- Email: [email protected]
Premonitory features and seizure self-prediction: Artifact or real?
Epilepsy Research (2011) 97, 231—235
journal homepage: www.elsevier.com/locate/epilepsyres
Premonitory features and seizure self-prediction: Artifact or real? Andreas Schulze-Bonhage a,∗, Sheryl Haut b a b
Epilepsy Center, University Hospital, Freiburg, Germany Albert Einstein College of Medicine, New York, USA
Received 10 February 2011; received in revised form 4 August 2011; accepted 8 September 2011
KEYWORDS Epilepsy; Seizure prediction; Prodrome; Trigger factor
Summary Seizure prediction is currently largely investigated by means of EEG analyses. We here report on evidence available on the ability of epilepsy patients themselves to predict seizures either by means of subjective experiences (‘‘prodromes’’), apparent awareness of precipitants, or a feeling of impending seizure (self-prediction). These data have been collected prospectively by paper or electronic diaries. Whereas evidence for a predictive value of prodromes is missing, some patients nevertheless can forsee impending seizures above chance level. Relevant cues and practical implications are discussed. © 2011 Published by Elsevier B.V.
Epilepsy is characterized by the spontaneous and unprovoked occurrence of seizures. For patients and observers, these seizures appear to come out of the blue, which has considerable consequences for patients. Even if seizures occur quite rarely (e.g. a few times per year), the fact that their timing is unknown has legal consequences which hinder patients from driving motor vehicles and working in certain professions. Furthermore, spontaneous seizures may present physical risk during everyday activities such as bathing, or during sports, and may lead to social avoidance behaviour and a feeling of loss of control which may result in depression (Schulze-Bonhage and Buller, 2008). As a patient put it, ‘‘It is the unpredictability of it that is really nerve-racking to live with’’ (Murray, 1993).
∗
Corresponding author. E-mail address: [email protected] (A. Schulze-Bonhage). 0920-1211/$ — see front matter © 2011 Published by Elsevier B.V. doi:10.1016/j.eplepsyres.2011.09.026
Yet, from the earliest descriptions of epilepsy by Hippocrates, ‘‘warnings’’ experienced by patients prior to a visible seizure have been mentioned, and the concept of ‘‘prodromes’’ is found in textbooks on epilepsy suggesting that at least some patients experience precursors of seizures. Furthermore, patients often report that they have a sense of when their risk for seizure is high. So the question arises if certain subgroups of epilepsy patients might be able to predict their own seizures. In principle, seizure predictions of patients could be based on two sources: subjective experiences of ‘‘warnings’’ preceding a seizure, and knowledge of factors which increase the probability of a subsequent seizure occurrence. Warnings can be divided into an epileptic aura which may precede a focal seizure by seconds to minutes, and into subjective feelings experienced longer before a seizure, so-called ‘‘prodromes’’. Auras can be very useful to patients, if they allow them to retreat from public or potential dangerous situations or even to take counter-measures
232 against seizures; they are considered as part of the ictal event, however, so that the experiences during an aura can be considered as subjective detection of a seizure, but not as a prediction. Prodromes, on the other hand, are believed to be nonictal events, i.e. events which are not accompanied by ictal discharges in the EEG. Generally, they are considered to precede a seizure by longer time periods of hours to days, and their physiological background has remained obscure so far. Prodromes may thus be preictal events based on which patients might be able to predict an impending seizure. Aside from such subjective experiences, patients may have knowledge of trigger factors or precipitants of their seizures, upon which they may base a prediction of an impending seizure. It has been hypothesized that true unprovoked seizures, unrelated to precipitants, may actually be rare (Rajna et al., 2008). Such precipitating factors could comprise aspects temporal effects such as circadian or catamenial patterns; acute precipitants such as sleep deprivation, stress or other emotional factors, or simply non-compliance with medication intake. Awareness of any of these factors could lead to valid predictions of increased seizure susceptibility at certain periods of time. Methodologically, the field of clinical prediction of seizures depends on reliable patient report. Examination of the relevant studies mandates acknowledgement that diary formats differ significantly. Questionnaires are commonly used to identify premonitory features and precipitants reported by patients in a cross-sectional manner. This format, while useful, is subject to recall bias, and the findings must be tested prospectively. Prospective paper diaries have been widely used, but are limited by the lack of time stamping, and the risk of backfilled or retrospectively entered data. Electronic diaries are increasingly utilized in studies of premonitory features and seizure prediction, but present other challenges of data transfer and increased cost. In the following, studies analysing prodrome- and knowledge-based seizure predictions are addressed separately.
