Ethics statement

The institutional review board at the College of Medicine and Health Sciences of Hawassa University (IRB/005/09) and the Regional Ethical Committee of Western Norway (2016/1900/REK vest) provided ethical clearance. The Gedeo Zone Health Department and District Education Office provided a letter of permission. School directors and teachers participated in discussions. We obtained informed written (signed) and verbal (thumb print) consent from study participants’ parents or guardians and permission (assent) from children aged 12 years and older before the interviews. The participants’ privacy and confidentiality were maintained. Children diagnosed as anemic and who tested positive for helminths infections were referred to the nearest health institution for treatment according to the standard national guidelines [26].

Study area, design, and participants

The study was conducted in the Wonago district of the Gedeo zone in the Southern Ethiopia. The district is 377 km south of Addis Ababa, the capital city of Ethiopia, and 13 km South of Dilla, the capital city of the Gedeo zone. The district has 17 rural and 4 urban kebeles, which is the smallest administrative units. In 2014, Wonago’s population was estimated to be 143,989 people: 71,663 (49.8%) men and 72,326 (50.2%) women. The district is among the most densely populated areas in Ethiopia, with 1,014 people per square kilometre of land area. The district has 26 government health facilities (6 health centers and 20 health posts), 2 private clinics, and 2 drug stores, and more than 36,000 students in 3 urban and 22 rural primary schools. Most residents depend on cash crops of coffee, fruit, and ensete (Ensete ventricosum).

We conducted this cross-sectional survey from February 2017 to June 2017. The study population was schoolchildren and their parents or guardians. Students aged 7–14 years were recruited in schools and their parents or guardian contacted by visiting their homes. Using a three-stage cluster sampling method, we randomly assigned 4 schools to level one, 24 classes (comprising 2,384) to level two, and 864 students to level three. We then randomly included 36 children from each class. When more than one child in a class was living together in the same household, one of them was selected randomly by a lottery method. The household of the student's parents or guardian was identified through a local guide. The study participants’ parents or guardians consented and children assented before enrolment. We replaced participants who dropped out of school after the selection process with participants of the same class, sex, and age. The recruitment process is shown in Fig 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart of study inclusion.

https://doi.org/10.1371/journal.pntd.0008002.g001

Sample size

Since this study was part of a large project aiming to identify school health problems, we considered multiple factors to calculate the sample size using OpenEpi software [27] based on single population proportion [28]. Assuming a 95% confidence interval (CI), the maximum sample size was calculated using proportions of different variables from previous studies (e.g., anemia [27%], stunting [30%], thinness [37%], helminths infection [27%], and skin infection [50%]), 5% precision, and a design effect of 2 to account for multistage sampling [12, 29–32]. We also calculated the sample size using outcome-associated variables, such as under-nutrition (32% prevalence of stunting among female participants), and helminths infection (50% among children who did not wash their hands before meals) [12, 29–32]. Finally, we obtained the maximum sample size using 50% of helminths infection among children who did not wash their hands before meals, and 50% of skin infection. The reason that we included skin infection in the sample size calculation was that this problem was one of our studies planned for a subsequent paper. After adding a 10% non-response rate, we reached a final sample size of 845, the minimum required sample size. We then randomly recruited 864 students.

Data collection tools and procedures

Ten trained enumerators conducted the interviews using a pretested, structured questionnaire that was adapted and developed in English and then translated into the local Gedeooffa language. The interviews were done with children at their schools and with parent at their homes. Training was provided for all personnel participated in data collection, supervision, and data entry process. To minimize potential bias and validate the measurement tools prior to actual data collection, a pre-test was conducted on 42 primary school-aged children in other schools not selected for this study. The supervisors checked data for completeness and consistency onsite.

