A macro- and micronutrient-fortified complementary food supplement reduced acute infection, improved haemoglobin and showed a dose-response effect in improving linear growth: A 12-month cluster randomised trial

Inadequate protein quality may be a risk factor for poor growth. To examine the effect of a macronutrient-micronutrient supplement KOKO Plus (KP), provided to infants from 6 to 18 months of age, on linear growth, a single-blind cluster-randomised study was implemented in Ghana. A total of thirty-eight communities were randomly allocated to receive KP (fourteen communities, n 322), a micronutrient powder (MN, thirteen communities, n 329) and nutrition education (NE, eleven communities, n 319). A comparison group was followed cross-sectionally (n 303). Supplement delivery and morbidity were measured weekly and anthropometry monthly. NE education was provided monthly. Baseline, midline and endline measurements at 6, 12 and 18 months included venous blood draws, diet, anthropometry, morbidity, food security and socio-economics. Length-for-age Z-score (LAZ) was the primary outcome. Analyses were intent-to-treat using mixed-effects regressions adjusted for clustering, sex, age and baseline. No differences existed in mean LAZ scores at endline (-1·219 (sd 0·06) KP, -1·211 (sd 0·03) MN, -1·266 (sd 0·03) NE). Acute infection prevalence was lower in the KP than NE group (P = 0·043). Mean serum Hb was higher in KP infants free from acute infection (114·02 (sd 1·87) g/l) than MN (107·8 (sd 2·5) g/l; P = 0·047) and NE (108·8 (sd 0·99) g/l; P = 0·051). Compliance was 84·9 % (KP) and 87·2 % (MN) but delivery 60 %. Adjusting for delivery and compliance, LAZ score at endline was significantly higher in the KP v. MN group (+0·2 LAZ; P = 0·026). A macro- and micronutrient-fortified supplement KP reduced acute infection, improved Hb and demonstrated a dose-response effect on LAZ adjusting consumption for delivery. This was a cluster randomised single-blind study with three groups (KP, MN and NE). The study design ensured participant blinding and prevented contamination. One group received KP (KP group) with NE, a second group received MN (MN group) with NE and a third group received NE alone (NE group). Due to ethical considerations, there was no control group, but we followed a separate group of communities cross-sectionally to ascertain secular trends in the primary outcome, i.e. the LAZ. The ingredient composition and nutrient content of KP (sachet per d) are presented in Tables 1 and ​and2,2, respectively. The development of the formulation is discussed elsewhere(,20,21). KP contains soya powder, sugar and oil along with the essential amino acid lysine and a micronutrient premix. KP was formulated as a complement that aids meeting the WHO complementary feeding guidelines(,22), the FAO/WHO micronutrient recommended nutrient intakes (RNI) and the WHO protein and essential amino acid requirements for the 6- to 24-month age group(,8,23). It achieved 30 % of the total recommended energy requirement, 60 % of total protein and 40 % of total fat requirements from complementary foods. In addition, an assessment of amino acid and micronutrient composition shows that the supplement met 35−55 % of essential amino acids and 50–150 % RNI of micronutrient needs based on the total daily requirements. The micronutrient premix provided 50–150 % RNI in both KP and MN sachets. KP was produced in Ghana and the micronutrient premix for both supplements was produced in South Africa. Formulation of KOKO Plus (g per sachet) Macro- and micronutrient composition of KOKO Plus per sachet compared with macronutrient requirements from complementary food (per d) and amino acid and micronutrient needs (per d) The study protocol was reviewed and approved by the Institutional Review Boards of the Ghana Health Service (GHS) (Accra, Ghana) and the Noguchi Institute for Medical Research, Accra, Ghana. Written informed consent was obtained from both parents except in single parent households. A data safety monitoring board (DSMB) reviewed study outcomes on a quarterly basis. No interim analyses were planned or stopping rules defined. Sample size calculations were based on change in LAZ and diarrhoeal morbidity, with change in LAZ being the primary outcome. The sample size per group was 301, with thirteen clusters per group (about twenty-three participants/cluster, equal number of clusters). This would detect a 0·5 cm change in length (1·2 sd) in infants provided an energy-containing v. non-energy-containing micronutrient supplement using a design effect of 1·66, intraclass correlation of 0·03, power of 0·80, α of 0·05 and an attrition rate of 15 %. The sample size was also sufficient to detect a minimum of a 0·19 change (0·54 sd) in LAZ (required sample size: 298 per group). This LAZ change estimate was the average change observed by Adu-Afarwuah et al.