Prodromes and seizure prediction Prodromes have been the subject of descriptions in the literature (Dostoyevsky, 1868) and of scientific investigations. Most publications used questionnaires in which patients or at times also caregivers were asked if they would experience changes in their subjective perception or in their behaviour. Depending on the type of question and on the population investigated, rates of 6.9% up to 39% of patients with prodromes were reported (Giuccioli et al., 1990; Hughes et al., 1993; Rajna et al., 1997; Schulze-Bonhage et al., 2006; Scaramelli et al., 2009). A separation from auras was performed in most studies by defining a time interval before an upcoming seizure, which was mostly set as more than 30 min (Schulze-Bonhage et al., 2006), rarely also only more than 5 min before seizure onset. One study analyzed if the phenomenology of ‘‘prodromal’’ symptoms was similar to or different from auras; out of 15 patients who were able to describe their experiences with sufficient detail, 12 reported clearly different perceptions during the periods far from a seizure as compared to auras directly progressing
A. Schulze-Bonhage, S. Haut into a seizure (Schulze-Bonhage et al., 2006). Four studies analyzed the dependency of a prodrome on the classification of epilepsy; there was a concordantly higher frequency of prodromes in structural as compared to genetic epilepsy (Hughes et al., 1993; Rajna et al., 1997; Schulze-Bonhage et al., 2006; Scaramelli et al., 2009). The exact etiology in cases of structural epilepsy did not appear to play a central role (Rajna et al., 1997). Based on questionnaires, the phenomenology of subjective experiences during ‘‘prodromes’’ was reported. There was considerable variability between the statements of individual patients regarding their prodrome-sensation. Mostly, vegetative symptoms (e.g. palpitations, sweating, gastrointestinal symptoms), or emotional disturbances (e.g. irritability fatigue, anxiety, depression) are reported. Notably when summarizing the range of symptoms reported, there is wide variability between studies, and an impressively wide spectrum of possible complaints is reported (Table 1). The validity of prodromes as seizure precursors has recently been questioned. Methods used for an analysis of EEG-based seizure prediction methods (AschenbrennerScheibe et al., 2003; Maiwald et al., 2004; Mormann et al., 2005; Winterhalder et al., 2003; Schelter et al., 2006, 2007) generally were not applied to evaluate the validity and prediction performance of prodromes. Taylor (2007) accordingly pointed out that there are no studies which prove that prodromes are preictal events, and has put forward the hypothesis that prodromal experiences may erroneously be considered as seizure-related when epilepsy is considered to be a disease essentially consisting of seizures, and that they might be seizure-independent symptoms related to the neurobiological background manifesting in both, seizures and independent alterations in subjective experiences. At the Freiburg epilepsy center, a prospective study using handheld computers was performed which intended to identify patients who were able to predict their own seizures. Participants were recruited from a multicentric assessment in 500 patients to identify the subgroup convinced to experience seizure precursors (Schulze-Bonhage et al., 2006). Patients with a minimum seizure frequency of 1/month were assessed prospectively to state if they experienced a prodrome every 12 h, and they were asked to perform free entries whenever a prodrome or a seizure happened. Prediction performance was assessed using a methodology developed for EEG-based seizure prediction algorithms (Winterhalder et al., 2003). Data entries into the handheld occurred patient-initiated at any time, and at standardized points of time every 12 h according to an alarm given by the handheld to the patient. Patients entered prodromes and their type as well as seizures and indicated when either had occurred. Out of nine patients in whom at least 4 weeks of continuous entries were available, none had a predictive performance which was statistically better than to be expected from a random predictor, even when the false prediction rate was chosen according to the patient’s performance and when various seizure occurrence periods up to 24 h were analyzed (Maiwald et al., 2011). It is of interest that in this study not only the time of seizures and prodromes as reported by the patient was stored but also data entry times were time stamped. An analysis of time stamps of entries for seizures and prodromes
Premonitory features and seizure self-prediction: Artifact or real? Table 1
233
Premonitory symptoms as reported in studies on prodromes.
Giuccioli et al. (book art. 1990) n = 53
Hughes et al. Seizure (1993) n = 148
Rajna et al. Seizure (1997) n = 262
Schulze Bonhage et al. Epil Res (2006) n = 35
Scaramelli et al. Seizure (2009) n = 39
Absent-mindedness Cold hands/feet Depression Dizziness Fatigue Fearfulness Headache Heat waves Introversion Irritation Palpitations Polyuria Sleeplessness Speech disturbance Sweating Tiredness
Anxiety Confusion Depression Elation Fear ‘‘Funny feeling‘‘ Headache Irriability Menstrual symp. Polyuria Speech disturbance
Epigastric sensation ‘‘Funny feeling’’ Headache
Dizziness Headache Impaired concentration Malaise Nausea Restlessness Tiredness
Appetite change ‘‘Behavioural’’ change Cognitive disturbance Dysthermia Fatigue GI symptoms Headache Paresthesias Pain Sleep disturbance Speech disturbance Voiding changes
Note: High variability with symptoms related to pain, vegetative signs, changes in mood and cognitive impairments.
showed that some ‘‘prodromes’’ were entered only at a time when a seizure had already happened and retrospectively were reported as pre-ictal events according to the time when the patient stated it to have happened. Although this may be called ‘‘cheating’’, it may just correspond to the lack of specificity of the experiences which based on their phenomenology cannot be regarded as clear preictal phenomena. Only when a seizure follows these experiences, they may be retrospectively classified as a preictal sensation, conceptually reflected by the term ‘‘prodrome’’. As these experiences may not be entered as prodromes when no seizure follows, a statistical bias in favour of specificity can be the consequence of such retrospective reporting.
Seizure precipitants and self-prediction In the general epilepsy population, up to 90% of patients with epilepsy identify at least one seizure precipitant (Neugebauer et al., 1994; Frucht et al., 2000; Spector et al., 2000; Nakken et al., 2005; Sperling et al., 2008). Emotional stress and stressful life events, sleep deprivation, depression, anxiety, menstruation, and alcohol use are among the most commonly reported seizure precipitants. It seems likely that the presence of precipitants may make patients aware of the risk of an impending seizure. What follows is that if self-prediction of seizures is possible, at least part of this predictive ability is related to this awareness. One study addressed this possibility prospectively in a paper diary design (Haut et al., 2007a,b). Subjects provided data on potential precipitants which have been identified in other studies including hours of sleep, levels of stress and anxiety, menstruation and alcohol, on a daily basis. Seizure self-prediction was assessed by the following question, also completed on a daily basis: Do you think you will have a seizure in the next 24 h? Response options included extremely likely (1), somewhat likely (2), somewhat unlikely
(3) and extremely unlikely (4). Of note, premonitory feature report was not included in this study. Seventy-one subjects returned a total of 15,179 days followed with information regarding seizure occurrence the next day. Of these, fifty-seven subjects (80%) experienced at least one seizure during the period of diary followup. Overall sensitivity for the population was 31.9%, in that 475 of 1488 seizure days had an antecedent prediction of ‘‘extremely likely’’ or ‘‘somewhat likely’’. For the 13,691 seizure-free days for which there was a prediction, 11,388 had been rated ‘‘somewhat unlikely’’ or ‘‘extremely unlikely’’, yielding a population specificity of 83.2%. Adjusting for heterogeneity across subjects, an overall specificity of 87% and sensitivity of 21% were obtained. The Cochran—Mantel—Haenzel OR for positive prediction was 2.25 (1.91—2.65) indicating that seizures were twice as likely in the 24 epoch following a prediction. Noting the change in sensitivity with adjustment across subjects, the predictive ability of individuals was explored. Twelve subjects demonstrated significant individual predictability and were termed ‘‘predictors’’. The Mantel—Haenzel estimate of the OR for the predictor group was 3.14 (2.53—3.89), indicating that following a positive prediction, the subject was more than three times more likely to experience a seizure. Accuracy of seizure prediction for the subjects with significant predictive ability, adjusted for individual contribution, is presented in Fig. 1 (Haut et al., 2007a). In order to examine the factors related to this ability to predict an impending seizure, data on the potential precipitants were entered into predictive models (Haut et al., 2007b). Precipitants remaining significant in the model included reported levels of stress, anxiety and hours of sleep for the night prior to the seizure. When patient self seizureprediction was included, variables which remained in the model included self-prediction (OR 3.7; 95% CI 1.8—7.2), and hours of sleep for the night prior to the seizure (OR .90; 95% CI 82—.99). The fact that self-reported stress and
234
A. Schulze-Bonhage, S. Haut
Accuracy of prediction
Seizure No seizure
100% 80% 60% 40% 20% 0% extremely likely
somewhat likely
somewhat unlikely
extremely unlikely
Seizure prediction
Figure 1 Accuracy of seizure predictors. PPV and NPV values for the subgroup (n = 12) of subjects demonstrating significant seizure prediction. Values are adjusted for within person correlation. Table 2
Hypothetical scenario based on precipitant report.