We assessed individual, parent, household, and school level exposure variables. Individual and household factors were collected from the child or the child’s parents or guardians via interviews and observations of the housing conditions. The individual child factors included sex, age, hygiene behavior, loss of appetite in the past month, de-worming treatment in the past 6 months. Measurements such as weight, height, and haemoglobin were done for children in the school. To measure weight, a digital portable scale (Seca 877, Seca GmbH, Germany) was calibrated to the nearest 0.1 kg. Children were weighed in light clothing and no shoes. To measure height, a measuring board (Seca 213, Seca GmbH, Germany) was calibrated to the nearest 0.1 cm. Children were measured while standing barefoot with parallel feet, heels, buttocks, and shoulders, with their heads held upright, the backs of their heads touching the measuring board, and their hands hanging by their sides. Capillary blood samples were taken for haemoglobin measurement, processed, and examined using standard procedures by trained and experienced laboratory technicians using a HemoCue Analyser Hb 301 (Angelholm, Sweden). Parent factors included the educational level of the mother and father. Household factors included the wealth index, which was constructed using principal component analysis of 14 household assets (electricity, radio, television, mobile phone, table, chair, bed, separate kitchen, cooking place, own land, bank account, toilet facility, floor type, and roof type); family size; source of drinking water; container used to store water; and use of treated water. School factors included access to health education on personal hygiene, absence from school in the past month, and participation in the school food program.

Laboratory procedures

Stool samples were collected, processed, and examined using standard procedures [33, 34]. Samples were collected at school in the early morning and stored in stool cups labelled with an identification code, name, sex, age, and date. The specimens were transported in a cold-box with frozen ice-packs to the nearest health facility, Dilla University Teaching and Referral Hospital, where Kato-Katz and formalin-ether concentration (FEC) techniques were used to conduct stool tests. Single, 41.7-mg thick, Kato-Katz smears were prepared from each stool sample on the same day of specimen collection. Then, 1 g of stool was preserved in 10% formalin solution and processed using the FEC technique [35]. All Kato-Katz slides were examined within one hour of preparation to minimize the risk of hookworm egg disappearance, and then the reading of slides were re-performed to detect and count egg of other helminth infections. The slides were examined by three experienced laboratory technicians. The result of each helminths species from two diagnostic techniques was recorded separately. To reduce possible bias introduced during outcome measurement, 10% of 850 of the test results of Kato-Katz and FEC were randomly selected and re-examined in a blinded fashion to ensure reproducibility of the results. To ensure accuracy of helminth results, the quality control was performed based on World Health Organization (WHO) guideline [36]. According to WHO guideline, re-reading is required if the expert identifies a difference in the egg count of more than 10% and more than four eggs between the readings and discuss the reasons for the discrepancy [36]. However, in the WHO guideline, there is no clear information how to handle differences in presence or absence of helminth eggs [37]. Therefore, we compared the helminths egg count of initial reading with second quality control reading. According to WHO guideline, when the difference in helminths egg count exceed four eggs, re-reading of slides was performed by the third senior laboratory technician. In case of a discordant results between the initial reading and re-reading were confirmed by the third senior laboratory technician. However, the WHO quality control guideline is mainly for quantitative diagnostic methods such as Kato-Katz method, there is no guideline for judging discrepancy of faecal egg counts obtained by the FEC semi-quantitative method for quality control. Hence, we performed the quality control for the FEC method based on the presence or absence of helminth eggs. For the FEC method, a discordant result was considered when there is difference in the presence or absence helminths egg from initial reading to the quality control reading. Results were classified as false-positive if the original result was positive for a specific helminths infection, but the results from the second reading, as well as from the third reading, were negative. Results were classified as false-negative if the original result was negative, but the quality control as well as the result from the third reading, were positive.

Statistical analysis

The data were entered into a database using the double-entry system in Epi-data version 3.1 (EpiData Association; Odense Denmark, 2004). Inconsistencies were cleaned and missing values addressed before analysis. After validation, the data were exported to SPSS version 20 (IBM Corp, 2011) and STATA 14 software (StataCorp LP, College Station, TX, 2015) for analysis.

