Menu
Menu
Menu
Menu
Background Three-quarters of the burden of mental health problems occurs in low-and-middle-income countries, but few epidemiological studies of these problems in preschool children from sub-Saharan Africa have been published. Behavioural and emotional problems often start in early childhood, and this might be particularly important in Africa, where the incidence of perinatal and early risk factors is high. We therefore aimed to estimate the prevalence and risk factors of behavioural and emotional problems in young children in a rural area on the Kenyan coast. Methods We did a population-based epidemiological study to assess the burden of behavioural and emotional problems in preschool children and comorbidities in the Kilifi Health and Demographic Surveillance System (KHDSS, a database formed of the population under routine surveillance linked to admissions to Kilifi County Hospital). We used the Child Behaviour Checklist (CBCL) to assess behavioural and emotional problems. We then determined risk factors and medical comorbidities associated with behavioural and emotional problems. The strength of associations between the risk factors and the behavioural and emotional problems was estimated using generalised linear models, with appropriate distribution and link functions. Findings 3539 families were randomly selected from the KHDSS. Of these, 3273 children were assessed with CBCL. The prevalence of total behavioural and emotional problems was 13% (95% CI 12–14), for externalising problems was 10% (9–11), and for internalising problems was 22% (21–24). The most common CBCL syndrome was somatic problems (21%, 20–23), whereas the most common DSM-IV-oriented scale was anxiety problems (13%, 12–14). Factors associated with total problems included consumption of cassava (risk ratio 5·68, 95% CI 3·22–10·03), perinatal complications (4·34, 3·21–5·81), seizure disorders (2·90, 2·24–3·77), and house status (0·11, 0·08–0·14). Seizure disorders, burn marks, and respiratory problems were important comorbidities of behavioural and emotional problems. Interpretation Behavioural and emotional problems are common in preschool children in this Kenyan rural area and are associated with preventable risk factors. Behavioural and emotional problems and associated comorbidities should be identified and addressed in young children. Funding Wellcome Trust.
This study was done within the Kilifi Health and Demographic Surveillance System (KHDSS), which is located in Kilifi County, about 60 km north of Mombasa city.12 The KHDSS is both an area (divided into enumeration zones under regular surveillance) and a database (formed of the population under routine surveillance linked to admissions to Kilifi County Hospital). The KHDSS has a population of about 280 000 residents who are predominantly of the Mijikenda tribe and has an estimated 50 000 children aged 1–6 years. The KHDSS has a northern and southern region covering an area of 891 km2. Epilepsy and neurodevelopmental clinics at Kilifi County Hospital provide therapeutic interventions and counselling services. Screening in stage 1 was done by trained fieldworkers who read out the content of the questionnaires to the parents owing to low literacy levels in this area, taking short breaks every 10 min. The three questionnaires (for risk factors, behavioural and emotional problems, and seizures) in stage 1 were administered in a random order. A random sample of those with and without behavioural and emotional problems predetermined through a sample calculation was invited in stage 2 for a clinical evaluation study to determine medical comorbidities. This study was approved by the Scientific and Ethics Review Unit of the Kenya Medical Research Institute and written informed consent was obtained from parents or carers of children participating in the study. Behavioural and emotional problems were assessed in stage 1 with the Child Behaviour Checklist (CBCL), which was adapted and piloted in the local population and languages.13 The CBCL is used in children aged 1·5–5·5 years,2 and has been applied on 1–6-year-old children; it identifies seven syndromes and five DSM-IV-oriented scales.14 The CBCL has acceptable psychometric properties on a sample of Kenyan preschool children in this area.13 The CBCL items had an internal consistency Cronbach’s α of 0·95, and inter-informant agreement (Pearson’s correlation coefficient, r>0·80), test–retest reliability (r=0·76), and the fit index of the seven-CBCL syndromes (eg, root mean square error of approximation <0·05) were within acceptable ranges.13 Because of the literacy challenges in this area, CBCL questions were read out to the respondents (parents or guardians) by trained neuropsychological assessors fluent in the local languages. The language of administration was primarily Kiswahili (lingua franca), but Giriama was also used for a few respondents who could not comprehend Kiswahili. We used a systematic approach of translation and adaptation of the tools. The initial translation was done by two independent translators fluent in the original language (English) and the target language (Kiswahili and Giriama). These translations were then back translated into English by two independent translators and inconsistencies were resolved. The scoring system used and the grouping of the CBCL items into syndromes and externalising and internalising subscales followed recommendations by the Achenbach System of Empirically Based Assessments.2 The total score for the CBCL was formed by summing ratings from all of the 99 items. Items that formed the seven syndromes of the CBCL, externalising and internalising scores, and the