Stress
Anxiety
Hours of sleep
Risk of seizure %
Low Moderate High
Low Moderate High
High Moderate Low
4.1 5.9 8.5 Prediction
Any Any Any
Any Any Any
High Moderate Low
Low Moderate High
2.9 6.7 13.1
anxiety levels did not remain significant in the model once self-prediction was included led to the conclusion that subjects felt most likely to experience a seizure when they were reporting greater stress and anxiety. A number of hypothetical scenarios (Table 2) were calculated for a 40-year old female subject. In this scenario, the highest risk of seizure is modeled with report of fewer hours of sleep and high certainty of self-prediction. As discussed in an accompanying editorial (Litt and Krieger, 2007), this study was limited by the paper diary format. The study was then repeated with electronic diaries and extended sampling strategies including report of premonitory features, and more extensive mood measurements. We anticipate that these data will further clarify the accuracy of modeling seizure occurrence from patient reported data. Seizure self-prediction has been similarly reported in a recent inpatient study (Dubois et al., 2010). In this study, 83 subjects provided self-predictions of impending seizures during video-EEG monitoring. Following a positive prediction, the probability of a seizure was 0.320 (95% CI: 0.149—0.558), significantly increased over the probability of a seizure (0.151, 95% CI: 0.71—0.22) following a negative prediction.
Discussion The ability of patients to self-predict seizures has so far remained a matter of controversy. The recent investigation of Maiwald et al. (2011) suggests that subjective experiences of patients do not have a predictive value for seizures to
follow within a period of 1—24 h. The practical value of prodromes for seizures had already been called into question by Petitmengin et al. (2006) who pointed out that the subjective sensations called prodromes may be of limited value in practical life due to the lack of specificity as seizure precursors. The PDA-based prospective analysis of Maiwald et al. (2011) showed that predictions based on prodromes were not superior to random predictions taking into consideration relatively long seizure occurrence periods. It remains an open question as to whether the subjective sensations of patients are insufficient as seizure predictors as they are in fact unrelated to subsequent seizures or only due to the fact that they are not specific enough in character to be identified as preictal events. It has to be stated, that thus far, a prodrome-based predictability of seizures has not been proven. Whereas the majority of seizures that occur at present are not clearly predicted by patients, the recent studies by Haut et al. and others suggest that a subgroup of patients may be aware of impending seizures. These patients appear to be able to use knowledge about trigger factors to predict seizures. Further analyses on the factors relevant for these predictions, particularly seizure triggers or precipitants, may result in a more quantitative, individualized approach and may be helpful for patients in their everyday life. Patients may not recognize all of their seizures, and in fact, the accuracy of seizure diaries and seizure self-report continues to be debated (Blum et al., 1996; Tatum et al., 2001; Hoppe et al., 2007). Methodologically, the paucity of studies investigating prodromes and prediction in relation to EEG-confirmed seizures introduces the possibility that seizures are underreported, and that the relationship between prediction and seizure occurrence may be overestimated. Dubois et al. (2010) however recently demonstrated that patient awareness of seizures was high. This issue will remain critical for planning interventional studies. In a practical sense, the value of patient-based seizure prediction at the currently reported sensitivities is limited. Even for patients able to predict their seizures above chance level, sensitivity is below 50%, which is lower than the requirements suggested for future EEG-based prediction methods (Schulze-Bonhage et al., 2010). It is an interesting question, however, whether education and training may improve the performance of patient-based predictions compared to the naive state, in which both studies on prodromes and on unspecified predictions were performed. If so, this might allow both behavioural and possibly pharmacologic interventions, including extra precautions and temporally targeted interventions during periods of high risk.
References Aschenbrenner-Scheibe, R., Maiwald, T., Winterhalder, M., Voss, H.U., Timmer, J., Schulze-Bonhage, A., 2003. How well can epileptic seizures be predicted? An evaluation of a nonlinear method. Brain 126, 2616—2626. Blum, D.E., Eskola, J., Bortz, J.J., Fisher, R.S., 1996. Patient awareness of seizures. Neurology 47, 260—264. Dostoyevsky, F., 1868. The Idiot, Part II, 5.
