements, returning diaries covering a total of 15,635 daily entries.

The authors determined that a positive prediction of seizures was associated with a twofold risk of seizures within 24 hours, with an odds ratio (Cochran-Mantel-Haenszel method) of 2.25 (95% confidence interval, 1.91 to 2.65).

The overall specificity for positive prediction was 83.2%; the sensitivity was 31.9%. Among all predictors, 12 of the patients demonstrated significant within-person odds ratios: 3.14 (95% CI, 2.53 to 3.89).

Neither gender nor epilepsy localization was associated with prediction, but the authors did find that predictors were significantly younger (P= 0.026) and had a higher seizure rate (P= 0.003) than non-predictors, although the last finding may have been confounded by the statistical power of the study, Dr. Litt and Dr. Krieger pointed out in their editorial.

"It may be that because of too few seizures during the sample period, the authors actually misclassified some patients with lower seizure frequencies who were good predictors as non-predictors, because of insufficient data," the editorialists wrote.

They pointed out several potential sources of bias:

• "First, the pool of 134 subjects is not randomly chosen from a well-defined population raising the question of selection bias.

• "Nonresponse bias is potentially of more concern. Specifically, it is not clear that the 71 responders are representative of all patients enrolled, and whether failure to send back diaries was influenced by a lack of ability to predict seizures. If this is the case, then the results of the study might reflect the outcomes of subjects who are better predictors, and the number of predictors begins to approach what might be expected by chance alone.

• "The third and perhaps most serious concern is response bias. The data in this study are self reported. A willingness of subjects to please investigators might have led to reports that improperly demonstrated prediction."

Dr. Haut and colleagues suggested that self-prediction of seizures could be combined with quantitative EEG analysis to improve the efficacy of both techniques.

"Further investigation into seizure self-prediction may provide the potential to develop pre-emptive treatments as an additional strategy for reducing the burden of refractory epilepsy," they concluded.

The authors acknowledged that the study was limited by the lack of a precise time stamp on the paper diaries used, and by a failure to include in the diaries questions about premonitory symptoms, which might have helped to improve the sensitivity of the predictions.