- Email: [email protected]
Predictive risk factors of seizure-related injury in persons with epilepsy
Journal of the Neurological Sciences 285 (2009) 59–61
Contents lists available at ScienceDirect
Journal of the Neurological Sciences j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j n s
Predictive risk factors of seizure-related injury in persons with epilepsy Somsak Tiamkao a,⁎, Kittisak Sawanyawisuth a, Thanin Asawavichienjinda b, Prapun Yaudnopakao c, Suwanna Arunpongpaisal d, Warinthorn Phuttharak d, Narong Auevitchayapat d, Suda Vannaprasaht d, Siriporn Tiamkao d, Kutcharin Phunikhom d, Aporanee Chaiyakum d, Jiamjit Saengsuwan d, Suthipun Jitpimolmard d a
Department of Medicine, Srinagarind Hospital, Faculty of Medicine, Khon Kaen University, Khon Kaen, Thailand Maharaj Nakhon Ratchasima Hospital, Nakhon Ratchasima, Thailand c Surin Hospital, Surin, Thailand d Epilepsy Research Group, Srinagarind Hospital, Faculty of Medicine, Khon Kaen University, Khon Kaen, Thailand b
a r t i c l e
i n f o
Article history: Received 18 March 2009 Received in revised form 13 May 2009 Accepted 19 May 2009 Available online 10 June 2009 Keywords: Seizure-related injury Predictive Model Online Epilepsy
a b s t r a c t Objective: The clinical risk factors for seizure-related injuries (SRI) in adult persons with epilepsy (PWE) were studied and analyzed to develop a predictive model. Methods: We enrolled 300 consecutive cases from three epilepsy clinics in Northeast, Thailand. Subjects were eligible if reported to have at least one seizure attack during the past 12 months. Face-to-face questionnaire was used to evaluate SRI, baseline characteristics and other seizure-related variables. Results: There were 247 and 91 PWE who met a criterion and had SRI, respectively. By multivariate logistic regression method, GTC seizure type, having history of seizure attacks at least 12 times/year, and daytime seizure were significant risk factors of having SRI with odds ratio of 2.376, 2.460, and 3.562, respectively. We developed the predictive model for having SRI in PWE and it gave 90.3% sensitivity and 46.7% specificity on the occurrence of SRI. The estimated probability of SRI can be found online at http://sribykku.webs.com/. Conclusions: The significant predictive factors for SRI in PWE were the occurrence of GTCs, seizures at least 12 times/year or daytime seizures. Clinicians or PWE can easily evaluate the risk of having SRI in individuals by the online predictive model. © 2009 Elsevier B.V. All rights reserved.
1. Introduction Epilepsy is considered as one of the major public health problems in the world [1]. As other diseases, many studies showed that it contributes to poor quality of life. The impact of epileptic seizures will be more if the seizures are uncontrolled and increase the risk of injuries. Seizure-related injuries (SRI) are common and reported with a wide range of events. It can be just minor trauma to intracranial hematoma [2], extensive severe burns [3] or even death [4]. The European Cohort study showed that 21% of persons with epilepsy (PWE) reported accidents compared to 14% of control subjects [5]. Among the reported injuries, the relative risk for concussion was the highest at 2.6. The risk of trauma was also related to seizure type [3] and the frequency of seizure was the predictive factor of seizure-related injury [6]. These bring up an overprotecting
⁎ Corresponding author. Division of Neurology, Department of Medicine, Faculty of Medicine, Khon Kaen University, Khon Kaen, 40002, Thailand. Tel.: +66 43 363 664; fax: +66 43 374 542. E-mail address: [email protected] (S. Tiamkao). 0022-510X/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.jns.2009.05.021
issue in PWE that may result in social isolation, physical inactivity and psychological dependency. Here, we did a study to evaluate the risk and predictive factors for having SRI in PWE by logistic regression analysis. In addition, we would like to create an online predictive model to assess the risk of having SRI in individuals. Therefore, PWE or clinicians can assess their own risks for SRI electronically. 2. Methods 2.1. Study population We studied 300 consecutive adult PWE at the out-patient clinic at three provincial hospitals in the Northeast, Thailand in 2004. The inclusion criterion was epileptic patients who reported at least one seizure attack during the past 12 months. We used a face-to-face standardized questionnaire by one of the authors to acquire the risk factors for SRI. The questionnaire was used and reported elsewhere [7]. The questionnaire contents included demographics, age of onset and type of seizure, apparent causes, history of falling during seizure attack, seizure attack time whether in daytime or nighttime, total number of seizures during the past 12 months, current numbers of