DSM-oriented scales are outlined in the appendix. The 90th percentile of the CBCL scores was used as the cutoff according to recommendations from the developers of the tool, and produced cutoff scores similar to those applied in children in the USA.15 These cutoffs were piloted and found to discriminate between children with and without adverse perinatal events.13 Parents and carers of children assessed with the CBCL were interviewed using a parental risk factor questionnaire. The risk factor questionnaire consisted of the following items: maternal sociodemographic information such as employment, schooling, religion, and ethnicity; pregnancy and perinatal histories; socioeconomic status indicators such as housing status; medical histories such as seizure disorders or other chronic illnesses; and child health and nutrition habits, such as food types consumed, eating soil, and snoring at night. A thorough literature search informed the choice of risk factors included in the questionnaire. About 20% of the children were invited to take part in stage 2 of the study. One clinician saw 243 children with CBCL scores of more than 60 and 377 children with CBCL scores of less than 60. The children were randomly selected from those screened in stage 1 using the RAND command of MySQL (Oracle Corp, USA). The clinician who was blinded to the screening status in stage 1 asked questions about the history of seizures to determine whether the seizures were acute seizures or epilepsy. The clinician was blinded to the status of the children screening positive for seizures in stage 1 in the community. The clinician did a clinical examination to assess for gross and fine motor deficits, sensory function, abnormal limb activity, cognitive or mental status, cranial nerve function, sensory ability, and skin integrity. Anthropometric measures of the child were also taken and included head circumference, mid-upper arm circumference, height, and weight. Abnormal pregnancy was defined as premature or prolonged labour, post-dated pregnancy, pre-eclampsia, eclampsia, or any other health problems during pregnancy.16 Adverse perinatal events were defined as delays in crying, breathing, and breastfeeding after birth. Seizure disorders included both epilepsy and febrile acute seizures, with febrile seizures defined as seizures associated with a febrile illness or fever in those younger than 6 years, and epilepsy as a history of two unprovoked seizures.17 Intellectual disability was assessed by a clinician observing young children who had problems performing the standardised test of a locally adapted developmental inventory.18 Malnutrition was defined as a weight-for-age z score value of −2 or lower or a mid-upper arm circumference less than 11·5 cm. Sensory function was considered impaired if a child could not localise touch from cotton wool or a painful stimulus. Motor impairments were defined as an inability to hold toys and walk or sit upright if of an appropriate age. We estimated that screening approximately 3500 children aged between 1 and 6 years, randomly selected from a surveillance database of 50 000 children using RAND command would be sufficient to identify psychopathology with a precision of 1%. We assessed whether the CBCL scores had a normal distribution by plotting quintile, Kernel density (of predicted regression residuals), and histogram plots (appendix) on raw and (natural) log-transformed scores. Prevalence of behavioural problems was computed first as a probability (where those with CBCL problems are treated as positive and those without CBCL problems as negative), applying the inverse logit function to the intercept coefficient of a logistic regression model to provide outcome probabilities on the logit (log odds) scale. The probability was then multiplied by 100 to obtain the prevalence. Prevalence estimates stratified by age group and sex were computed, fitting fractional polynomial equations to smooth the prevalence by age. Risk factors associated with behavioural and emotional problems were determined using log binomial regression models implemented in a generalised linear model, with robust variance-covariance matrix of the estimators. β coefficients for each risk factor above were then computed with log-transformed CBCL scores as the dependent variable using a generalised linear model that assumes a normal distribution and has an identity link function, and a robust variance–covariance matrix. We built penultimate models that accounted for child factors (age, sex, schooling, and region of residence) and final models that accounted for both child factors and maternal factors (age, marital status, economic or employment status, education level, ethnicity, and religion). A test for linear trend was performed for risk factors categorised into three or more levels using likelihood ratio tests. Age for the child and mother were entered into the risk factor models as first degree fractional polynomials. We measured associations between discrete variables using Pearson's χ2 tests, or Fisher's exact tests, where the number of observations in a cell was less than five. Student's t test or the Mann-Whitney U test were used to compare the distribution of behavioural and emotional scores. Internal consistency was computed using the cialpha command in Stata. A p value of 0·05 or less was considered significant for exploratory comparisons, whereas associations were significant if the lower CIs did not include 1. All analyses were done with Stata, version 13. The funders of the study had no role in study design, data collection, data analysis, data interpretation, or the writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.
