Prevalence, risk factors and behavioural and emotional comorbidity of acute seizures in young Kenyan children: a population-based study

This study was conducted in Kilifi, which is on the Kenyan coast, about 60 km north of Mombasa city [15]. The Kilifi Health and Demographic Surveillance System (KHDSS) has data on a population of about 270,000 residents, who are predominantly of the Mijikenda tribe. It is estimated that the population of children aged 1–6 years is about 50,000. The KHDSS has a northern and southern region divided by the Kilifi Creek, and it covers an area of about 891 km². Kilifi County Hospital (KCH) is the only referral hospital in the area, and it draws admissions from Kilifi. There is an epilepsy and neurodevelopmental clinic, which provides treatment for seizures and counselling services for mental health problems [16].

We estimated that randomly selecting and screening 7587 children from 50,000 children aged 1–6 years within KHDSS in stage I would give the prevalence of seizure disorders (of at least 5%) with a precision of <1%. We had estimated that this survey would generate at least 250 children with acute seizures and an equal number of controls and that these numbers would be sufficient to measure an association between acute seizures and behavioural/emotional problems (with a risk ratio of ~2) with at least 80% at the 5% significance level. Those positive for seizures in stage I were invited for further clinical, neurological and anthropometrical evaluation in stage II, details for which are provided below. The 7587 children to participate in the study were randomly selected from a KHDSS database of 50,000 children living in the community aged 1 years using the RAND command of MySQL (Oracle Corp, USA). A parental risk factor questionnaire with items on sociodemographic, socioeconomic and medical history information was administered to all those with acute seizures and to a predetermined sample of those without seizures, who served as controls.

Seizures were screened in stage I by trained fieldworkers using a modified seven-question seizure questionnaire, which was piloted before the study (Additional file 1). The questions were based upon previous screening questionnaires for epilepsy studies [17] but were modified to capture seizures with fever occurring in the childhood period. Additional details on seizure onset, duration and localisation and family history were obtained in stage I, which could be used for those screening positive who failed to turn up for clinical evaluation in stage II (subsequently referred to as attrition). Electroencephalography (EEG) was performed on children with acute seizures who cooperated as previously described [18].

About half of the children screened for the presence or absence of a history of seizures in stage I were randomly selected for assessment for behavioural or emotional problems using the Child Behaviour Checklist (CBCL), which had been adapted and validated for use in this rural population and showed excellent psychometric properties in 301 children, some of whom were young [14, 19]. The strength of the CBCL is that it has questions that apply to young children such as ‘cries a lot’, ‘clings to adults or too dependent’, ‘doesn’t want to sleep alone’ etc. [19]. Administration of the CBCL (for behavioural and emotional assessment) was done on the same day of screening for seizures in stage I by the same fieldworker. As previously described [14], we used the CBCL scores, completed by the mother or primary caregiver, to form total CBCL problems, externalising problems, internalising problems, CBCL syndromes and Diagnostic and Statistical Manual of Mental Disorders IV (DSM-IV) oriented scales. A similar parental risk factor questionnaire was administered in stage I for those who screened negative for seizures and in stage II during clinical evaluation of those who screened positive in stage I.

Children with a history of seizures and a random proportion (9%) of those without seizures were seen by a clinician in stage II for further clinical evaluation. The clinical evaluation in stage II was scheduled to happen within a week of screening in stage I. Those who failed to turn up were followed up and scheduled for evaluation as soon as possible. The clinician, who was blinded to seizures status in stage I, asked questions on the history of seizure disorders and determined whether the seizures were acute seizures or established epilepsy (according to the International League Against Epilepsy guidelines). The clinician obtained further information about onset of seizures, seizure type/frequency and duration. Definitions for seizure disorders and other conditions are in Additional file 2: Table S1. A clinician assessed cognitive impairments by observing how children performed on a number of items in a standardised developmental inventory [20]. A neurological examination was performed on all children invited for further review.

Causes of acute seizures were identified by asking the parent about the febrile illness that was associated with the seizures, and symptoms and signs at the time of the seizure (Additional file 2: Table S2). This followed the World Health Organization algorithms for Integrated Management of Child Infections (http://​www.​who.​int/​maternal_​child_​adolescent/​topics/​child/​imci/​en/​), which were piloted among children admitted with malaria and correctly identified the majority of these children for treatment [21]. The community data on acute seizures were linked with hospital admissions to identify additional causes of acute seizures. Anthropometric measures of the child were also taken by neuropsychological assessors and included head circumference, mid-upper arm circumference, height and weight.

Acute seizures (referred to, interchangeably, as provoked seizures) were febrile seizures if they were not associated with a diagnosis of malaria, meningitis or encephalopathy as identified with causes recorded in the hospital database or Integrated Management of Child Infections guidelines, while the remainder were presumed to be acute symptomatic seizures, since malaria is ubiquitous in this area. Convulsive seizures were classified into either focal or generalised, with detailed definitions provided in Additional file 2: Table S1.