60
S. Tiamkao et al. / Journal of the Neurological Sciences 285 (2009) 59–61
antiepileptic drugs (AEDs), and the SRI events. The SRI was identified by the occurrence of traffic accidents, near drowning, burns, fractures, joint dislocations, head traumas, or soft tissue injuries during seizure attacks in the past 12 months. The investigators attended questionnaire interview trainings prior to the beginning of the study. The study protocol was approved by the institutional ethical committee. All participants signed the informed consent prior to the beginning of the study. 2.2. Data analysis We analyzed data by using the statistical function of the SAS software version 8.2 (SAS Institute Inc., Cary, NC, USA). Descriptive statistic was used to show the difference between those who had or did not have SRI. The numerical variables including age, age onset of epilepsy, and duration of having epilepsy were presented in median and range. The categorical variables were shown in number of patients (%) including number of male patients and number of one who had generalized tonic–clonic seizures (GTCs), had seizure attacks at least 12 times a year, had daytime or nighttime seizure, had history of falling during seizure attack, and had AEDs from none to 3 or more. The Wilcoxon rank sum test and Chi square test were used to compare differences in numbers and proportions, respectively, of the 2 groups. Difference was considered significant if p b 0.05. We tested all independent variables by univariate logistic analysis. Predictive model was executed using a multivariate logistic regression model. Possible explanatory variables were included and fit a backward elimination method to predict the occurrence of SRI. The retention criterion for the final model was 0.10. Data presented in crude odds ratio (OR), adjusted odds ratio, and 95% confidence interval (C.I.). The Hosmer–Lemeshow goodness of fit statistic was used to assess the fitness of the final model [8]. To demonstrate the discrimination of the model, c statistics or the area under the receiver operating characteristic (ROC) curve was calculated [9]. 3. Results There were 247 of 300 cases who had seizure attack in the past 12 months. Of those, 91 (36.84%) had SRI. The differences of baseline characters between those who had or did not have SRI were shown in Table 1. There was no difference between both groups in genders, age, age onset of epilepsy, duration of having epilepsy, and numbers of cases with nighttime seizure. SRI group had more GTCs cases and
Table 1 Baseline characteristics between no seizure-related injury (no SRI) and seizure-related injuries (SRI) group. Variable
No SRI N = 156
SRI N = 91
Male gender Age (years), median (range) Age onset of epilepsy (years), median (range) Duration of having epilepsy (years), median (range) Seizure type; GTCsa Seizure attacks ≥12 times/year Daytime seizure Nighttime seizure History of fallingb Number of antiepileptic drugs 0 1 2 ≥3
70 (44.9) 34.0 (13–79) 21 (0–78) 6 (0–55)
45 (49.5) 33 (16–91) 21 (0–90) 8 (0–44)
30 (19.2) 39 (25.0) 119 (76.3) 93 (59.6) 117 (75.0)
31 45 84 45 83
(34.1) (49.5) (92.3) (50.0) (91.2)
5 (3.2) 112 (71.8) 34 (21.8) 5 (3.2)
2 (2.2) 49 (53.9) 33 (36.3) 7 (7.7)
Table 2 Crude odds ratio for developing seizure-related injuries by univariate analysis. Variable
Crude OR
95% confidence interval
p value
Male gender Age (years) Age onset of epilepsy (years) Duration of having epilepsy (years) Seizure type; GTCsa Seizure attacks ≥ 12 times/year Daytime seizure Nighttime seizure History of fallingb Number of antiepileptic drugs 0 1 2 ≥3
1.202 1.002 1.000 1.002 2.170 2.934 3.731 0.677 3.458 1.906 1.000 1.094 2.426 3.400
0.716 0.984 0.985 0.980 1.205 1.696 1.587 0.402 1.537 1.242
2.018 1.020 1.015 1.025 3.909 5.075 8.770 1.142 7.782 2.927
0.4867 0.8677 0.9903 0.8586 0.0099 0.0001 0.0025 0.1440 0.0027 0.0032
0.205 0.440 0.473
5.832 13.392 25.892
0.9165 0.3092 0.2200
Note. OR; odds ratio, aGTCs; generalized tonic–clonic seizures, bHistory of falling during seizure attack.
