Study design and setting

Community-based cross-sectional surveys were implemented as part of a larger study investigating the disease ecology of tungiasis in Matuga and Msambweni sub-counties in Kwale, Kenya; Ugenya sub-county in Siaya; and Bugiri in Uganda. The regions have various cultures and ethnicities, including livestock-keeping practices, soil features, and closeness to animal habitats, yet their climate conditions are comparable. The recruitment of participants and data collection was done between February 2020 and April 2021.

Study size

The study aimed to test if the mean of outcomes were significantly different in the two group. Therefore, a sample size for a two-sample means test was computed. The study used category fluency as reference outcome. Previous study in a similar setting but with younger population (mean age = 5.2 years) reported a mean category fluency of 15.97 [10]. Assuming a common standard deviation of 2 the study required at least 506 (253 infected and 253 uninfected) participants to detect at least 0.05 difference in means at α = 0.05 and power of 0.8. The sample size was calculated using Stata [11]. However, the actual study size was 454 (220 infected and 234 uninfected), mean difference of category fluency between the two group was 2.6 and a common standard deviation of 7, giving the study a power of 0.97.

Study population and sampling procedure

The study targeted eight to fourteen-year-old children, the most susceptible to tungiasis infection [2]. The study established inclusion criteria that specified residency in a household with a natural soil floor, given its known association with increased risk of tungiasis, and the availability of an adult caregiver for informed consent and interviews. In addition to those specified for infected participants, eligibility criteria for uninfected participants also required the absence of infected family members. In stage I, sixteen public primary schools were to be randomly chosen within each region from a list of all existing public primary schools. However due to the low prevalence of tungiasis in some regions, additional 12 schools in Siaya and four schools in Kwale were randomly selected. Moreover, due to the exceptionally low prevalence of tungiasis in the Bugiri region, the decision was made to conclude data collection after surveying only eight schools. As a result, the selection outcome for Stage I comprised 28 schools from Siaya, 20 schools from Kwale, and eight schools from Bugiri. In stage II, up to a maximum of 102 school-going children were quasi-randomly selected in each school. This process resulted in a total of 5331 pupils. Hands, and feet of the 5331 pupils were visually inspected for tungiasis. Out of the 5331, 589 pupils were infected while 4742 pupils were uninfected. In each school, up to 10 infected and 20 uninfected pupils were then quasi-randomly selected as index pupils from those with tungiasis and those without, respectively. This process resulted in selection of 361 from 589 infected pupils and 729 pupils out of 4742 uninfected pupils as index pupils as shown in Fig. 1. These index pupils were to participate in the larger study. In stage III, six infected and six uninfected children were quasi-randomly selected in each school from the pool of index pupils. The section of infected pupils at this stage was based on severity of infection-aiming at three with severe (> 10 fleas) and three with a mild infection (< 10 fleas) where possible. Conversely, the selection of uninfected children was done through a simple random process. Overall, stage III resulted in selection of 253 of the 361 infected pupils and 523 of the 729 uninfected pupils. This total of 506 pupils (253 infected and 253 uninfected pupils) formed the the final study group for neurocognitive and mental health assessments (Fig. 1).

Fig. 1

Participant selection flow diagram. Orange represents infected children; green represents uninfected children. Yes, indicates pupils selected for the next stage. No, indicates pupils excluded from the study

Study variables

Neurocognitive function and mental health problems were main outcome variables. Five domains of neurocognitive functioning were: attention, memory, language, perceptual-motor, and executive function. These domains are detailed in S1 of the Additional file 1. The explanatory variable of interest was tungiasis status. Other explanatory variables included in each model as potential confounders were nutritional status (underweight, stunting, wasting), disability, perinatal complications, residence (Siaya, Kwale, Bugiri), socioeconomic status (SES), school absenteeism (school days missed in the week preceding data collection), orphanhood, household size, ill family member, and both household head-related factors (sex, relation to participant, age) and caregiver-related factors (sex, relation to participant, age, spending time with the participant, exposure to hugging or cuddling, correction method, caregiver depression, and caregiver stress).

Data sources and methods of measurementDiagnosis of tungiasis and classification of infection

Trained community health workers washed children’s feet to expose embedded fleas. The feet and fingers of the children were visually inspected embedded flea. The selected participants were categorised infected if they had at least an imbedded flea and uninfected if they did not present with a flea. The embedded fleas were manually counted and the infected further categorised into infection status as mild infection if they had less than 10 imbedded fleas and severe infection if they presented with ten or more imbedded fleas.

Neurocognitive and mental health measures

Participants underwent approximately two-hour battery of tests administered by trained research assistants to assess their neurocognitive abilities across multiple domains. Language function was evaluated with the Early Grade Reading Assessment (EGRA) [12] and Category Fluency Test (CFT) [13], while attention was assessed with the Comprehensive Trail-Making Test (CTMT) [14] and Stroop Color and Word Test (SCWT) [15]. Working memory was evaluated using the backward digit span task [16], and fine motor control was assessed with the bead threading test [17]. The battery also included the Early Grade Maths Assessments (EGMA) [18], Standard and Coloured Raven Progressive Matrices (RPM) [19] to evaluate numeracy and nonverbal intelligence, respectively. The lower scores in these cognitive tests indicate poor neurocognitive function. These tests are valid and reliable for assessing their respective domains and have been adapted for use in Kenya [20] and Uganda [10]. The Child Behavior Checklist [CBCL] [21] was used to assess mental health outcomes. In this study, the total score was used to assess mental health problems, with higher scores indicating more problems. The neurocognitive and mental health measures are detailed in S1 in supplementary material.