Premonitory features and seizure self-prediction: Artifact or real? Dubois, J.M., Boylan, L.S., Shiyko, M., Barr, W.B., Devinsky, O., 2010. Seizure prediction and recall. Epilepsy Behav. 18 (1—2), 106—109. Frucht, M.M., Quigg, M., Schwaner, C., Fountain, N.B., 2000. Distribution of seizure precipitants among epilepsy syndromes. Epilepsia 41, 1534—1539. Giuccioli, D., Czuczwara, H., Finkler, J., Leitenberger, J., May, T., Nothbaum, N., Wolf, P., 1990. Epilepsie und Prodromi: eine prospektive Untersuchung. In: Epilepsie 89. Einhorn-Presse Verlag, Reinbeck, pp. 312—321. Haut, S.R., Hall, C.B., LeValley, A.J., Lipton, R.B., 2007a. Can patients with epilepsy predict their seizures? Neurology 68, 262—266. Haut, S.R., Hall, C.B., Masur, J., 2007b. Lipton RB seizure occurrence: precipitants and prediction. Neurology 69, 1905—1910. Hoppe, C., Poepel, A., Elger, C.E., 2007. Accuracy of patient seizure counts. Arch. Neurol. 64, 1595—1599. Hughes, J., Devinsky, O., Feldmann, E., Bromfield, E., 1993. Premonitory symptoms in epilepsy. Seizure 2, 201—203. Litt, B., Krieger, A., 2007. Of seizure prediction, statistics, and dogs: a cautionary tail. Neurology 68, 250—251. Maiwald, T., Winterhalder, M., Aschenbrenner-Scheibe, R., Voss, H.U., Schulze-Bonhage, A., Timmer, J., 2004. Comparison of three nonlinear seizure prediction methods by means of the seizure prediction characteristic. Physica D 194, 357—368. Maiwald, T., Blumberg, J., Timmer, J., Schulze-Bonhage, A., 2011. Are prodromes preictal events? A prospective PDA-based study. Epilepsy Behav. 21, 184—188. Mormann, F., Kreuz, T., Rieke, C., Andrzejak, R.G., Kraskov, A., David, P., Elger, C.E., Lehnertz, K., 2005. On the predictability of epileptic seizures. Clin. Neurophysiol. 116, 569—587. Murray, J., 1993. Coping with the uncertainty of uncontrolled epilepsy. Seizure 2, 167—178. Nakken, K.O., Solaas, M.H., Kjeldsen, M.J., et al., 2005. Which seizure-precipitating factors do patients with epilepsy most frequently report? Epilepsy Behav. 6, 85—89. Neugebauer, R., Paik, M., Hauser, W.A., et al., 1994. Stressful life events and seizure frequency in patients with epilepsy. Epilepsia 35, 336—343. Petitmengin, C., Baulac, M., Navarro, V., 2006. Seizure anticipation: are neurophenomenological approaches able to detect preictal symptoms? Epilepsy Behav. 9, 298—306. Rajna, P., Clemens, B., Csibri, E., Dobos, E., Geregely, A., Gottschal, M., Gyorgy, I., Horvath, A., Horvath, F., Mezofi, L., Velkey, I.,
235
Veres, J., Wagner, E., 1997. Hungarian multicentre epidemiologic study of the warning and initial symptoms (prodrome, aura) of epileptic seizures. Seizure 6, 361—368. Rajna, P., Sólyom, A., Mezofi, L., et al., 2008. Are there real unprovoked/unprecipitated seizures? Med. Hypotheses 71, 851—857. Scaramelli, A., Braga, P., Avellanal, A., Bogacz, A., Camejo, C., Rega, I., Messano, T., Arciere, B., 2009. Prodromal symptoms in epileptic patients: clinical characterization of the pre-ictal phase. Seizure 18, 246—250. Schelter, B., Winterhalder, M., Maiwald, T., Brandt, A., Schad, A., Schulze-Bonhage, A., Timmer, J., 2006. Testing statistical significance of multivariate time series analysis techniques for epileptic seizure prediction. Chaos, 16 (013108-1-013108-10). Schelter, B., Winterhalder, M., Drentrup, H.F., Wohlmuth, J., Nawrath, J., Brandt, A., Schulze-Bonhage, A., Timmer, J., 2007. Seizure prediction: the impact of long prediction horizons. Epilepsy Res. 73, 213—217. Schulze-Bonhage, A., Buller, A., 2008. Unpredictability of seizures and the burden of epilepsy. In: Schelter, B., Timmer, J., SchulzeBonhage, A. (Eds.), Seizure Prediction in Epilepsy: From Basic Mechanisms to Clinical Applications. Wiley, Berlin, pp. 1—10. Schulze-Bonhage, A., Kurth, C., Carius, A., Steinhoff, B.J., Mayer, T., 2006. Seizure anticipation by patients with focal and generalized epilepsy: a multicentre assessment of premonitory symptoms. Epilepsy Res. 70, 83—88. Schulze-Bonhage, A., Sales, F., Wagner, K., Teotonio, R., Carius, A., Schelle, A., Ihle, M., 2010. Views of patients with epilepsy on seizure prediction devices. Epilepsy Behav. 18, 388—396. Spector, S., Cull, C., Goldstein, L.H., 2000. Seizure precipitants and perceived self-control of seizures in adults with poorly controlled epilepsy. Epilepsy Res. 38, 207—216. Sperling, M.R., Schilling, C.A., Glosser, D., et al., 2008. Selfperception of seizure precipitants and their relation to anxiety level, depression, and health locus of control in epilepsy. Seizure 17, 302—307. Tatum 4th, W.O., Winters, L., Gieron, M., et al., 2001. Outpatient seizure identification: results of 502 patients using computerassisted ambulatory EEG. J. Clin. Neurophysiol. 18, 14—19. Taylor, D.C., 2007. Whatever happened to the epileptic prodrome? Epilepsy Behav. 11, 251—252. Winterhalder, M., Maiwald, T., Voss, H.U., Aschenbrenner-Scheibe, R., Timmer, J., Schulze-Bonhage, A., 2003. The seizure prediction characteristic: a general framework to assess and compare seizure prediction methods. Epilepsy Behav. 4, 318—325.