higher number of patients who had seizure attacks, daytime seizures, history of falling during seizure attacks, and number of AEDs. The univariate analysis (Table 2) showed that the independent significant factors related to have SRI were GTCs (OR 2.170 95% C.I. 1.205–3.909), total seizure attacks at least 12 times a year (OR 2.934, 95% C.I. 1.696–5.075), daytime seizure (OR 3.731, 95% C.I. 1.587– 8.770), having history of falling during seizure attack (OR 3.458, 95% C. I. 1.537–7.782), and number of AEDs (OR 1.906, 95% C.I. 1.242–2.927). By the multivariate logistic regression method, there were five remaining variables but only three variables were statistically significant in the final model (Table 3). GTC seizure type, having history of seizure attacks at least 12 times/year, and daytime seizure had the risk of having SRI 2.376, 2.460, and 3.562 times, respectively. The Hosmer–Lemeshow goodness-of-fit test statistic of the final multivariate logistic model was 13.6 (p = 0.09). The area under ROC curve or c statistic of the final model was 0.715. The sensitivity and specificity of the final model were 90.3% and 46.7%, respectively. The predictive model of having SRI can be found below Table 3 or online at http://sribykku.webs.com/. 4. Discussion We previously reported that soft tissue injury was the most common type of injury reported in the U.K. and the Northeast, Thailand [7,10]. The common sites of injury were head, face or arms. Nearly all reported injuries were mild. Similar results were also reported in Europe by the European Cohort [5,11]. The risk of SRI is of great concern to both physicians and PWE for the prevention of possible injury.
p Value 0.4865 0.8923 0.9615 0.2879 0.0091 b0.0001 0.0015 0.1433 0.0017 0.0198
Note. Data are number (%) of patients, unless otherwise indicated. aGTCs; generalized tonic–clonic seizures, bHistory of falling during seizure attack. Data for no SRI and SRI group may not total 156 and 90, respectively, because of missing data.
Table 3 Adjusted odds ratio for developing seizure-related injuries of all remaining covariates in the final model by multiple logistic regression analysis. Variable
Adjusted OR
95% confidence interval
p value
Age Numbers of antiepileptic drugs Seizure type; GTCsa Seizure attacks ≥ 12 times/year Daytime seizure
1.018 1.554 2.376 2.460 3.562
0.998 0.959 1.257 1.329 1.433
0.0846 0.0739 0.0077 0.0042 0.0063
1.039 2.485 4.493 4.552 8.852
Note. aGTCs; generalized tonic–clonic seizures, OR; odds ratio. The predicted probability of having SRI (P) can be calculated for any given patient: P (occurrence of SRI) = 1 / [1 + exp (Z)]. Z equal 3.4096 − (0.0177 × age in years) − (0.4341 × no. AEDs) − (0.8655 × GTCs) − (0.9000 × frequency) − (1.2702 × daytime seizure), where numbers of AEDs = 0 if no. AEDs taken, 1 if one AED taken, 2 if two AEDs taken, 3 if three or more AEDs taken; GTCs = 1 if seizure type is GTCs, 0 if not; frequency = 1 if average seizure attack more than 1 event/month, 0 if not; daytime seizure = 1 if having daytime seizure, 0 if not; exp = exponential.
