Study design and sites
The study was performed as a single-blind placebo controlled, parallel group study [1:1]. Public primary schools were purposely selected according to the following criteria: (a) school meal program in place, (b) public day school, (c) at least 280 children aged 6–12 years, and (d) accessibility by car. The primary schools with comparatively large number of students in the study area where the authorities and the head teachers expressed support and were open for the intervention were approached.
Under these criteria, Kakumuti Pre- and Primary School was selected in rural Kitui-West (Sub-county), Kitui (County), Eastern Province of Kenya, approximately 165 km away from Nairobi. About 430 children attended the school, which had a self-governed school meal program. Kitui County is of marginal agricultural potential, prone to droughts [30] and the stunting prevalence in children under five is among the highest in the country [2]. Kitui is considered a low-risk area for malaria transmission [9].
Kitui belongs to the baobab belt [29], except of certain sub-counties such as Kitui-West where the few baobab trees do not produce fruits. Around the school, there were no baobab products identified in the markets, and baobab fruits used for the study were sourced from another area, namely Kyamatu location in Kitui-East Sub-county. The intervention started in May 2018, which is generally the end of the long rainy season. The average rainfall was above normal in 2018, and general food security improved during the intervention period, which fell in the postharvest season of staples and pulses.
Sampling study participants
After obtaining official research permits and consent from the school administration, locally trained project assistants described the study in the local Kikamba language to caregivers of the children eligible for screening. Only children whose caregivers provided written informed consent (signature or fingerprint) were invited for the screening. The assistants orally informed these children about the study objective and procedure of the upcoming exercise, and children approved the reception of the information with their signature. Children’s oral consent and their signature were prerequisite for any further interview and examination. Registered nurses and laboratory technicians performed the clinical screening in a separate room and administered a dewormer (Albendazole USP 400) to all children. Thus, intestinal blood loss due to helminth infections was prevented. Eligible participants were apparently healthy children aged 6–12 years with lowest adjusted Hb level at screening. Exclusion criteria are shown in Fig. 1.
Intervention
The intervention took place daily for a total of 83 days from May to July 2018. In addition to a standardized school meal, study children received either one cup of a drink with BFP or one cup of an isoenergy drink without BFP. The standardized portion of the school meal had an estimated iron content of 7.6 mg per portion, mainly from beans (NutriSurvey2007) (Table 1). BFP is rich in vitamin C [22, 31]; therefore, we expected an improvement in the bioavailability of iron from the school meal.
Table 1 Energy and nutrients of one portion of school meal (calculated in NutriSurvey©) The preparation of the school meal (mixed beans, maize, iodized salt, and vegetable oil) was standardized. Baobab fruits were delivered from Kyamatu and processed on a daily basis by trained local field assistants. They cleaned the fruits with a wire brush first and with a soft brush afterwards to remove the hair from the outer shell. The fruits were then cracked with a machete, and those with any spots (insects, mold, etc.) inside the fruit were discarded applying the two man rule. The pulp-seed mix was removed from the shell, ground in a mortar to separate pulp (which is in the form of a powder) from the seeds, and the powder sieved twice in succession.
About 20–30 min prior to distributing the drinks to the children, all ingredients for the intervention drink were blended. A weighted cup of intervention drink contained 20 g BFP, 5 g honey, 7 drops of Mango Liquid Flavour Drops (SygLabs, Germany), and 200 ml of bottled water. The isoenergy control drink consisted of 3 g commercially available corn starch, 10 g honey, 5 Mango Liquid Flavour Drops (SygLabs, Germany) and 220 ml of bottled water. The corn starch was boiled in 2 l of bottled water and mixed with the remaining ingredients after cooling. The field assistants weighed 220 g of either baobab drink or control drink in cups. The cups were coded with different colors to differentiate between intervention and control drink. Table 2 shows the nutrient composition of the intervention and control drink. The field assistants observed the children during the consumption to avoid any exchange of food and drinks and recorded the amount of food and drink consumed by each child.
Table 2 Energy and nutrition composition of 220 ml intervention and control drink (calculated in NutriSurvey©) During the intervention, eight BFP samples were taken, stored in the fridge, and protected from light until a laboratory analysis was performed. The vitamin C was determined in triplicate using the method of Vikram et al. [32] with slight modifications. The samples was analyzed using a Shimadzu HPLC (20A Model, Tokyo, Japan), fitted with a ODS-C18 (250 cm × 4.6 mm × 5 µl) column, CTO-10AS VP oven, SPD-M20A diode array detector, DGU-20ASR prominence degassing unit, CBM-20A prominence communications bus module, SIL-20A HT prominence auto sampler and an LC-20AD prominence liquid chromatograph. The mobile phase contained 0.8% metaphosphoric acid at a flow rate of 0.8 ml/min. The injection volume used was 20 µl at a wavelength of 266 nm and oven temperatures of 30 °C. The retention time of pure ascorbic acid was used to identify ascorbic peaks in sample chromatographs. Iron, zinc, calcium, and magnesium were analyzed in duplicate with an inductive coupled plasma-optic emission spectrometer as described by Habte et al. [33]. Table 3 shows the BFP composition.