All analyses were performed using STATA (version 13, StataCorp, TX, USA). The prevalence of acute seizures was computed as a probability derived from the inverse link function of a logit model. Acute seizures are viewed as relatively benign in this community and are not treated at hospital, which may cause attrition between stage I and II. Attrition was addressed using multiple imputation by chained equations [22], implemented in STATA by constructing a model with age and seizure variables obtained in stage I. The assumption in the multiple imputation was that the missing data for acute seizure status in stage II occurred at random because failure to report for further clinical evaluation in the second stage also occurred at random. We included age and common phenotypes of seizures (e.g. prolonged seizures, focal seizures and repetitive seizures) to infer for missing acute seizure status in stage II because these variables were documented for all children in stage I and are often correlated with acute seizures [2].

The generalised linear models (GLMs) for risk factors for acute seizures had a logit link function to compute odds ratios (ORs), which closely approximate risk ratios when the outcome is relatively rare (<10%) [23]. The association of behavioural and emotional problems with acute seizures was determined by fitting a GLM of the binomial family (with a log link). Effect estimates for the association of behavioural and emotional problems with acute symptomatic seizures was hypothesised to be greater than that with febrile seizures, which was tested by computing ratios of the risk ratios (ROR) as previously described [24].

We investigated if the association between acute seizures and behavioural and emotional problems is mediated by a co-diagnosis of epilepsy using the Sobel–Goodman mediation tests [25]. The mediation analysis was necessary to understand which of the two seizure disorders (acute seizures or epilepsy) is associated with behavioural and emotional problems, since acute seizures can occur in children with a history of epilepsy [10] and both conditions can have poor behavioural outcomes. In those with acute seizures, risk factors for behavioural and emotional problems were determined with GLM models, with a binomial distribution and log link (when the outcome was binary), and with a Gaussian distribution and identity link (when the log-transformed CBCL scores were the outcome). In these models, McFadden’s pseudo R² was used to test the variance of behavioural and emotional comorbidity of acute seizures explained by significant risk factors. Cohen’s d statistics was used to compare CBCL scores between groups. Risk factors reaching a p value ≤0.25 in the bivariate analysis were entered into a multivariable model. Pearson’s chi squared was used to compare discrete variables between groups, while Student’s t-test or the Mann–Whitney U test was used to compare continuous variables, such as age and behavioural/emotional scores. A p value of 0.05 or less was considered significant (unless a Bonferroni corrected cut-off is indicated), while an ROR > 1.0 signified a statistically significant difference between two ORs.

There were 614 children screening for any seizures in stage I (Fig. 1), of whom 56 (9.1%) were treated for seizures in KCH, suggesting that the treatment gap for seizures is >90%. The total number of provoked seizures after accounting for attrition (between stages I and II) was 429 (261 from clinical history or hospital records, 145 from multiple imputation and 23 extrapolated from the assessment of screen negatives) (Fig. 1), which is an overall prevalence of 6.1 per 100 (95% CI: 5.5–6.8). The prevalence ratio of provoked seizures as a function of age-year was 1.10 (95% CI: 1.04–1.17) (Additional file 2: Figure S1). Of the 261 children with provoked seizures confirmed by a clinician, 126 (48%) met the definition for febrile seizures, while the remainder were acute symptomatic seizures. This translates to a prevalence of 2.9 per 100 (95% CI: 2.6–3.3) for febrile seizures and 3.2 (95% CI: 2.9–3.5) for acute symptomatic seizures based on the numbers in Fig. 1.

The median age of onset of all provoked seizures was 13 months (interquartile range, IQR, 6–25). Most children experienced generalised tonic-clonic seizures (221/261 or 85%), few had multiple seizure types (17/261 or 7%), while seizure type could not be established in the remainder. Complex provoked seizures (focal, repetitive and/or prolonged) occurred in 240/261 (84%) of all seizures, in 114/135 (84%) of acute symptomatic seizures and in 106/126 (84%) of febrile seizures (Additional file 2: Figure S2). Provoked seizures preceded a history of epilepsy in 30/261 (11%) children. Of the 261 provoked seizures diagnosed by a clinician, 158 (61%) children agreed to undergo an EEG recording, with no differences in age (p = 0.08), sex (p = 0.77) or complex phenotypes (p = 0.83) between those with and without an EEG assessment. Abnormalities were detected in 43/158 (27%) of the EEGs and 22/43 (51%) of the abnormalities were focal. An abnormal EEG was neither correlated with focal seizure semiology (tetrachoric rho = 0.12, p = 0.42) nor with neurological or motor deficits (tetrachoric rho = 0.15, p = 1.00).