Descriptive statistics including frequency, percentage, mean, median, range, interquartile range (IQR), and standard deviation (SD) were calculated to describe relevant variables. Cross tabulation was used to calculate the proportions of categorical variables in relation to outcome variables for any helminths infection, for T. trichiura infection, and for A. lumbricoides infection. A wealth index was constructed by using principal component analysis [38] to code the previously listed 14 household assets as 0 (absent) or 1 (present). The internal consistency of the 14 variables was determined (Cronbach alpha of 0.78 and Kaiser-Meyer-Olkin sampling adequacy of 0.8). The socioeconomic indictors (poor, middle, and rich) were categorized based on the first component explaining 28.3% of the variance in the data with an Eigen value of 4.1. We used WHO AnthroPlus 1.0.4 software to calculate height-for-age, weight-for-age, and body mass index-for-age Z scores according to the standard reference for children aged 5–19 year [39]. Stunting was defined as height-for-age Z scores < -2 SD, and thinness was defined as body-mass-index-for-age Z scores < -2 SD [40]. Anaemia was recorded according to the WHO guidelines for school-aged children: haemoglobin < 11.5 g/dl for those aged 5–11 years and < 12 g/dl for those aged 12–14 years. Anaemia was estimated using the haemoglobin value adjusted for altitude [41].

Kappa statistics were used to estimate reliability of the inter-rater agreement of the two readers using 10% of the test results in either Kato-Katz or FEC method. Kappa values were defined as follows: poor = 0.01–0.2; fair = 0.21–0.4; moderate = 0.41–0.6; good = 0.61–0.8; and perfect = 0.81–1 [42]. Kappa values were considered statistically at P < 0.05.

We used a multivariate, multilevel, mixed-effect, logistic regression model to analyze three separate binary outcome variables: the presence or absence of any helminth infections, the presence or absence of T. trichiura, and the presence or absence of A. lumbricoides. Any helminth infections in this study includes (T. trichiura, A. lumbricoides, Taenia species, hookworm species, Strongyloides stercoralis, and Hymenolepis nana). In addition, the eggs detected per Kato-Katz slide (41.7 mg of faeces) was multiplied by a factor 24 to obtain a standard measure of eggs per gram (epg) of stool [43]. The epg was then used as a proxy for estimation of helminths infection intensity. Infection intensity was defined as light (< 5,000 epg for A. lumbricoides; < 1,000 epg for T. trichiura; and < 2,000 epg for hookworm) or moderate (≥ 5,000 epg for A. lumbricoides; ≥ 1,000 epg for T. trichiura; and ≥ 2,000 epg for hookworm) [44]. Two separate count models also were constructed for the T. trichiura and A. lumbricoides fecal egg counts. Thus, using a zero-inflated negative binomial (ZINB) regression model, we performed two part model; the first part has an interpretation as binary outcome model and the second part as a count model. The association between the over-dispersed count outcome with excess zeros and the potential predictors of T. trichiura and A. lumbricoides infection was determined using a ZINB regression model [25].

Multilevel, mixed-effect, logistic regression for modelling infection risk

We used three data hierarchies: school-level, class-level, and individual-level (child or parent). Student participants were clustered within the same class, and classes were nested within schools. We included school and class levels during the analysis and assessed potential confounding and effect modifications using a multivariate, multilevel, mixed-effect, logistic regression model and stratified analysis. Prior to the multivariate regression, we checked the collinearity among exposure variables. We used the presence and absence of any helminths, of T. trichiura infection, and of A. lumbricoides infection as separate outcome variables and conducted the analysis using a multilevel logistic regression model. For all predictors, we applied a simple, bivariate, logistic regression without considering random effects, and a multilevel, logistic regression model with random school and class effects. We also estimated the intra-cluster correlation coefficient for each model within school and class.

