Effectiveness of screening and treatment of children with severe acute malnutrition by community health workers in Simiyu region, Tanzania: a quasi-experimental pilot study

Health system constraints hamper treatment of children with severe acute malnutrition (SAM) in Tanzania. This non-inferiority quasi-experimental study in Bariadi (intervention) and Maswa (control) districts assessed the effectiveness, coverage, and cost-effectiveness of SAM treatment by community health workers (CHWs) compared with outpatient therapeutic care (OTC). We included 154 and 210 children aged 6–59 months with SAM [mid-upper arm circumference (MUAC) < 11.5 cm] without medical complications in the control and intervention districts, respectively. The primary treatment outcome was cure (MUAC ≥ 12.5 cm). We performed costing analysis from the provider’s perspective. The probability of cure was higher in the intervention group (90.5%) than in the control group (75.3%); risk ratio (RR) 1.17; 95% CI 1.05, 1.31 and risk difference (RD) 0.13; 95% CI 0.04, 0.23. SAM treatment coverage was higher in the intervention area (80.9%) than in the control area (41.7%). The cost per child treated was US$146.50 in the intervention group and US$161.62 in the control group and that per child cured was US$161.77 and US$215.49 in the intervention and control groups, respectively. The additional costs per an additional child treated and cured were US$134.40 and US$130.92, respectively. Compared with OTC, treatment of children with uncomplicated SAM by CHWs was effective, increased treatment coverage and was cost-effective. This study was conducted in Simiyu Region, northern Tanzania, where in 2018, 4.6% of children were acutely malnourished with 0.5% being severely acutely malnourished16. According to the 2012 census, Simiyu Region had a population of about 1.6 million inhabitants and was divided into five districts. In consultation with the local health authorities, three rural wards (Sakwe, Ihusi and Mwadobana) in Bariadi District and three rural wards (Malampaka, Busilili, Shishiyu) in Maswa District were selected purposively as intervention and control areas, respectively. In selecting the intervention and control areas, the following factors were considered: study logistics, distance between the two areas to minimize contamination, comparability between the two areas in terms of the population size, expected number of SAM cases, and health infrastructure—number of CHWs working on the Next Generation Programme, number of health facilities, and distance between wards and SAM treatment centres. The intervention wards had a population of about 45,200 people distributed in 11 villages and served by 13 CHWs, three dispensaries and one health centre. The control wards had a population of about 35,800 people distributed in nine villages and served by 11 CHWs, three dispensaries and one health centre. This is a parallel two-arm non-inferiority quasi-experimental pilot study. All children aged 6–59 months with SAM and without medical complications were eligible for inclusion if their primary caretakers provided consent. Eligible children were recruited in the community by CHWs in the intervention wards or by formal health workers in health facilities in the control wards. Only children with good appetite, without severe oedema and no underlying medical condition and/or complications were eligible for enrolment in the study. In the intervention area, CHWs screened children for SAM by measuring their mid-upper arm circumference (MUAC) and those with MUAC < 11.5 cm or mild/moderate oedema were classified as having SAM and treated at home using RUTF, with the dosage based on a child’s body weight. CHWs followed up enrolled children through weekly home visits to replenish their RUTF and to monitor their progress by assessing their weight, MUAC, and medical symptoms. In the control wards, CHWs screened and referred malnourished children to nearby health facilities for treatment by health workers according to the standard national guidelines10. Caretakers could also take their children directly to health facilities. Health workers enrolled children in the study using criteria similar to that used in the intervention district. Supplemental Fig. S1 shows the flow chart used in this study (adapted from national guidelines10) for screening and management of children with acute malnutrition by CHWs. All enrolled children were followed up—either by CHWs in the intervention wards or health care workers through OTC clinics in the control wards—until they exited the study after experiencing one of the study outcomes. Prior to the intervention, CHWs and their supervisors (who included the program staff and health facility staff who usually supervise CHWs in their catchment areas) were adequately trained to screen and manage children with SAM. The training, which covered both theory and practice, was delivered by nutritionists from the Tanzania Food and Nutrition Centre and aimed to impart knowledge and skills in management of SAM among children younger than five years old at the community level. CHWs and their supervisors from the intervention area received further training on home treatment of children with SAM without medical complications. We defined study outcomes in a standard way in both the intervention and control groups. The primary study outcome was cure from SAM, defined as MUAC ≥ 12.5 cm. The secondary study outcomes were default, defined as absence on three consecutive visits; non-response, defined as failure to attain discharge criteria after three months on treatment; transfer to inpatient therapeutic care (ITC); or death. The criteria for ITC transfer were loss of appetite, development of medical complications, development of oedema, weight loss or static weight on three consecutive visits, and request by the caregiver. Other secondary outcomes were length of stay, defined as the number of days from treatment initiation to recovery and average