journal homepage: www.elsevier.com/locate/epilepsyres
Premonitory features and seizure self-prediction: Artifact or real? Andreas Schulze-Bonhage a,∗, Sheryl Haut b a b
Epilepsy Center, University Hospital, Freiburg, Germany Albert Einstein College of Medicine, New York, USA
Received 10 February 2011; received in revised form 4 August 2011; accepted 8 September 2011
KEYWORDS Epilepsy; Seizure prediction; Prodrome; Trigger factor
Summary Seizure prediction is currently largely investigated by means of EEG analyses. We here report on evidence available on the ability of epilepsy patients themselves to predict seizures either by means of subjective experiences (‘‘prodromes’’), apparent awareness of precipitants, or a feeling of impending seizure (self-prediction). These data have been collected prospectively by paper or electronic diaries. Whereas evidence for a predictive value of prodromes is missing, some patients nevertheless can forsee impending seizures above chance level. Relevant cues and practical implications are discussed. © 2011 Published by Elsevier B.V.
Epilepsy is characterized by the spontaneous and unprovoked occurrence of seizures. For patients and observers, these seizures appear to come out of the blue, which has considerable consequences for patients. Even if seizures occur quite rarely (e.g. a few times per year), the fact that their timing is unknown has legal consequences which hinder patients from driving motor vehicles and working in certain professions. Furthermore, spontaneous seizures may present physical risk during everyday activities such as bathing, or during sports, and may lead to social avoidance behaviour and a feeling of loss of control which may result in depression (Schulze-Bonhage and Buller, 2008). As a patient put it, ‘‘It is the unpredictability of it that is really nerve-racking to live with’’ (Murray, 1993).
∗
Corresponding author. E-mail address: [email protected] (A. Schulze-Bonhage). 0920-1211/$ — see front matter © 2011 Published by Elsevier B.V. doi:10.1016/j.eplepsyres.2011.09.026
Yet, from the earliest descriptions of epilepsy by Hippocrates, ‘‘warnings’’ experienced by patients prior to a visible seizure have been mentioned, and the concept of ‘‘prodromes’’ is found in textbooks on epilepsy suggesting that at least some patients experience precursors of seizures. Furthermore, patients often report that they have a sense of when their risk for seizure is high. So the question arises if certain subgroups of epilepsy patients might be able to predict their own seizures. In principle, seizure predictions of patients could be based on two sources: subjective experiences of ‘‘warnings’’ preceding a seizure, and knowledge of factors which increase the probability of a subsequent seizure occurrence. Warnings can be divided into an epileptic aura which may precede a focal seizure by seconds to minutes, and into subjective feelings experienced longer before a seizure, so-called ‘‘prodromes’’. Auras can be very useful to patients, if they allow them to retreat from public or potential dangerous situations or even to take counter-measures
232 against seizures; they are considered as part of the ictal event, however, so that the experiences during an aura can be considered as subjective detection of a seizure, but not as a prediction. Prodromes, on the other hand, are believed to be nonictal events, i.e. events which are not accompanied by ictal discharges in the EEG. Generally, they are considered to precede a seizure by longer time periods of hours to days, and their physiological background has remained obscure so far. Prodromes may thus be preictal events based on which patients might be able to predict an impending seizure. Aside from such subjective experiences, patients may have knowledge of trigger factors or precipitants of their seizures, upon which they may base a prediction of an impending seizure. It has been hypothesized that true unprovoked seizures, unrelated to precipitants, may actually be rare (Rajna et al., 2008). Such precipitating factors could comprise aspects temporal effects such as circadian or catamenial patterns; acute precipitants such as sleep deprivation, stress or other emotional factors, or simply non-compliance with medication intake. Awareness of any of these factors could lead to valid predictions of increased seizure susceptibility at certain periods of time. Methodologically, the field of clinical prediction of seizures depends on reliable patient report. Examination of the relevant studies mandates acknowledgement that diary formats differ significantly. Questionnaires are commonly used to identify premonitory features and precipitants reported by patients in a cross-sectional manner. This format, while useful, is subject to recall bias, and the findings must be tested prospectively. Prospective paper diaries have been widely used, but are limited by the lack of time stamping, and the risk of backfilled or retrospectively entered data. Electronic diaries are increasingly utilized in studies of premonitory features and seizure prediction, but present other challenges of data transfer and increased cost. In the following, studies analysing prodrome- and knowledge-based seizure predictions are addressed separately.