S. Tiamkao et al. / Journal of the Neurological Sciences 285 (2009) 59–61
The significant independent variables in the final logistic regression were the occurrence of GTCs, seizures at least 12 times/year and daytime seizures. As previously reported, frequent seizure attacks enhanced chances of having SRI [6]. From our data, if epileptic patients had seizure attacks approximately once per month, the risk of having SRI increased by 246 folds. GTC type of seizure introduced unconsciousness and aggravated SRI 2.3 times compared with non-GTC type. Finally, the daytime seizure had the highest risk for developing SRI. Daytime period has much more activities than nighttime period. Some activities such as driving, transportation or recreation activities contribute directly to SRI. The predicted probability of the occurrence of SRI for an individual was shown below Table 3 and online at http://sribykku.webs.com/. This predictive model was created by 5 potential clinical variables remained in the final model; it will be helpful in clinical practice to identify a risk of SRI in PWE. Even though age and numbers of AEDs were not statistically significant in the final model, they included in the predictive model due to their clinical relevance to SRI and the borderline p values (0.08 and 0.07, respectively). The goodness-of-fit of the final model was shown by the Hosmer and Lemeshow method (p value 0.09). The c statistic value or the area under ROC curve of 0.715 suggested a fair discrimination effect of the final model. The predictive model had a good prediction of the occurrence of SRI by 90.3%. This high sensitivity predictive model can be a good tool for screening or ruling out SRI. Clinicians or PWE will know confidently the risk for SRI in individuals. Therefore, low predictive risk patients may not anymore be restricted to some risky activities such as driving or recreation activities. The limitations of our study were interviewer bias, overestimated bias, and recall bias. Participating subject and the interviewer knew each other. To reduce the interview bias, the investigator trainings were arranged prior to the actual study. PWE should be encouraged to live as normal a life as possible if epilepsy is controllable. To avoid fear of doing some activities or
61
overprotection, knowing their own risks for SRI is crucial. With our high sensitivity of the predictive model, physicians can clinically judge PWE who are at risk for the occurrence of SRI.
Acknowledgement The authors would like to thank Professor Simon D. Shorvon for the initiation of the study and Dr. Chatlert Pongchaiyakul for helping in statistical analysis.
References [1] World Health Organization. Epilepsy: aetiology, epidemiology and prognosis; February 2001. Published online: Available from: http://www.who.int/mediacentre/factsheets/fs165/en/. [2] Neufeld MY, Vishne T, Chistik V, Korczyn AD. Life-long history of injuries related to seizures. Epilepsy Res 1999;34:123–7. [3] Unglaub F, Woodruff S, Demir E, Pallua N. Patients with epilepsy: a high-risk population prone to severe burns as a consequence of seizures while showering. J Burn Care Rehabil 2005;26:525–8. [4] Nei M, Bagla R. Seizure-related injury and death. Curr Neurol Neurosci Rep 2007;7:335–41. [5] Van den Broek M, Beghi E, REST-1 Group. Accidents in patients with epilepsy: types, circumstances, and complications; a European cohort study. Epilepsia 2004;45:667–72. [6] Lawn ND, Bamlet WR, Radhakrishnan K, O'Brien PC, So EL. Injuries due to seizure in persons with epilepsy: a population-based study. Neurology 2004;63:1565–70. [7] Tiamkao S, Shorvon SD. Seizure-related injury in an adult tertiary epilepsy clinic. Hong Kong Med J 2006;12:260–3. [8] Hosmer DW, Hosmer T, Le Cessie S, Lemeshow S. A comparison of goodness-of-fit tests for the logistic regression model. Stat Med 1997;16:965–80. [9] Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143:29–36. [10] Tiamkao S, Amornsin O, Pongchaiyakul C, Asawavichienjinda T, Yaudnopakao P, Jitpimolmard S, et al. Seizure-related injuries in Northeast Thailand. J Med Assoc Thai 2006;89:608–13. [11] Cornaggia CM, Beghi M, Moltrasio L, Beghi E, RESt-1 Group. Accidents at work among people with epilepsy. Results of a European prospective cohort study. Seizure 2006;15:313–9.