Among the identifiable causes of provoked seizures, falciparum malaria was the most important cause of provoked seizures (112/261 or 43%), followed by respiratory tract infections (68/261 or 26%). The causes were not statistically different between complex and simple provoked seizures (Additional file 2: Table S3). In the multivariable analysis, risk factors for provoked seizures included family history of febrile seizures (adjusted odds ratio (aOR) = 3.19; 95% CI: 2.03–5.01) and previous hospitalisation (aOR = 6.65; 95% CI: 4.60–9.63) (Additional file 2: Table S4).

Of the 3273 children who received the CBCL, 200 (6.1%) had provoked seizures while 26 (0.7%) had had provoked seizures preceding epilepsy. The mean and median scores were significantly higher in those with provoked seizures than those without seizures for total problems, externalising problems, internalising problems and for the six CBCL syndromes and five DSM-IV-oriented scales (Table 2 and Fig. 2). However, effect sizes were variable for comparisons between provoked seizures and those without these seizures, and appeared greatest for internalising problems (Cohen’s d = 0.44; 95% CI: 0.29–0.58) (Table 2).

The median total CBCL scores for children with acute symptomatic seizures (37; IQR: 23–63) and febrile seizures (35; IQR: 23–62) were higher than for those without seizures (27; IQR: 18–21); p < 0.001 for both comparisons. This difference in scores was consistent with the effect sizes (Cohen’s d = 0.59, 95% CI: 0.32–0.80, for acute symptomatic seizures) and (Cohen’s d = 0.35, 95% CI: 0.10–0.61, for febrile seizures).

The crude prevalence of total CBCL problems was 27% (95% CI: 21–34%) for those with provoked seizures, 30% (95% CIs: 20–43%) for those with acute symptomatic seizures and 25% (95% CI: 15–38%) for those with febrile seizures; all being significantly greater than for those without seizures (11%; 95% CI: 11–12%; chi-squared p ≤ 0.001). The crude prevalence of externalising problems was significantly greater in provoked seizures (18%; 95% CI: 13–24%), in acute symptomatic seizures (23%; 95% CI: 14–35%) and in febrile seizures (22%; 95% CI: 12–35%) than in those without seizures (12%; 95% CI: 11–13%); chi-squared p < 0.001 for all comparisons. Among those with provoked seizures, total CBCL problems were not significantly increased in those with and without focal provoked seizures (35% vs 30%, X² p = 0.41), neurological deficits (25% vs 39%, Fisher’s exact p value = 1.00) or abnormal EEG (11% vs 30%, Fisher’s exact p = 0.09).

Provoked seizures were associated with total CBCL problems (adjusted risk ratio (aRR) = 1.92; 95% CI: 1.34–2.77), externalising problems (aRR = 1.82; 95% CI: 1.21–2.75) and internalising problems (aRR = 1.57; 95% CI: 1.22–2.02) after accounting for socio-demographic factors (e.g. child’s age) and other non-seizure confounders (e.g. perinatal complications) (Table 3). The associations for total CBCL problems were greater for acute symptomatic seizures (RR = 3.64; 95% CI: 1.94–6.84) than for febrile (aRR = 2.00; 95% CI: 1.02–3.92), which was supported by the ROR for the two associations (1.82; 95% CI: 1.31–2.53).

In similar adjusted models, specific CBCL syndromes remained associated with provoked seizures, particularly anxious or depressed problems (aRR = 1.90; 95% CI: 1.34–2.70) and so were the DSM-IV-oriented scales, particularly affective problems (aRR = 2.05; 95% CI: 1.18–3.54) (Table 3). These associations remained significant even with continuous CBCL scores as the response variable (Additional file 2: Table S5). The proportion of the association between behavioural and emotional problems and provoked seizures mediated by a co-diagnosis of epilepsy was small (15.3%; 95% CI: 4.5–34.9%) (Additional file 2: Table S6).

In those with provoked seizures, five risk factors were significantly associated with total CBCL problems in the multivariable model, notably family history of febrile seizures (RR = 3.36; 95% CI: 1.34–8.41). These five significant risk factors explained 30% of the variation in the model (Additional file 2: Table S7). In those with provoked seizures, nine risk factors were significantly associated with externalising problems in the multivariable model, among them repetitive provoked seizures (β = 0.36; 95% CI: 0.15–0.57). These nine significant risk factors explained 79% of the variation in the model (Additional file 2: Table S8). In those with provoked seizures, four risk factors were significantly associated with internalising problems in the multivariable model, among them focal provoked seizures (RR = 1.80; 95% CI: 1.05–3.08). These four significant risk factors explained 48% of the variation in the model (Additional file 2: Table S9).