Five models were constructed for each outcome variable (any helminths or T. trichiura, or A. lumbricoides infection). Model 1 (empty) had no covariate indicating whether to consider the random-effect model. Model II contained the individual child factors. Model III contained the individual child and individual parent factors. Model IV contained household, individual child, and parent factors. Finally, Model V used multilevel, multivariate, logistic regression to assess individual, household, and school factors. Exposure variables with P values < .25 in the bivariate multilevel logistic regression model were introduced into the model II, model III and model IV.

Variables in model V for any helminths included individual child factors (sex, age, loss of appetite in past month, nail trimming, handwashing with soap before meals, thinness, and anemia), individual parent factors (mother’s educational status), household factors (wealth, source of drinking water, container used to store water, treated water), and school factors (participation in school feeding program). Covariate variables (e.g., sex, age, nail trimming, handwashing with soap before meals, wealth, source of drinking water, using treated water, and participation in a school feeding program) with P values >.25 in the bivariate regression model were retained in the final model to control for confounding.

Variables in model V for T. trichiura infection included individual child factors (sex, age, loss of appetite in past month, eating uncooked vegetable, thinness, and anemia), individual parent factors (mother’s educational status), household factors (wealth, family size, and container used to store water), and school factors (participation in school feeding program). Covariate variables (e.g., sex, age, and wealth) with P values >.25 in the bivariate regression model were retained in the final model to control for confounding.

Variables in model V for A. lumbricoides infection included individual child factors (sex, age, nail trimming, dirt in fingernail, loss of appetite in past month, handwashing with soap after using latrine, de-worming treatment in past 6 months, and anemia), individual parent factors (mother’s educational status), household factors (wealth and source of drinking water), and school factors (participation in school feeding program). Covariate variables (e.g., sex, nail trimming, and wealth) with P values >.25 in the bivariate regression model were retained in the final model to control for confounding.

Zero-inflated negative binomial regression for modelling infection intensity

To examine potential factors associated with infection intensity, a count model was applied using the fecal egg counts for T. trichiura and A. lumbricoides infections. A Poisson model was appropriate for count data, but the assumption of equal variance and mean did not fit to our data, because the mean of A. lumbricoides and T. trichiura eggs was higher than the variance. We have evidence of over-dispersion for T. trichiura A. lumbricoides egg counts. Moreover, zero fecal egg counts for these two infections were more than expected if a Poisson distribution was used [45]. Furthermore, excess number of zeros in our data exceeded those expected under a standard negative-binomial (NB) distribution [45]. Thus, one-part models tend to underestimate the frequencies of zeros and to bias estimation of the covariate effect size [46]. The Vuong test favors ZINB over NB for T. trichiura (Z = 17.5; P < .001), and for A. lumbricoides (Z = 9.4; P < .001) eggs count model, indicating the presence of excess zeroes to be accounted. We thus choose ZINB as the best fitting model. The ZINB model is a two-part model that models an over-dispersed count outcomes with inflated zeros [47, 48]. The ZINB model assumes that the excess zero counts come from a logit model and the counts from a negative binomial model [49]. In other words, the excess zeros in T. trichiura and A. lumbricoides egg count among participants were generated from two separate process. The first process produced only zero counts, corresponds to participants that are free of T. trichiura and A. lumbricoides egg, zero counts for these group of participants considered as true zeros. The second process consisted of participants where T. trichiura and A. lumbricoides egg is actually present but not reported due to sampling zeros or diagnostic technique, zero counts for these group of participants considered as false zeros.

Measure of effect and model fitness

Results were calculated as crude odds ratios and adjusted odds ratios with a 95% CI. Predictors with P values < .05 in the final multilevel, multivariate regression model were reported as statistically significant. The Vuong test was used to compare the ZINB model with a standard negative binomial model and a likelihood ratio test to compare ZINB with a zero-inflated Poisson (ZIP) regression model. The Vuong and likelihood ratio tests with P < .05 favored the ZINB model [46]. Model fitness was checked using -2 log likelihood (deviance) and Akaike information criterion (AIC). The model with the lowest deviance and AIC was used as the final model [46, 50].