(,24) in a three-arm intervention study with a cross-sectional non-intervention group comparing one macro- and micronutrient-fortified spread with two micronutrient formulations(,24,25). The subjects were from communities in three districts of the Central region of Ghana. These were districts with the highest rates of moderate and severe acute malnutrition. A population size greater than 1000 households was defined as the minimum criterion for study inclusion. At total of sixty-one communities, each serving as a cluster, fulfilled the criterion. A total of thirty-nine communities were randomly selected using the Microsoft Excel random number function (RAND) by a research associate. Following this, a new random sequence was generated using RAND followed by block randomisation (number of blocks = 4) and the clusters were randomly assigned to one of three groups (KP, MN and NE) by the same research associate. Another eleven communities were randomly but separately selected from the remaining list. Changes occurred to the total number of clusters per group as study implementation began. During the community sensitisation process, we found one of the clusters in the KP group subdivided into two different communities and four clusters in the NE group merged into two. Thus, the total number of clusters were thirty-eight not thirty-nine, with fourteen in the KP, thirteen in the MN (original allocation) and eleven in NE group. Fig. 1 shows the actual study flow, loss to follow-up and ‘drop outs’ by individual participants. The study was conducted from January 2013 through to February 2015 when the last infant graduated from the intervention study. Study participants and follow-up by group (KOKO Plus (KP), micronutrient powder (MN) and nutrition education (NE)). * Lost to follow-up includes deaths and severe acute malnutrition (SAM). The University of Ghana implemented the study. The intervention period was 12 months from infant age 6−18 months and was delivered at the community level to ensure blinding and prevent contamination. The KP and MN supplements were formulated for daily consumption with instructions for use to mothers in the communities assigned to KP and MN groups, respectively. The distribution was conducted by a local non-governmental organisation (NGO) working with community health volunteers. The distribution team was expected to visit each community on a weekly basis and deliver the supplements through the community health volunteers. All the mothers enrolled in the study were followed by the community health volunteers who live in the communities. The NE materials were adapted from the Good Life project, a US Agency for International Development behaviour change project conducted from 2009 to 2013, to support GHS in areas of family planning, maternal and child health, malaria, nutrition, water and sanitation. Modifications were made to the training materials with specific modules on supplement use(,26). The NE component included monthly sessions with mothers and infants with role-plays, activities and cooking demonstrations conducted in conjunction with GHS volunteers in each community irrespective of treatment group. In each community irrespective of intervention group, all mothers with newborn infants (0–3 months of age) who attended the mother support group were invited to participate in the study. This was to encourage mothers to participate in monthly nutrition education sessions and to continue exclusive breastfeeding. When eligible (at 6 months of age), dyads were enrolled into the intervention. Inclusion criteria were singleton term birth, exclusively or predominantly breastfed, parents planning to live in the community for a period of 12 months and willing to participate for the entire study period and written informed consent. Exclusion criteria included severe anaemia (Hb <70 g/l) or severe acute malnutrition (MUAC <110 mm)(,27,28). Infants were assessed for severe anaemia and acute malnutrition at each time point (baseline (B), midline (M) and endline (E)) and, if diagnosed, referred for routine medical care and excluded from participation. Anthropometric measurements were collected monthly and included length (Infant/Child ShorrBoard®; Weigh and Measure, LLC; http://www.weighandmeasure.com/), weight (Seca 874 digital scale; http://www.seca.com/en_mw/products/all-products/product-details/seca874.html), MUAC (Child MUAC Tape; Weigh and Measure, LLC), subscapular and triceps skinfolds (Holtain T/W skinfold caliper; http://www.holtain.com/tw.php) and head and chest circumference. The digital scales were tested weekly for accuracy using standard weights. Motor development assessments were conducted using the WHO motor development skills framework(,29). A single 24-h diet recall and semi-structured questionnaires were administered to assess change in diet, socio-economic status, infant and young child feeding practices, morbidity and household food security at B, M and E. The 24-h diet recall was developed and implemented in different studies within the University of Ghana and was contextualised to the local diet. A series of locally tested and validated household measures were used for ascertaining portion sizes. Dietary data were cleaned with all data in household measures converted into grams. Supplement compliance and morbidity questionnaires were administered weekly. Both paper and electronic forms were utilised. Data were uploaded daily through the cell phone network, stored on Formhub and ONA (the Formhub system began having problems in mid-2014 and