The prevalence of febrile seizures (2.9%) was higher than found in other studies of febrile seizures, including Tanzania (2.1%) [7] and the United Kingdom (2.3%), but lower than in the United States of America (3.9%) [26]. The Tanzanian study was conducted within administrative boundaries in which vital statistics are updated less often than the demographic surveillance areas used in our study, which can potentially affect prevalence estimates. Most seizures with fever in Kilifi are acute symptomatic seizures [10], including in this study (3.2%), perhaps explaining the low prevalence of febrile seizures in our study compared to the American study. Some febrile seizures could be symptomatic, considering this study was in a malaria-endemic area, where parasites may sequester into the brain [10]. The prevalence estimates from our study confirm that hospital incidence [27] grossly underestimates the true prevalence of acute seizures in the community. The large treatment gap for acute seizures in this area should be addressed, since only 9% of those screening for seizures in stage I had ever sought treatment for seizures at the only referral hospital in this area.

Like studies from resource-rich countries [28], a family history of febrile seizures was documented in up to 40% of children with acute seizures, suggesting a genetic susceptibility or shared environmental factors. The association with previous hospitalisation likely points to febrile illnesses that cause acute seizure, e.g. falciparum malaria (identified as the most common cause of acute seizures in this study; 43%). Cassava consumption, which may be a marker of poverty, is associated with neuropsychiatric problems [29]. Our study was not designed to investigate the possible neurotoxicity of cassava.

The relatively lower age of onset, compared to other studies [3], reflects the susceptible period for the infectious causes of acute seizures in this area. Complex acute seizures were common (85%), comparable to previous studies in this area (84%) [13]. EEGs were abnormal in about 30%, suggesting a propensity to develop epilepsy [9], and the explanation of why acute seizures preceded epilepsy in 11% of children in this study. These features may affect mental health outcomes in children with acute seizures.

Acute seizures remained associated with behavioural and emotional problems, externalising problems and internalising problems after accounting for potential demographic and non-seizure confounders. This was consistent with the frequency and effect size differences (using Cohen’s d statistics) between those with acute seizures and those without. Acute symptomatic seizures contributed significantly greater behavioural and emotional comorbidities than febrile seizures since the ROR and its corresponding lower confidence interval were greater than 1. A small proportion of the association (15%) was mediated by a co-diagnosis of epilepsy, a marker of underlying cortical damage. These associations do not imply a causal association, as this was a cross-sectional study, although we asked about behavioural and emotional problems in the past 2 months, when some children would have experienced their first acute seizures. Alternatively, acute seizures can presumably create anxiety in the child and the mother [30].

Multivariable models identified significant risk factors that explained the substantial variation in these models, e.g. up to 80% for externalising problems. These factors can explain the behavioural and emotional comorbidity of acute seizures in three ways: (i) direct biological effect, (ii) shared pathway and (iii) confounding or bias. A family history of febrile seizures (associated with total CBCL and externalising problems) suggests there is an underlying genetic susceptibility [31]. Repetitive or focal seizures can have a direct biological effect through damaging parts of the brain or changing the maturation of neuronal networks, possibly explaining the association between behavioural and emotional problems and acute seizures. Head injury and cognitive impairment could be markers of a shared pathway, e.g. underlying neurological damage [32] associated with both behavioural problems and acute seizures. The number of siblings can increase family or parental distress [33], thereby confounding the association between behavioural and emotional problems and acute seizures. Snoring, when caused by an upper airway obstruction, can have a direct hypoxic effect on the brain that manifests as behavioural and emotional problems in these children with seizures [14]. The role of eating soil in the behavioural and emotional comorbidity of acute seizures is unclear and requires further investigation, since the association could be due to residual confounding by the effects of heavy metals in soils, which may be associated with neurodevelopmental impairments [34].

To our knowledge, this is the largest epidemiology study of acute seizures and behavioural emotional problems in young children within the community in Africa using a robust demographic surveillance system. The two-stage methodology ensured that most acute seizures were diagnosed by a clinician. Behavioural and emotional problems were assessed with an internationally applied behavioural and emotional assessment tool (the CBCL), which was validated and adapted to the local population [19].

Attrition between stages I and II (addressed with multiple imputation) and the limited sampling of clinical assessment and EEG may affect the interpretation of results. Although we leveraged on a demographic surveillance that is linked to admissions to KCH to establish the acute seizure status of children who did not turn up for clinical evaluation in stage II, our demographic surveillance is not linked to other peripheral health facilities and private clinics, where some children may have been treated for acute seizures. The use of multiple imputation to account for attrition may introduce some bias if there were some unmeasured factors that could make failure to report for assessment occur at random. Maternal mental health status, quality of parenting, HIV status and family functioning may influence behavioural and emotional outcomes [30], but they were not assessed in our study.