Inter-rater agreement for helminths data

We performed reliability test using 10% of 850 sample. The inter-rater agreement of the two readers either Kato-Katz or FEC was checked with Kappa statistics. In case of discordant results, a third reader was used to confirm the analysis. We observed a good agreement between the initial reading and re-reading in either Kato-Katz or FEC method (P < .001). The Kappa values in the Kato-Katz reading were as follows: A. lumbricoides, 0.83 [95% CI: 0.72–0.95]; T. trichiura, 0.88 [95% CI: 0.78–0.98]; Taenia species, 0.87 [95% CI: 0.76–0.98]; and hookworm, 0.86 [95% CI: 0.74–0.98]. Among discordant results between the two readers, 7 were for A. lumbricoides, 5 for T. trichiura, 5 for Taenia, and 5 for hookworm. The proportion of false positive results in the Kato-Katz smear, 1.2% (1/85) were for A. lumbricoides, 5.9% (5/85) for T. trichiura, 3.5% (3/85) for Taenia, and 2.4% (2/85) for hookworm. The proportion of false negative results in the Kato-Katz smear, 7.1% (6/85) were for A. lumbricoides, 0 for T. trichiura, 2.4% (2/85) for Taenia, 3.5% (3/85) for hookworm. According to the WHO guideline, in the Kato-Katz smear, helminth egg counts between initial reading and second quality control reading was compared. Discrepancies in helminth egg counts in the Kato-Katz smear were detected, 8.2% (7/85) for A. lumbricoides, 5.9% (5/85) for T. trichiura, and 5.9% (5/85) for Taenia species. For hookworm, the observed differences in egg counts did not exceed 4 eggs per slide (S7 Table).

The Kappa values in the FEC reading were as follows: A. lumbricoides, 0.91 [95% CI: 0.82–0.99]; T. trichiura, 0.86 [95% CI: 0.75–0.98]; Taenia species, 0.84 [95% CI: 0.71–0.97]; hookworm, 0.86 [95% CI: 0.67–1.00]; and S. stercoralis, 0.81 [95% CI: 0.62–0.99]. Among discordant results between the two readers, 4 were for A. lumbricoides, 5 for T. trichiura, 5 for Taenia, 2 for hookworm, and 4 for S. stercoralis. The proportion of false positive results in the FEC method, 2.4% (2/85) were for A. lumbricoides, 3.5% (3/85) for T. trichiura, 2.4% (2/85) for Taenia, 1.2% (1/85) for hookworm, and 2.4% (2/85) for S. stercoralis. The proportion of false negative results in the FEC method, 2.4% (2/85) were for A. lumbricoides, 2.4% (2/85) for T. trichiura, 3.5% (3/85) for Taenia, 1.2% (1/85) for hookworm, and 2.4% (2/85) for S. stercoralis (S8 Table).

Variation and risk factors for any helminths infection

Predictors of any helminths infection were estimated using a multivariate, multi-level, mixed-effect, logistic regression analysis. The intra-cluster correlation coefficient (ICC) value, calculated in the empty model with no covariate, was 1.2% at the school and class levels and 0.1% in the final model, indicating unexplained variations of any helminths infection prevalence at the school- and class-levels. See the S10 Table of supplementary data for the results of subsequent multilevel models.

In the bivariate, multi-level, mixed-effect, logistic regression model; the following factors had significant associations with any helminths infection: loss of appetite in the past month, thinness, anemia, having a mother or guardian with no formal education, and using an open container for water storage (S9 Table). In the multivariate, multi-level, mixed-effect, logistic regression model analysis, the risk of any helminths infection was higher among children with loss of appetite in the past month (AOR: 1.89 [95% CI: 1.16, 3.08]), thinness (AOR: 1.73 [95% CI: (1.04, 2.90]), anemia (AOR: 1.45 [95% CI: 1.04, 2.03]), a mother or guardian with no formal education (AOR: 2.08 [95% CI: 1.25, 3.47]), and open containers for water storage (AOR: 2.06 [95% CI: 1.07, 3.99]). However, no significant differences were observed between any helminths infection and sex, age, nail trimming, handwashing before meals, eating undercooked vegetables, wealth, source of drinking water, using treated water at home, or participation in a school feeding program. Table 3 and S10 Table shows the details.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Multivariate, multilevel, mixed-effect, logistic regression analysis of any helminths infection among schoolchildren in the Wonago district of Southern Ethiopia, 2017.