stopped being maintained by developers. ONA is an identical system, which made the switch seamless). All data cleaning and analysis was done using Stata 13.1 (StataCorp LLC). One venous blood draw (3 ml) and a fingerprick (Hemocue 301), to assess severe anaemia (<70 g/l) (for screening purposes), were collected at B, M and E(,30). Sample collection utilised butterfly needles (21/23 gauge) and K2EDTA vacutainers (catalogue no. 368841; Becton, Dickinson and Company; https://www.bd.com/resource.aspx?IDX=7220) for whole blood and plasma analyses and Trace Element Serum Separator Tube vacutainers (BD; catalogue no. 368380) for serum analyses. Samples were immediately placed in a super cooler tube rack to keep sample temperature at 4°C(,31) and transported back to the laboratory within 5 h of collection. A quantity of 50 μl of K2EDTA blood was aliquoted into an Eppendorf tube for whole blood count including Hb concentration (ABX Pentra-60 haematology machine; Horiba Medical; http://www.horiba.com/us/en/medical/products/hematology/abx-pentra-60/). Hb concentrations measured clinically were used for subsequent statistical analyses while Hb measured with HemoCue were used for screening infants. The rest was centrifuged at 3000 g (14 min at 4°C) and plasma aliquoted into Eppendorf tubes and stored at −20°C for plasma amino acid analysis (to be reported in a separate paper). Samples collected in trace element-free tubes were rested at room temperature for 30 min, centrifuged at 3000 g (15 min, 25°C), and aliquoted using sterile metal-free pipette tips (200 µl) into one DNase, pyrogen-free microcentrifuge tube (Thomas Scientific) stored at −80°C for Zn analysis and three Eppendorf tubes stored at −30°C. Serum Zn (Elemental Analysis Lab, David Killilea, University of California) was measured using HPLC, IGF-1, serum cortisol and prealbumin using ELISA (IGF-1 and Cortisol kits from DRG; Prealbumin kit from Abcam) (University of Cape Coast, Ghana), serum transferrin receptors, serum retinol binding protein, serum ferritin, CRP and AGP using sandwich ELISA (Juergen Erhardt)(,32). The primary outcome of this study was change in mean LAZ from 6 to 18 months of age in infants of the KP group compared with those of the MN and NE groups. Anthropometric indices (LAZ, WAZ and weight-for-length Z-score (WLZ)) were computed using the WHO 2006 growth reference charts (WHO macro, STATA)(,33) as were the household insecurity access score (HFIAS)(,34), maternal BMI and dietary diversity scores(,35). Duplicate anthropometric measurements were checked for discrepancies. A total of six observations were flagged as implausible Z-scores by WHO 2006 definitions and were coded as missing. Outlier tests on an infant's longitudinal measurements were conducted using a linear regression model which computed standardised residuals for each data point. A standardised residual ≥ +2·5 or ≤ −2·5 was considered as an outlier and replaced through imputation. A new regression was run and used to impute the omitted length measurement with a new predicted value. This process was repeated using a less strict cut-off of ≥ +3·0 or ≤ −3·0 standardised residual. Based on the review of the findings, the study statistician recommended using the more stringent cut-off of ≥ +2·5 or ≤ −2·5. Across all the anthropometric data and intervention groups (n 9161 observations, including measurements at baseline (B), midline (M) and endline (E) and monthly measurements), a total of seventy-six data points were identified as outliers and replaced through imputation. All analyses were intent to treat. To verify the randomisation assumption, differences in mean values across three groups at B were tested using linear mixed-effects regression models adjusting for clustering. The difference in difference across groups and intervention period was tested using mixed-effects linear regression models adjusting for clustering, intervention group, B values and repeated measures. For all models (both primary and secondary outcomes), random effects included in the model were the cluster variable and the individual ID while baseline and demographic variables (determined using step-wise regression and vary by each model) were included as fixed effects. Any additional covariates in the models were included based on the specific outcome measure. The primary outcome was LAZ. Secondary outcomes included WAZ, WLZ, MUAC, serum Hb, anaemia prevalence, serum ferritin (unadjusted and adjusted for inflammation), serum Zn, serum cortisol, serum IGF-1, serum retinol binding protein, serum CRP and serum AGP, acute and chronic infection status and prevalence, and prevalence of fever and diarrhoea. For LAZ and all other anthropometric outcomes, we computed two separate models. One is the B-E model, where we estimated changes from B (6 months) to M (12 months) and B to E (18 months of age) using the B, M and E data for anthropometry. The second is the monthly model where we computed marginal treatment effects over an 18-month period to examine the rate of change in LAZ, WAZ and WLZ using the monthly anthropometric data. Separately to assess secular differences between the three arms and the comparison group, we computed LAZ, WAZ and WHZ scores for each of the time points (6, 12 and 18 months) and examined change at E adjusting for B, age, sex and clustering using mixed-effects models. We used the two inflammation markers CRP and AGP to determine prevalence of acute and chronic inflammation. Onset of acute inflammation was defined as CRP >5 mg/l while presence of chronic inflammation was defined as AGP >1 g/l as defined and utilised by several studies(,36–40). Prevalence of acute and chronic inflammation was computed for B, M and E and change in prevalence was tested using mixed-effects logistic regression analysis. We further utilised the CRP and AGP data and above-noted cut-offs to compute a variable for the four infection stages as defined by Thurnham et al.