Contents lists available at ScienceDirect
Journal of the Neurological Sciences j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j n s
Predictive risk factors of seizure-related injury in persons with epilepsy Somsak Tiamkao a,⁎, Kittisak Sawanyawisuth a, Thanin Asawavichienjinda b, Prapun Yaudnopakao c, Suwanna Arunpongpaisal d, Warinthorn Phuttharak d, Narong Auevitchayapat d, Suda Vannaprasaht d, Siriporn Tiamkao d, Kutcharin Phunikhom d, Aporanee Chaiyakum d, Jiamjit Saengsuwan d, Suthipun Jitpimolmard d a
Department of Medicine, Srinagarind Hospital, Faculty of Medicine, Khon Kaen University, Khon Kaen, Thailand Maharaj Nakhon Ratchasima Hospital, Nakhon Ratchasima, Thailand c Surin Hospital, Surin, Thailand d Epilepsy Research Group, Srinagarind Hospital, Faculty of Medicine, Khon Kaen University, Khon Kaen, Thailand b
a r t i c l e
i n f o
Article history: Received 18 March 2009 Received in revised form 13 May 2009 Accepted 19 May 2009 Available online 10 June 2009 Keywords: Seizure-related injury Predictive Model Online Epilepsy
a b s t r a c t Objective: The clinical risk factors for seizure-related injuries (SRI) in adult persons with epilepsy (PWE) were studied and analyzed to develop a predictive model. Methods: We enrolled 300 consecutive cases from three epilepsy clinics in Northeast, Thailand. Subjects were eligible if reported to have at least one seizure attack during the past 12 months. Face-to-face questionnaire was used to evaluate SRI, baseline characteristics and other seizure-related variables. Results: There were 247 and 91 PWE who met a criterion and had SRI, respectively. By multivariate logistic regression method, GTC seizure type, having history of seizure attacks at least 12 times/year, and daytime seizure were significant risk factors of having SRI with odds ratio of 2.376, 2.460, and 3.562, respectively. We developed the predictive model for having SRI in PWE and it gave 90.3% sensitivity and 46.7% specificity on the occurrence of SRI. The estimated probability of SRI can be found online at http://sribykku.webs.com/. Conclusions: The significant predictive factors for SRI in PWE were the occurrence of GTCs, seizures at least 12 times/year or daytime seizures. Clinicians or PWE can easily evaluate the risk of having SRI in individuals by the online predictive model. © 2009 Elsevier B.V. All rights reserved.
1. Introduction Epilepsy is considered as one of the major public health problems in the world [1]. As other diseases, many studies showed that it contributes to poor quality of life. The impact of epileptic seizures will be more if the seizures are uncontrolled and increase the risk of injuries. Seizure-related injuries (SRI) are common and reported with a wide range of events. It can be just minor trauma to intracranial hematoma [2], extensive severe burns [3] or even death [4]. The European Cohort study showed that 21% of persons with epilepsy (PWE) reported accidents compared to 14% of control subjects [5]. Among the reported injuries, the relative risk for concussion was the highest at 2.6. The risk of trauma was also related to seizure type [3] and the frequency of seizure was the predictive factor of seizure-related injury [6]. These bring up an overprotecting
⁎ Corresponding author. Division of Neurology, Department of Medicine, Faculty of Medicine, Khon Kaen University, Khon Kaen, 40002, Thailand. Tel.: +66 43 363 664; fax: +66 43 374 542. E-mail address: [email protected] (S. Tiamkao). 0022-510X/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.jns.2009.05.021
issue in PWE that may result in social isolation, physical inactivity and psychological dependency. Here, we did a study to evaluate the risk and predictive factors for having SRI in PWE by logistic regression analysis. In addition, we would like to create an online predictive model to assess the risk of having SRI in individuals. Therefore, PWE or clinicians can assess their own risks for SRI electronically. 2. Methods 2.1. Study population We studied 300 consecutive adult PWE at the out-patient clinic at three provincial hospitals in the Northeast, Thailand in 2004. The inclusion criterion was epileptic patients who reported at least one seizure attack during the past 12 months. We used a face-to-face standardized questionnaire by one of the authors to acquire the risk factors for SRI. The questionnaire was used and reported elsewhere [7]. The questionnaire contents included demographics, age of onset and type of seizure, apparent causes, history of falling during seizure attack, seizure attack time whether in daytime or nighttime, total number of seizures during the past 12 months, current numbers of