Prodromes and seizure prediction Prodromes have been the subject of descriptions in the literature (Dostoyevsky, 1868) and of scientific investigations. Most publications used questionnaires in which patients or at times also caregivers were asked if they would experience changes in their subjective perception or in their behaviour. Depending on the type of question and on the population investigated, rates of 6.9% up to 39% of patients with prodromes were reported (Giuccioli et al., 1990; Hughes et al., 1993; Rajna et al., 1997; Schulze-Bonhage et al., 2006; Scaramelli et al., 2009). A separation from auras was performed in most studies by defining a time interval before an upcoming seizure, which was mostly set as more than 30 min (Schulze-Bonhage et al., 2006), rarely also only more than 5 min before seizure onset. One study analyzed if the phenomenology of ‘‘prodromal’’ symptoms was similar to or different from auras; out of 15 patients who were able to describe their experiences with sufficient detail, 12 reported clearly different perceptions during the periods far from a seizure as compared to auras directly progressing
A. Schulze-Bonhage, S. Haut into a seizure (Schulze-Bonhage et al., 2006). Four studies analyzed the dependency of a prodrome on the classification of epilepsy; there was a concordantly higher frequency of prodromes in structural as compared to genetic epilepsy (Hughes et al., 1993; Rajna et al., 1997; Schulze-Bonhage et al., 2006; Scaramelli et al., 2009). The exact etiology in cases of structural epilepsy did not appear to play a central role (Rajna et al., 1997). Based on questionnaires, the phenomenology of subjective experiences during ‘‘prodromes’’ was reported. There was considerable variability between the statements of individual patients regarding their prodrome-sensation. Mostly, vegetative symptoms (e.g. palpitations, sweating, gastrointestinal symptoms), or emotional disturbances (e.g. irritability fatigue, anxiety, depression) are reported. Notably when summarizing the range of symptoms reported, there is wide variability between studies, and an impressively wide spectrum of possible complaints is reported (Table 1). The validity of prodromes as seizure precursors has recently been questioned. Methods used for an analysis of EEG-based seizure prediction methods (AschenbrennerScheibe et al., 2003; Maiwald et al., 2004; Mormann et al., 2005; Winterhalder et al., 2003; Schelter et al., 2006, 2007) generally were not applied to evaluate the validity and prediction performance of prodromes. Taylor (2007) accordingly pointed out that there are no studies which prove that prodromes are preictal events, and has put forward the hypothesis that prodromal experiences may erroneously be considered as seizure-related when epilepsy is considered to be a disease essentially consisting of seizures, and that they might be seizure-independent symptoms related to the neurobiological background manifesting in both, seizures and independent alterations in subjective experiences. At the Freiburg epilepsy center, a prospective study using handheld computers was performed which intended to identify patients who were able to predict their own seizures. Participants were recruited from a multicentric assessment in 500 patients to identify the subgroup convinced to experience seizure precursors (Schulze-Bonhage et al., 2006). Patients with a minimum seizure frequency of 1/month were assessed prospectively to state if they experienced a prodrome every 12 h, and they were asked to perform free entries whenever a prodrome or a seizure happened. Prediction performance was assessed using a methodology developed for EEG-based seizure prediction algorithms (Winterhalder et al., 2003). Data entries into the handheld occurred patient-initiated at any time, and at standardized points of time every 12 h according to an alarm given by the handheld to the patient. Patients entered prodromes and their type as well as seizures and indicated when either had occurred. Out of nine patients in whom at least 4 weeks of continuous entries were available, none had a predictive performance which was statistically better than to be expected from a random predictor, even when the false prediction rate was chosen according to the patient’s performance and when various seizure occurrence periods up to 24 h were analyzed (Maiwald et al., 2011). It is of interest that in this study not only the time of seizures and prodromes as reported by the patient was stored but also data entry times were time stamped. An analysis of time stamps of entries for seizures and prodromes
Premonitory features and seizure self-prediction: Artifact or real? Table 1
233
Premonitory symptoms as reported in studies on prodromes.
Giuccioli et al. (book art. 1990) n = 53
Hughes et al. Seizure (1993) n = 148
Rajna et al. Seizure (1997) n = 262
Schulze Bonhage et al. Epil Res (2006) n = 35
Scaramelli et al. Seizure (2009) n = 39
Absent-mindedness Cold hands/feet Depression Dizziness Fatigue Fearfulness Headache Heat waves Introversion Irritation Palpitations Polyuria Sleeplessness Speech disturbance Sweating Tiredness
Anxiety Confusion Depression Elation Fear ‘‘Funny feeling‘‘ Headache Irriability Menstrual symp. Polyuria Speech disturbance
Epigastric sensation ‘‘Funny feeling’’ Headache
Dizziness Headache Impaired concentration Malaise Nausea Restlessness Tiredness
Appetite change ‘‘Behavioural’’ change Cognitive disturbance Dysthermia Fatigue GI symptoms Headache Paresthesias Pain Sleep disturbance Speech disturbance Voiding changes
Note: High variability with symptoms related to pain, vegetative signs, changes in mood and cognitive impairments.
showed that some ‘‘prodromes’’ were entered only at a time when a seizure had already happened and retrospectively were reported as pre-ictal events according to the time when the patient stated it to have happened. Although this may be called ‘‘cheating’’, it may just correspond to the lack of specificity of the experiences which based on their phenomenology cannot be regarded as clear preictal phenomena. Only when a seizure follows these experiences, they may be retrospectively classified as a preictal sensation, conceptually reflected by the term ‘‘prodrome’’. As these experiences may not be entered as prodromes when no seizure follows, a statistical bias in favour of specificity can be the consequence of such retrospective reporting.
Seizure precipitants and self-prediction In the general epilepsy population, up to 90% of patients with epilepsy identify at least one seizure precipitant (Neugebauer et al., 1994; Frucht et al., 2000; Spector et al., 2000; Nakken et al., 2005; Sperling et al., 2008). Emotional stress and stressful life events, sleep deprivation, depression, anxiety, menstruation, and alcohol use are among the most commonly reported seizure precipitants. It seems likely that the presence of precipitants may make patients aware of the risk of an impending seizure. What follows is that if self-prediction of seizures is possible, at least part of this predictive ability is related to this awareness. One study addressed this possibility prospectively in a paper diary design (Haut et al., 2007a,b). Subjects provided data on potential precipitants which have been identified in other studies including hours of sleep, levels of stress and anxiety, menstruation and alcohol, on a daily basis. Seizure self-prediction was assessed by the following question, also completed on a daily basis: Do you think you will have a seizure in the next 24 h? Response options included extremely likely (1), somewhat likely (2), somewhat unlikely
(3) and extremely unlikely (4). Of note, premonitory feature report was not included in this study. Seventy-one subjects returned a total of 15,179 days followed with information regarding seizure occurrence the next day. Of these, fifty-seven subjects (80%) experienced at least one seizure during the period of diary followup. Overall sensitivity for the population was 31.9%, in that 475 of 1488 seizure days had an antecedent prediction of ‘‘extremely likely’’ or ‘‘somewhat likely’’. For the 13,691 seizure-free days for which there was a prediction, 11,388 had been rated ‘‘somewhat unlikely’’ or ‘‘extremely unlikely’’, yielding a population specificity of 83.2%. Adjusting for heterogeneity across subjects, an overall specificity of 87% and sensitivity of 21% were obtained. The Cochran—Mantel—Haenzel OR for positive prediction was 2.25 (1.91—2.65) indicating that seizures were twice as likely in the 24 epoch following a prediction. Noting the change in sensitivity with adjustment across subjects, the predictive ability of individuals was explored. Twelve subjects demonstrated significant individual predictability and were termed ‘‘predictors’’. The Mantel—Haenzel estimate of the OR for the predictor group was 3.14 (2.53—3.89), indicating that following a positive prediction, the subject was more than three times more likely to experience a seizure. Accuracy of seizure prediction for the subjects with significant predictive ability, adjusted for individual contribution, is presented in Fig. 1 (Haut et al., 2007a). In order to examine the factors related to this ability to predict an impending seizure, data on the potential precipitants were entered into predictive models (Haut et al., 2007b). Precipitants remaining significant in the model included reported levels of stress, anxiety and hours of sleep for the night prior to the seizure. When patient self seizureprediction was included, variables which remained in the model included self-prediction (OR 3.7; 95% CI 1.8—7.2), and hours of sleep for the night prior to the seizure (OR .90; 95% CI 82—.99). The fact that self-reported stress and
234
A. Schulze-Bonhage, S. Haut
Accuracy of prediction
Seizure No seizure
100% 80% 60% 40% 20% 0% extremely likely
somewhat likely
somewhat unlikely
extremely unlikely
Seizure prediction
Figure 1 Accuracy of seizure predictors. PPV and NPV values for the subgroup (n = 12) of subjects demonstrating significant seizure prediction. Values are adjusted for within person correlation. Table 2
Hypothetical scenario based on precipitant report.