Table 3 Baobab fruit pulp composition Allocation into the intervention and control groups
The allocation of participating children into either the intervention or control group was done using the stratified random sampling in SPSS. Participants were stratified according to sex (30 male and 36 female), Hb level above and below median for male (md = 11.9 g/dl) and female (md = 11.8 g/dl), respectively, resulting in four blocks. Among each block, a random allocation in intervention and control group was performed with the Mersenne Twister random number generator conducted in SPSS (V 24) according to age in years.
Sample size
A total of 33 children were allocated into each group, with an assumed dropout of 10%, and a prevalence of homozygote and mixed forms of sickle cell disease and α-thalassemia of 6% (own data), and 76% of the children with Hb-levels > 11.5 g/dl [9]; we aimed to have data of 56 children be available at the endpoint. Given this sample size, we expected to detect medium to strong effects (Cohen’s d = 0.76) with alpha = 0.05 and power = 80, two-sided. The number of probands was expected to translate to 15% decrease in mean sTfR in the intervention group with an unchanged mean sTfR in the control group (mean baseline and control groups: 8.48; mean intervention group at endline: 7.208) with a standard deviation of 1.32 at both time points and a correlation of 0.25 between time points. These values were copied from Perignon et al. [34] as we did not have our own data when the study was planned; the assumed correlation of 0.25 is a conservative assumption.
We initially aimed to screen a total of 273 children but found 249 children aged 6–12 years only, of whom 223 children participated in the screening. When we selected the school, we were only given the numbers of children per class. Information on child age was provided only after the school had been selected.
Blood sample collection and analysis
To minimize any discomfort, a local anesthetic ointment containing lidocaine and prilocaine (EMLA™, AstraZeneca, Cambridge, UK) was applied onto the area of skin to be numbed prior to pricking. During screening, capillary blood samples of children were taken for two subsequent Hb measurements using a HemoCue HB 301 photometer device (HemoCue AB, Ängelholm, Sweden). The maximum tolerated difference between the measurements was 0.5 g/dl. The mean value was used to determine individual Hb levels at screening.
At baseline, registered nurses took from each child a non-fasting venous blood sample, which was spun within 30 min to obtain 50–100 µl serum. The serum was pipetted into labeled 0.2 ml Multiply® PCR tubes (Sarstedt Inc., US). In the field, samples were either stored at low temperature for a maximum of 7 days and then put into a freezer or stored in a freezer on the same day [35]. The samples were analyzed for serum ferritin (FER), soluble transferrin receptor (sTfR), acidic glycoprotein (AGP), and C-reactive protein (CRP) levels using a Sandwich ELISA at the VitMin Lab, Willstaett, Germany, [36]. Hb concentrations were measured immediately after phlebotomy using a HemoCue HB 301 photometer device (HemoCue AB, Ängelholm, Sweden).
Hb was adjusted for altitude and anemia, which is defined as adjusted Hb < 11.5 g/dl in children aged 7–11 years and < 12 g/dl in children aged 12 years [37]. Iron deficiency was defined by depleted iron stores (adjusted FER < 15 µg/L) [38] and tissue iron deficiency by high serum sTfR (> 8.3 mg/L) [36].
CRP and AGP were assessed for the identification and classification of inflammation: incubation (CRP levels > 5 mg/L and AGP levels ≤ 1 g/L), early convalescence (CRP levels > 5 mg/L and AGP levels > 1 g/L), and late convalescence (CRP levels ≤ 5 mg/L and AGP levels > 1 g/L). FER was adjusted for inflammation stage with correction factors for each inflammation stage [39].
Genotyping for sickle cell trait and the 3.7 kb α-globin deletion that most commonly causes α +-thalassemia in African populations was conducted by PCR [40, 41] at the KEMRI-Wellcome Trust Research Laboratories in Kilifi, Kenya, as described in detail previously.
Anthropometric measurements
At screening, nurses received an additional instruction on how to assess the mid-upper arm circumference (MUAC) with a measuring tape that allows for an assessment to the nearest 0.1 cm. Moderate undernutrition was defined at MUAC < 14.5 cm and < 18.5 cm for children aged 6–9 years and 10–12 years, respectively [42].
To control for a potential influence of anthropometric developments from baseline to endline, we assessed weight and height at baseline and endline. Children were checked for edema and weighed without shoes and in light clothing to the nearest 0.1 kg, using a Seca® UNICEF scale (SECA 874, Hamburg, Germany). Body height was measured to the nearest 0.5 cm using a calibrated SECA® height scale (SECA 213, Hamburg, Germany). Weight and height measurements were repeated twice with a maximum tolerable difference of 0.1 kg for weight and 0.5 cm for height.
The weight-for-age z-score (WAZ), body mass index-for-age z-score (BAZ), and height-for-age z-score (HAZ) were calculated using Anthro Plus, the anthropometric calculator module based on the 2007 WHO reference for children aged 5–19 years [43, 44]. Stunting, underweight, and thinness were defined by HAZ, WAZ, and BAZ below − 2 SD, respectively. The school provided data on the age of the children, which was crosschecked with primary caregivers. If the primary caregiver could not verify the date of birth, WAZ, BAZ, and HAZ were not calculated.