60
S. Tiamkao et al. / Journal of the Neurological Sciences 285 (2009) 59–61
antiepileptic drugs (AEDs), and the SRI events. The SRI was identified by the occurrence of traffic accidents, near drowning, burns, fractures, joint dislocations, head traumas, or soft tissue injuries during seizure attacks in the past 12 months. The investigators attended questionnaire interview trainings prior to the beginning of the study. The study protocol was approved by the institutional ethical committee. All participants signed the informed consent prior to the beginning of the study. 2.2. Data analysis We analyzed data by using the statistical function of the SAS software version 8.2 (SAS Institute Inc., Cary, NC, USA). Descriptive statistic was used to show the difference between those who had or did not have SRI. The numerical variables including age, age onset of epilepsy, and duration of having epilepsy were presented in median and range. The categorical variables were shown in number of patients (%) including number of male patients and number of one who had generalized tonic–clonic seizures (GTCs), had seizure attacks at least 12 times a year, had daytime or nighttime seizure, had history of falling during seizure attack, and had AEDs from none to 3 or more. The Wilcoxon rank sum test and Chi square test were used to compare differences in numbers and proportions, respectively, of the 2 groups. Difference was considered significant if p b 0.05. We tested all independent variables by univariate logistic analysis. Predictive model was executed using a multivariate logistic regression model. Possible explanatory variables were included and fit a backward elimination method to predict the occurrence of SRI. The retention criterion for the final model was 0.10. Data presented in crude odds ratio (OR), adjusted odds ratio, and 95% confidence interval (C.I.). The Hosmer–Lemeshow goodness of fit statistic was used to assess the fitness of the final model [8]. To demonstrate the discrimination of the model, c statistics or the area under the receiver operating characteristic (ROC) curve was calculated [9]. 3. Results There were 247 of 300 cases who had seizure attack in the past 12 months. Of those, 91 (36.84%) had SRI. The differences of baseline characters between those who had or did not have SRI were shown in Table 1. There was no difference between both groups in genders, age, age onset of epilepsy, duration of having epilepsy, and numbers of cases with nighttime seizure. SRI group had more GTCs cases and
Table 1 Baseline characteristics between no seizure-related injury (no SRI) and seizure-related injuries (SRI) group. Variable
No SRI N = 156
SRI N = 91
Male gender Age (years), median (range) Age onset of epilepsy (years), median (range) Duration of having epilepsy (years), median (range) Seizure type; GTCsa Seizure attacks ≥12 times/year Daytime seizure Nighttime seizure History of fallingb Number of antiepileptic drugs 0 1 2 ≥3
70 (44.9) 34.0 (13–79) 21 (0–78) 6 (0–55)
45 (49.5) 33 (16–91) 21 (0–90) 8 (0–44)
30 (19.2) 39 (25.0) 119 (76.3) 93 (59.6) 117 (75.0)
31 45 84 45 83
(34.1) (49.5) (92.3) (50.0) (91.2)
5 (3.2) 112 (71.8) 34 (21.8) 5 (3.2)
2 (2.2) 49 (53.9) 33 (36.3) 7 (7.7)
Table 2 Crude odds ratio for developing seizure-related injuries by univariate analysis. Variable
Crude OR
95% confidence interval
p value
Male gender Age (years) Age onset of epilepsy (years) Duration of having epilepsy (years) Seizure type; GTCsa Seizure attacks ≥ 12 times/year Daytime seizure Nighttime seizure History of fallingb Number of antiepileptic drugs 0 1 2 ≥3
1.202 1.002 1.000 1.002 2.170 2.934 3.731 0.677 3.458 1.906 1.000 1.094 2.426 3.400
0.716 0.984 0.985 0.980 1.205 1.696 1.587 0.402 1.537 1.242
2.018 1.020 1.015 1.025 3.909 5.075 8.770 1.142 7.782 2.927
0.4867 0.8677 0.9903 0.8586 0.0099 0.0001 0.0025 0.1440 0.0027 0.0032
0.205 0.440 0.473
5.832 13.392 25.892
0.9165 0.3092 0.2200
Note. OR; odds ratio, aGTCs; generalized tonic–clonic seizures, bHistory of falling during seizure attack.