Stress
Anxiety
Hours of sleep
Risk of seizure %
Low Moderate High
Low Moderate High
High Moderate Low
4.1 5.9 8.5 Prediction
Any Any Any
Any Any Any
High Moderate Low
Low Moderate High
2.9 6.7 13.1
anxiety levels did not remain significant in the model once self-prediction was included led to the conclusion that subjects felt most likely to experience a seizure when they were reporting greater stress and anxiety. A number of hypothetical scenarios (Table 2) were calculated for a 40-year old female subject. In this scenario, the highest risk of seizure is modeled with report of fewer hours of sleep and high certainty of self-prediction. As discussed in an accompanying editorial (Litt and Krieger, 2007), this study was limited by the paper diary format. The study was then repeated with electronic diaries and extended sampling strategies including report of premonitory features, and more extensive mood measurements. We anticipate that these data will further clarify the accuracy of modeling seizure occurrence from patient reported data. Seizure self-prediction has been similarly reported in a recent inpatient study (Dubois et al., 2010). In this study, 83 subjects provided self-predictions of impending seizures during video-EEG monitoring. Following a positive prediction, the probability of a seizure was 0.320 (95% CI: 0.149—0.558), significantly increased over the probability of a seizure (0.151, 95% CI: 0.71—0.22) following a negative prediction.
Discussion The ability of patients to self-predict seizures has so far remained a matter of controversy. The recent investigation of Maiwald et al. (2011) suggests that subjective experiences of patients do not have a predictive value for seizures to
follow within a period of 1—24 h. The practical value of prodromes for seizures had already been called into question by Petitmengin et al. (2006) who pointed out that the subjective sensations called prodromes may be of limited value in practical life due to the lack of specificity as seizure precursors. The PDA-based prospective analysis of Maiwald et al. (2011) showed that predictions based on prodromes were not superior to random predictions taking into consideration relatively long seizure occurrence periods. It remains an open question as to whether the subjective sensations of patients are insufficient as seizure predictors as they are in fact unrelated to subsequent seizures or only due to the fact that they are not specific enough in character to be identified as preictal events. It has to be stated, that thus far, a prodrome-based predictability of seizures has not been proven. Whereas the majority of seizures that occur at present are not clearly predicted by patients, the recent studies by Haut et al. and others suggest that a subgroup of patients may be aware of impending seizures. These patients appear to be able to use knowledge about trigger factors to predict seizures. Further analyses on the factors relevant for these predictions, particularly seizure triggers or precipitants, may result in a more quantitative, individualized approach and may be helpful for patients in their everyday life. Patients may not recognize all of their seizures, and in fact, the accuracy of seizure diaries and seizure self-report continues to be debated (Blum et al., 1996; Tatum et al., 2001; Hoppe et al., 2007). Methodologically, the paucity of studies investigating prodromes and prediction in relation to EEG-confirmed seizures introduces the possibility that seizures are underreported, and that the relationship between prediction and seizure occurrence may be overestimated. Dubois et al. (2010) however recently demonstrated that patient awareness of seizures was high. This issue will remain critical for planning interventional studies. In a practical sense, the value of patient-based seizure prediction at the currently reported sensitivities is limited. Even for patients able to predict their seizures above chance level, sensitivity is below 50%, which is lower than the requirements suggested for future EEG-based prediction methods (Schulze-Bonhage et al., 2010). It is an interesting question, however, whether education and training may improve the performance of patient-based predictions compared to the naive state, in which both studies on prodromes and on unspecified predictions were performed. If so, this might allow both behavioural and possibly pharmacologic interventions, including extra precautions and temporally targeted interventions during periods of high risk.
References Aschenbrenner-Scheibe, R., Maiwald, T., Winterhalder, M., Voss, H.U., Timmer, J., Schulze-Bonhage, A., 2003. How well can epileptic seizures be predicted? An evaluation of a nonlinear method. Brain 126, 2616—2626. Blum, D.E., Eskola, J., Bortz, J.J., Fisher, R.S., 1996. Patient awareness of seizures. Neurology 47, 260—264. Dostoyevsky, F., 1868. The Idiot, Part II, 5.