https://doi.org/10.1371/journal.pntd.0008002.t003

Variation and risk factors for T. trichiura and A. lumbricoides infections

The intra-cluster correlation value calculated in Model V for T. trichiura was low and insignificant, indicated that the variability in this infection prevalence was not attributable to class or school factors. The S11 Table of supplementary data shows the results of these tests. Similarly, the variability in A. lumbricoides prevalence at the school-level was insignificant. However, the intra-cluster correlation value calculated in Model V for A. lumbricoides infection indicated that 3.2% of the variability in this infection prevalence was attributable to class factors. The S12 Table of supplementary data shows the results of these tests.

For T. trichiura model, all significant variables in the bivariate, multilevel, mixed-effect model also were significant in the multivariate model. The risk of T. trichiura infection was higher among children with loss of appetite in the past month (AOR: 1.76 [95% CI: 1.15, 2.71]), thinness (AOR: 1.73 [95% CI: 1.07, 2.78]), anemia (AOR: 1.53 [95% CI: 1.11, 2.12]), a mother or guardian with no formal education (AOR: 1.94 [95% CI: 1.18, 3.19]), and participation in the school food program (AOR: 1.55 [95% CI: 1.13, 2.12]). Furthermore, there were no statistically significant differences between T. trichiura infection and sex, age, nail trimming, handwashing, eating uncooked vegetable, receiving de-worming treatment in the past 6 months, or wealth (S9 and S11 Tables).

For A. lumbricoides model, all significant variables in the bivariate, multilevel, mixed-effect model also were significant in the multivariate model. The odds of A. lumbricoides infection increased by 92% among anemic children (AOR: 1.92 [95% CI: 1.29, 2.88]). The odds were lower among children who received de-worming treatment in the past 6 months [AOR: 0.57 (95% CI: 0.33, 0.98)] and who used water from a protected source [AOR: 0.46 (95% CI: 0.28, 0.77)]. There were no statistically significant differences between A. lumbricoides infections and age, nail trimming, handwashing, wealth, source of drinking water, and participation in a school food program. The S9 and S12 Tables of supplementary data shows the results.

Zero-inflated negative binomial regression

Negative binomial count model for T. trichiura and A. lumbricoides infections.

Intensity of T. trichiura infection increased among girls (AOR: 1.23 [95% CI: 1.04, 1.45]), and in those using open container for water storage at home (AOR: 1.59 [95% CI: 1.14, 2.22]). The intensity of infection with A. lumbricoides (AOR: 1.08 [95% CI: 1.01, 1.16]) increased with increasing age. Unclean fingernails (AOR: 1.47 [95% CI: 1.07, 2.03]) were associated with increased intensity of A. lumbricoides infection. A habit of nail trimming (AOR: 0.56 [95% CI: 0.39, 0.79]) and handwashing with soap after using the latrine (AOR: 0.68 [95% CI: 0.48, 0.95]) lowered the intensity of A. lumbricoides. The intensity of A. lumbricoides was higher among children in school feeding programs (AOR: 1.97 [95% CI: 1.49, 2.61]) (Tables 4 and 5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Zero-inflated negative binomial regression model for T. trichiura fecal egg count among schoolchildren in the Wonago district, Southern Ethiopia, 2017 (n = 850).

https://doi.org/10.1371/journal.pntd.0008002.t004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Zero-inflated negative binomial regression model for A. lumbricoides fecal egg count among schoolchildren in the Wonago district, Southern Ethiopia, 2017 (n = 850).

https://doi.org/10.1371/journal.pntd.0008002.t005