higher number of patients who had seizure attacks, daytime seizures, history of falling during seizure attacks, and number of AEDs. The univariate analysis (Table 2) showed that the independent significant factors related to have SRI were GTCs (OR 2.170 95% C.I. 1.205–3.909), total seizure attacks at least 12 times a year (OR 2.934, 95% C.I. 1.696–5.075), daytime seizure (OR 3.731, 95% C.I. 1.587– 8.770), having history of falling during seizure attack (OR 3.458, 95% C. I. 1.537–7.782), and number of AEDs (OR 1.906, 95% C.I. 1.242–2.927). By the multivariate logistic regression method, there were five remaining variables but only three variables were statistically significant in the final model (Table 3). GTC seizure type, having history of seizure attacks at least 12 times/year, and daytime seizure had the risk of having SRI 2.376, 2.460, and 3.562 times, respectively. The Hosmer–Lemeshow goodness-of-fit test statistic of the final multivariate logistic model was 13.6 (p = 0.09). The area under ROC curve or c statistic of the final model was 0.715. The sensitivity and specificity of the final model were 90.3% and 46.7%, respectively. The predictive model of having SRI can be found below Table 3 or online at http://sribykku.webs.com/. 4. Discussion We previously reported that soft tissue injury was the most common type of injury reported in the U.K. and the Northeast, Thailand [7,10]. The common sites of injury were head, face or arms. Nearly all reported injuries were mild. Similar results were also reported in Europe by the European Cohort [5,11]. The risk of SRI is of great concern to both physicians and PWE for the prevention of possible injury.
p Value 0.4865 0.8923 0.9615 0.2879 0.0091 b0.0001 0.0015 0.1433 0.0017 0.0198
Note. Data are number (%) of patients, unless otherwise indicated. aGTCs; generalized tonic–clonic seizures, bHistory of falling during seizure attack. Data for no SRI and SRI group may not total 156 and 90, respectively, because of missing data.
Table 3 Adjusted odds ratio for developing seizure-related injuries of all remaining covariates in the final model by multiple logistic regression analysis. Variable
Adjusted OR
95% confidence interval
p value
Age Numbers of antiepileptic drugs Seizure type; GTCsa Seizure attacks ≥ 12 times/year Daytime seizure
1.018 1.554 2.376 2.460 3.562
0.998 0.959 1.257 1.329 1.433
0.0846 0.0739 0.0077 0.0042 0.0063
1.039 2.485 4.493 4.552 8.852
Note. aGTCs; generalized tonic–clonic seizures, OR; odds ratio. The predicted probability of having SRI (P) can be calculated for any given patient: P (occurrence of SRI) = 1 / [1 + exp (Z)]. Z equal 3.4096 − (0.0177 × age in years) − (0.4341 × no. AEDs) − (0.8655 × GTCs) − (0.9000 × frequency) − (1.2702 × daytime seizure), where numbers of AEDs = 0 if no. AEDs taken, 1 if one AED taken, 2 if two AEDs taken, 3 if three or more AEDs taken; GTCs = 1 if seizure type is GTCs, 0 if not; frequency = 1 if average seizure attack more than 1 event/month, 0 if not; daytime seizure = 1 if having daytime seizure, 0 if not; exp = exponential.