Premonitory features and seizure self-prediction: Artifact or real? Dubois, J.M., Boylan, L.S., Shiyko, M., Barr, W.B., Devinsky, O., 2010. Seizure prediction and recall. Epilepsy Behav. 18 (1—2), 106—109. Frucht, M.M., Quigg, M., Schwaner, C., Fountain, N.B., 2000. Distribution of seizure precipitants among epilepsy syndromes. Epilepsia 41, 1534—1539. Giuccioli, D., Czuczwara, H., Finkler, J., Leitenberger, J., May, T., Nothbaum, N., Wolf, P., 1990. Epilepsie und Prodromi: eine prospektive Untersuchung. In: Epilepsie 89. Einhorn-Presse Verlag, Reinbeck, pp. 312—321. Haut, S.R., Hall, C.B., LeValley, A.J., Lipton, R.B., 2007a. Can patients with epilepsy predict their seizures? Neurology 68, 262—266. Haut, S.R., Hall, C.B., Masur, J., 2007b. Lipton RB seizure occurrence: precipitants and prediction. Neurology 69, 1905—1910. Hoppe, C., Poepel, A., Elger, C.E., 2007. Accuracy of patient seizure counts. Arch. Neurol. 64, 1595—1599. Hughes, J., Devinsky, O., Feldmann, E., Bromfield, E., 1993. Premonitory symptoms in epilepsy. Seizure 2, 201—203. Litt, B., Krieger, A., 2007. Of seizure prediction, statistics, and dogs: a cautionary tail. Neurology 68, 250—251. Maiwald, T., Winterhalder, M., Aschenbrenner-Scheibe, R., Voss, H.U., Schulze-Bonhage, A., Timmer, J., 2004. Comparison of three nonlinear seizure prediction methods by means of the seizure prediction characteristic. Physica D 194, 357—368. Maiwald, T., Blumberg, J., Timmer, J., Schulze-Bonhage, A., 2011. Are prodromes preictal events? A prospective PDA-based study. Epilepsy Behav. 21, 184—188. Mormann, F., Kreuz, T., Rieke, C., Andrzejak, R.G., Kraskov, A., David, P., Elger, C.E., Lehnertz, K., 2005. On the predictability of epileptic seizures. Clin. Neurophysiol. 116, 569—587. Murray, J., 1993. Coping with the uncertainty of uncontrolled epilepsy. Seizure 2, 167—178. Nakken, K.O., Solaas, M.H., Kjeldsen, M.J., et al., 2005. Which seizure-precipitating factors do patients with epilepsy most frequently report? Epilepsy Behav. 6, 85—89. Neugebauer, R., Paik, M., Hauser, W.A., et al., 1994. Stressful life events and seizure frequency in patients with epilepsy. Epilepsia 35, 336—343. Petitmengin, C., Baulac, M., Navarro, V., 2006. Seizure anticipation: are neurophenomenological approaches able to detect preictal symptoms? Epilepsy Behav. 9, 298—306. Rajna, P., Clemens, B., Csibri, E., Dobos, E., Geregely, A., Gottschal, M., Gyorgy, I., Horvath, A., Horvath, F., Mezofi, L., Velkey, I.,
235
Veres, J., Wagner, E., 1997. Hungarian multicentre epidemiologic study of the warning and initial symptoms (prodrome, aura) of epileptic seizures. Seizure 6, 361—368. Rajna, P., Sólyom, A., Mezofi, L., et al., 2008. Are there real unprovoked/unprecipitated seizures? Med. Hypotheses 71, 851—857. Scaramelli, A., Braga, P., Avellanal, A., Bogacz, A., Camejo, C., Rega, I., Messano, T., Arciere, B., 2009. Prodromal symptoms in epileptic patients: clinical characterization of the pre-ictal phase. Seizure 18, 246—250. Schelter, B., Winterhalder, M., Maiwald, T., Brandt, A., Schad, A., Schulze-Bonhage, A., Timmer, J., 2006. Testing statistical significance of multivariate time series analysis techniques for epileptic seizure prediction. Chaos, 16 (013108-1-013108-10). Schelter, B., Winterhalder, M., Drentrup, H.F., Wohlmuth, J., Nawrath, J., Brandt, A., Schulze-Bonhage, A., Timmer, J., 2007. Seizure prediction: the impact of long prediction horizons. Epilepsy Res. 73, 213—217. Schulze-Bonhage, A., Buller, A., 2008. Unpredictability of seizures and the burden of epilepsy. In: Schelter, B., Timmer, J., SchulzeBonhage, A. (Eds.), Seizure Prediction in Epilepsy: From Basic Mechanisms to Clinical Applications. Wiley, Berlin, pp. 1—10. Schulze-Bonhage, A., Kurth, C., Carius, A., Steinhoff, B.J., Mayer, T., 2006. Seizure anticipation by patients with focal and generalized epilepsy: a multicentre assessment of premonitory symptoms. Epilepsy Res. 70, 83—88. Schulze-Bonhage, A., Sales, F., Wagner, K., Teotonio, R., Carius, A., Schelle, A., Ihle, M., 2010. Views of patients with epilepsy on seizure prediction devices. Epilepsy Behav. 18, 388—396. Spector, S., Cull, C., Goldstein, L.H., 2000. Seizure precipitants and perceived self-control of seizures in adults with poorly controlled epilepsy. Epilepsy Res. 38, 207—216. Sperling, M.R., Schilling, C.A., Glosser, D., et al., 2008. Selfperception of seizure precipitants and their relation to anxiety level, depression, and health locus of control in epilepsy. Seizure 17, 302—307. Tatum 4th, W.O., Winters, L., Gieron, M., et al., 2001. Outpatient seizure identification: results of 502 patients using computerassisted ambulatory EEG. J. Clin. Neurophysiol. 18, 14—19. Taylor, D.C., 2007. Whatever happened to the epileptic prodrome? Epilepsy Behav. 11, 251—252. Winterhalder, M., Maiwald, T., Voss, H.U., Aschenbrenner-Scheibe, R., Timmer, J., Schulze-Bonhage, A., 2003. The seizure prediction characteristic: a general framework to assess and compare seizure prediction methods. Epilepsy Behav. 4, 318—325.