(,39). For the biochemical markers, models were adjusted for age, sex, B value, community clustering and the infection stage(,39). We tested two approaches with serum ferritin: one where the unadjusted serum ferritin was modelled with infection stage (at the different time points) as a covariate and a second where the serum ferritin itself was adjusted. For Hb, we examined the difference across groups between children with or without acute infection at E, adjusting for chronic infection and sex, given the high risk of malaria in this population(,41,42). Serum ferritin at each time point was adjusted for inflammation at that time point using the four-stages method as recommended by Thurnham et al.(,39). For the morbidity markers, while data on presence or absence of malarial parasite was not collected, we assessed morbidity using two common measures: prevalence of fevers (all fevers including malarial) and diarrhoeal episodes at B, M and E, again using mixed-effects models. Diarrhoea is defined as three or more loose or liquid stools per d. Episodes of diarrhoea are considered separate if there are three or more consecutive diarrhoea-free days. Prevalence of fever was also assessed at B, M and E as presence of any fever as reported by the caregiver in the past week. While one of the outcome measures was to examine diarrhoeal morbidity using the longitudinal (weekly) data, we were unable to compute these indicators due to significant missing data and thus report only on the B-M-E changes in both acute infection and diarrhoeal morbidity using mixed-effects logistic regression models. Nutrient intake analysis and dietary diversity score computations were conducted. Nutrient intake was calculated using the Research to Improve Infant Nutrition and Growth (RIING) food composition table of Ghanaian foods compiled from three different data sources including FAO, US Department of Agriculture (USDA) and data used in previous research in Ghana (RIING food composition database, Nutrition Department, University of Ghana). The database contains 306 foods with twenty-nine nutrients (macro- and micronutrients). Nutrient intakes were calculated using a SAS (version 9.3; SAS Institute) program. Mean intakes and corresponding standard deviations were estimated at B, M and E. Differences were tested using ANOVA. Compliance as defined by total used sachets divided by total delivered sachets was 86·2 % in the KP group and 88·4 % in the MN groups, indicating that if the mother received the sachet, the sachets were utilised at a similar rate across the two groups (Table 3). However, mean consumption (total used sachets) across both groups was 186 supplements (181 in KP, 190 in MN), much lower than the expected 365 supplements over 52 weeks. Thus, while compliance was high, this was a measure only of the total received supplements. A review of the delivery and distribution logs showed that, on average, mothers received only about two-thirds of the expected 365 supplements. Supplement delivery was hampered due to various reasons – including inaccessibility to mothers, inaccessibility to sites during the monsoons among others. Delivery and reported consumption of supplement and overall compliance during the trial period, in the KOKO Plus and micronutrient groups (Mean values and standard deviations; medians and percentages) Supplement consumption, i.e. study adherence, was thus affected by delivery. As high rates of non-adherence can lead to underestimates of treatment effect(,43), we examined the relationship of supplement consumption and each outcome measure (both primary and secondary). As short-term variation in supplement consumption (i.e. 1 week or 1 month) would not be observable, we included a variable that represents total supplements consumed by the child over the duration of the study. As the total supplement consumed by an infant over the study would have had a different effect on outcomes at M compared with E, this variable was modelled as an interaction with time and intervention/treatment group. The interaction term of time, treatment and consumption allowed for the effect of supplement consumption (a time-invariant variable) to be modelled differentially over time. Thus, we conducted consumption modelling using mixed-effects regression models and we estimated predicted outcomes at different time points across a range of different levels of supplement consumption which represented adjusted means over time at those different levels.