S. Tiamkao et al. / Journal of the Neurological Sciences 285 (2009) 59–61
The significant independent variables in the final logistic regression were the occurrence of GTCs, seizures at least 12 times/year and daytime seizures. As previously reported, frequent seizure attacks enhanced chances of having SRI [6]. From our data, if epileptic patients had seizure attacks approximately once per month, the risk of having SRI increased by 246 folds. GTC type of seizure introduced unconsciousness and aggravated SRI 2.3 times compared with non-GTC type. Finally, the daytime seizure had the highest risk for developing SRI. Daytime period has much more activities than nighttime period. Some activities such as driving, transportation or recreation activities contribute directly to SRI. The predicted probability of the occurrence of SRI for an individual was shown below Table 3 and online at http://sribykku.webs.com/. This predictive model was created by 5 potential clinical variables remained in the final model; it will be helpful in clinical practice to identify a risk of SRI in PWE. Even though age and numbers of AEDs were not statistically significant in the final model, they included in the predictive model due to their clinical relevance to SRI and the borderline p values (0.08 and 0.07, respectively). The goodness-of-fit of the final model was shown by the Hosmer and Lemeshow method (p value 0.09). The c statistic value or the area under ROC curve of 0.715 suggested a fair discrimination effect of the final model. The predictive model had a good prediction of the occurrence of SRI by 90.3%. This high sensitivity predictive model can be a good tool for screening or ruling out SRI. Clinicians or PWE will know confidently the risk for SRI in individuals. Therefore, low predictive risk patients may not anymore be restricted to some risky activities such as driving or recreation activities. The limitations of our study were interviewer bias, overestimated bias, and recall bias. Participating subject and the interviewer knew each other. To reduce the interview bias, the investigator trainings were arranged prior to the actual study. PWE should be encouraged to live as normal a life as possible if epilepsy is controllable. To avoid fear of doing some activities or
61
overprotection, knowing their own risks for SRI is crucial. With our high sensitivity of the predictive model, physicians can clinically judge PWE who are at risk for the occurrence of SRI.
Acknowledgement The authors would like to thank Professor Simon D. Shorvon for the initiation of the study and Dr. Chatlert Pongchaiyakul for helping in statistical analysis.
References [1] World Health Organization. Epilepsy: aetiology, epidemiology and prognosis; February 2001. Published online: Available from: http://www.who.int/mediacentre/factsheets/fs165/en/. [2] Neufeld MY, Vishne T, Chistik V, Korczyn AD. Life-long history of injuries related to seizures. Epilepsy Res 1999;34:123–7. [3] Unglaub F, Woodruff S, Demir E, Pallua N. Patients with epilepsy: a high-risk population prone to severe burns as a consequence of seizures while showering. J Burn Care Rehabil 2005;26:525–8. [4] Nei M, Bagla R. Seizure-related injury and death. Curr Neurol Neurosci Rep 2007;7:335–41. [5] Van den Broek M, Beghi E, REST-1 Group. Accidents in patients with epilepsy: types, circumstances, and complications; a European cohort study. Epilepsia 2004;45:667–72. [6] Lawn ND, Bamlet WR, Radhakrishnan K, O'Brien PC, So EL. Injuries due to seizure in persons with epilepsy: a population-based study. Neurology 2004;63:1565–70. [7] Tiamkao S, Shorvon SD. Seizure-related injury in an adult tertiary epilepsy clinic. Hong Kong Med J 2006;12:260–3. [8] Hosmer DW, Hosmer T, Le Cessie S, Lemeshow S. A comparison of goodness-of-fit tests for the logistic regression model. Stat Med 1997;16:965–80. [9] Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143:29–36. [10] Tiamkao S, Amornsin O, Pongchaiyakul C, Asawavichienjinda T, Yaudnopakao P, Jitpimolmard S, et al. Seizure-related injuries in Northeast Thailand. J Med Assoc Thai 2006;89:608–13. [11] Cornaggia CM, Beghi M, Moltrasio L, Beghi E, RESt-1 Group. Accidents at work among people with epilepsy. Results of a European prospective cohort study. Seizure 2006;15:313–9.