Covariates

Anthropometric measurements including height, weight, and Mid-arm circumference (MUAC) were used to assess nutritional status. Height was measured using a stadiometer, weight was measured using a calibrated scale, and MUAC was measured using a flexible tape measure. Height-for-age (HAZ) and weight-for-age (WAZ) z-scores were calculated according to the Centers for Disease Control and Prevention (CDC) [22], with HAZ z-scores < − 2 and WAZ z-scores < − 2 indicating stunting and underweight, respectively, while MUAC was used to evaluate wasting. Structured questionnaires were used to collect data on factors associated with poor neurocognitive and mental health outcomes, including disability child perinatal complications, region of residence, child age, child sex, and school grade level. Household socioeconomic status (SES) was assessed using tetrachoric principal component analysis (PCA) and the resulting wealth index was created based on eigenvalue and scree plot as detailed in S2, S3 and S4 in the Additional file 1. Psychosocial covariates including orphanhood, school absenteeism, household size, and caregiver information were also collected. Respondents’ relation to participant, age, spending time with the participant, exposure to hugging or cuddling, correction method, and caregiver mental health (depression, and stress).

Psychosocial covariates covered various topics such as orphanhood (also assessed using binary response options), school absenteeism (measured as the number of days absent from school in the week preceding data collection), household size (classified as either more than 2 adults or less than 2 adults), having a chronically ill family member (also assessed using binary response options), sex of household head, relation of the household head to the child (classified as either "child" or "not child"), age of household head, sex of caregiver, age of caregiver, and the relation of the caregiver to the child. Other factors assessed in the questionnaires included the amount of time the caregiver spent with the child (reported as "a lot of time" or "not a lot of time"), whether the child was hugged or cuddled (assessed using a binary response option of "yes" or "no"), and the caregiver's methods of correcting the child (reported as "beating" or "other methods").

Caregiver mental health was measured using Patient Health Questionnaire-9 (PHQ-9) and Parental Stress Scale (PSS). The PHQ-9 is a self-report questionnaire that measures depression by asking the respondent to rate the frequency of their symptoms over the past 2 weeks on a scale of 0–3 [23]. The questionnaire covers various areas related to depression, and scores range from 0 to 27. In this study depression cutoff was scores of 10 and above. The Parental Stress Scale (PSS), a questionnaire used to assess parental stress, includes 18 items covering various aspects of parenting, and respondents rate how often they experience stress related to each item on a 5-point scale [24]. The total score is calculated by summing the scores, with higher scores indicating higher levels of parental stress. In this study CBCL, PHQ-9 and PSS had acceptable alphas of 0.94, 0.87 and 0.68, respectively an indication of internal consistency.

Measures to address bias and errors

The outcome assessors were distinct from the infection assessors and were kept unaware of the participant's status to minimize potential biases, however, the status could be known in participants with visible signs of infection. Questionnaires and test score sheets were adapted to the Research Electronic Data Capture (REDCap) database [25], hosted at the International Centre of Insect Physiology and Ecology (ICIPE). Responses were recorded on tablets conditionally formatted to ensure the validity, accuracy, and completeness of the data. The study's data collection involved in-person interviews and assessments performed by assessors trained in neurocognitive evaluations. To accommodate diverse local languages, the questionnaires were initially written in English and subsequently translated into Kiswahili, Dholuo, and Kisoga. Information bias was reduced through pretesting, adjustments, and clarifications of the questions before commencing data collection. To promote consistency in the interview process, a workshop was held involving all assessors and investigators to establish a common interpretation of the responses. Moreover, to minimize misclassification bias, the assessors verified the presence of other infection indicators (like pain and itching) in infected.

Statistical analysis

Data were analysed utilizing STATA software version 15 [11]. The difference in distributions between the uninfected group and each infected group, as well as among all groups together, were compared using binomial and multinomial tests, respectively. Continuous data were presented as means and standard errors (SE) if normally distributed or medians and interquartile range (IQR) if skewed. For bivariate analyses of continuous data, Mann Whitney U Test, or the Kruskal–Wallis rank test was used to compare skewed distributions while Two Sample Students t test or analysis of variance (ANOVA) was used to compare means of normally distributed data. Categorical data were presented as frequencies with respective percentages and their proportions compared using the Pearson Chi-Square Test or Fisher's Exact Test. Analysis of covariates (ANCOVA), followed by the Scheffe Test was used for pairwise comparison of adjusted means. Covariates adjusted for included age, sex, grade, nutritional status, care giver education, disability, school absenteeism and SES.

For regression analyses, multilevel mixed effects generalized linear models with gaussian family and identity link were used for bivariable and multivariable analyses with the unique school identifier as a random effect to identify factors associated with neurocognitive and mental health outcome scores in children adjusting for age, sex, and grade as priori confounders. Multivariate analyses were conducted separately for each neurocognitive and mental health outcome. Backward stepwise selection was used to identify the most significant variables for the model. An exhaustive model containing all predictor variables was initially established. The variable with the highest p-value was subsequently eliminated from the model. This iterative process was continued until the stopping criterion (P < 0.05) was reached and the model selected as final.