Assessment of nutrient intake
To control dietary intake outside the study setting, we conducted 24 h recalls during the 1st (t1), 5th (t2) and 11th (t3) weeks. Interviewers with a formal qualification in nutrition or food science, as well as literate in English and the local language, were trained on applying standardized 24 h recalls with primary caregivers. The questionnaire and 24 h recalls were translated into the local Kikamba language and retranslated into English, reviewed during the 6-day interviewer training, pre-tested, and modified to ensure meaning equivalence of the questions. Pre-testing was carried out among households with children not involved in the study.
The interviews for the multiple pass 24 h recalls consisted of (a) listing all foods and drinks consumed the day before the interview, (b) gathering detailed information about each food or recipe for dishes, (c) estimated quantification of the amount of consumed food/drink and used ingredients for the recipes, and (d) reviewing the information with the respondent at the end of the recall. Specially designed photo books were developed to estimate the quantity of intake of food and drinks. The interviewer also used local measuring tools such as spoons and cups for quantifying portion sizes.
Table 4 shows the recommended dietary allowances for energy, vitamins, and trace elements for school-aged children. Individual energy adequacy ratios were calculated as total energy intake divided by sex, and age-specific energy requirements, based on the recommendations of the FAO/WHO/UNU expert committee on human energy requirements [45]. The nutrient adequacy ratio (NAR) was determined for vitamins C, iron, zinc, calcium, and magnesium. Individual NARs were calculated as a total intake of the nutrient divided by the recommended daily allowance (RDA) for that nutrient, based on intakes recommended by the Kenyan Ministry of Health [46]. Table 1 shows the energy and nutrients of one portion of school meal that was provided on a daily basis in addition to the drink.
Table 4 Recommended dietary allowances of energy, vitamin, minerals, and elements for school-aged children Data management and statistical analysis
Data entry and validation via double entry was performed for anthropometry and Hb, as well as for the 24 h recalls. The country-specific food database for Kenya was loaded into the NutriSurvey nutrient database. Missing food items were supplemented from the Tanzania Food Composition Tables [47] and the Food Data Central of the United States Department of Agriculture [48].
Data management and statistical analysis were performed using SPSS software (Version 24, IBM Corp., Armonk, NY, USA).
The mean intake of energy and nutrients, determined through NutriSurvey, at time points t1, t2, and t3 was calculated for each child. Normality of distributions was evaluated using the Shapiro–Wilk test. As most continuous variables (micronutrient status and energy and nutrient intake) had heavily skewed distributions, descriptive statistics for continual variables are presented in the median and interquartile range (IQR). For this data, a non-parametric median test was applied for comparing data from intervention and control groups at baseline (blood parameters and anthropometric data) and at t1, t2, and t3 (mean energy and nutrient intake). The strength of association was calculated with Cramer’s V, which equals r. For approximately normally distributed data, means and standard deviations are presented, and the independent t test was applied.
Outliers in development (baseline to endline) of Hb, FER, and sTfR were identified as described by Tukey [49] and excluded from the analysis (outliers: n(Hb) = 0; n(FER-intervention) = 3, n(FER-control) = 1; n(sTfR-intervention) = 0, n(sTfR-control) = 1).
The baseline and endline data on FER and sTfR were log transformed and used to calculate the development between baseline and endline to apply the independent samples’ t test for differences between groups and the paired t test for development within the group. The effect size for the independent t test was not calculated (differences not significant) and paired t test was calculated using Cohen’s d.
Friedman’s ANOVA was conducted to test differences in dietary intake between t1, t2, and t3 (related samples and pairwise comparison). The general linear model was used to evaluate the effects of time (baseline/endline), group (intervention/control), age (in years at baseline), change in weight (endline—baseline), sex (male/female), and genotype (heterozygote carrier of α-thalassemia/non-carrier) on Hb, LN(FER), and LN(sTfR) and the interaction of time with each variable, respectively. For Hb, we also analyzed the interaction time*group*genotype. Variables were tested for associations with non-parametric Spearman’s correlation. A p value of < 0.05 was considered statistically significant.
Ethical approval
The institutional review board of the Faculty of Medicine at Justus Liebig University Giessen, Germany (197/16) and the AMREF Ethics and Scientific Review Committee (AMREF- ESRC P313/2017) Kenya approved the Baobab Nutrition Intervention Study under the Kenyan National Commission for Science, Technology, and Innovation research permit (NACOSTI/P/18/60305/20841). The study was registered with the German Clinical Trials Registry (DRKS00011935). Official permission and approval from Kenya government authorities was obtained, and the municipal and governmental authorities in Kenya approved for the implementation of the study.
Written informed consent of primary caregivers and schoolchildren via signature or fingerprint was obtained prior to data collection. The ethics committees also approved the consent format prior to data collection. The management school board comprising the parent’s representative, representatives from the Kenya National Union of Teachers, church and local leaders were informed about the study and gave their verbal consent after participating in a stakeholder meeting to create awareness on the study.