weight gain, defined as weight change (g per kg per day) from treatment initiation to recovery. Baseline maternal and child’s sociodemographic data (child’s sex and age; mother’s vital status, age, education, and household wealth variables) and child’s physical assessment and health status data (MUAC, weight, exposure to HIV, type of admission, and presenting symptoms) were collected at enrolment. Child’s MUAC, weight and the amount of RUTF dispensed were recorded at each weekly visit. All collected data were recorded in case report forms contained in an enrolment and follow-up register. Children were enrolled into the study from August 2018 to December 2019 in the intervention group and from August 2018 to February 2020 in the control group. Follow-up ended on 26 March 2020. We obtained data to estimate coverage from SAM registers in health facilities in the control wards and from CHWs in the intervention wards. We also reviewed SAM registers at three health facilities (Maswa, Somanda and Songambele) offering ITC in the study districts and counted all children from the study wards who were treated in these health facilities. The main source of cost data was the accounting records of Doctors with Africa CUAMM (the implementing agency). We collected additional cost data on human resources (salaries and time allocation), capital and consumables using a questionnaire administered to health facility staff in the control areas. We estimated the minimum required sample size of 258 (129 per group) assuming that treatment of children with SAM by CHWs was non-inferior compared to treatment of children with SAM in health facilities, an overall proportion of cured children in both arms of 88% (pi = 0.88), a non-inferiority margin of 10% (delta = 0.1), a power of 80%, and a one-sided alpha of 0.025. We used the ssi module in Stata (College Station, TX, USA) to calculate the sample size. Data were entered in EpiData in duplicate, validated and exported to Stata 15 for cleaning and analysis, which was performed based on the intention-to-treat principle. Characteristics of participants were summarized using descriptive statistics and differences between intervention and control groups were compared using independent samples t-tests (for continuous variables) or chi-squared tests (for categorical variables). Six children in the control group had missing outcome data because follow-up ended before we could ascertain their outcomes, thus, we performed both complete-case analysis and analysis after multiple imputation to account for the missing data. We used multiple imputation with chained equations with 20 iterations based on all maternal and child characteristics listed in Table ​Table1.1. We calculated risk ratios (relative effects) and risk differences (absolute effects) with 95% CIs for cure and default using Poisson regression models with robust error variances17. We assessed the effect of the intervention on length of stay and weight gain using linear regression to obtain mean differences with 95% CIs. We adjusted the models for variables that showed some imbalance (P < 0.1) between control and invention groups. Estimates across imputed datasets were automatically combined using Rubin’s rules18. To evaluate non-inferiority of the intervention compared to the usual care, we compared the lower bound of the 95% CI for the effect of the intervention on cure with the pre-specified non-inferiority margin (− 10%). We did not assess the effect of the intervention on death, transfer and non-response to treatment because of a small number of observations. Characteristics of study participants at recruitment. Data are presented as n (%) for categorical variables or Mean ± SD for continuous variables. aDerived using principal components analysis of household assets, access to utilities and type of housing material. Because the results of both multiple imputation and complete-case analysis may be biased given that only the control group had children with missing outcome data, we performed sensitivity analysis (using the same approach as above) after excluding 31 children enrolled in the study during the same period as the children with missing data (i.e. after 28th December 2019). In other words, we restricted the analysis to only those children we could have potentially followed up for the maximum follow-up period of three months. The effect of the intervention on coverage, defined as the proportion of the children with SAM being reached with treatment in the intervention and control wards, was assessed using data on the number of children treated over a 12-month period from September 2018 to August 2019. We estimated coverage using an indirect method by dividing the number of children aged 6–59 months with SAM who received treatment (including ITC) by the expected number of children aged 6–59 months with SAM over the reference period (the annual SAM burden). Where AnnualSAMburden=numberofchildren6-59months∗SAMprevalence∗(1+K); K being the incidence correction factor, whose value was assumed to be 4.82 based on a meta-analysis of studies from three West African countries19 (similar data for Tanzania/East Africa are not available). We used a SAM prevalence 0.5% for Simiyu Region based on the National Nutrition Survey 201816. To estimate the effect of the intervention on coverage, we calculated both relative and absolute changes in coverage. We performed cost-effectives analysis from the provider’s perspective. The time horizon was 1 year: from September 2018 to August 2019. We calculated costs using the activity-based costing method by identifying the activities of the project, determining the cost of each activity and calculating the overall and unit costs. Cost analysis focused on treatment of children with SAM without complications at the ward level. Thus, we did not consider ITC costs. We included costs related to sensitization and mobilization, training of CHWs and their supervisors (transportation of trainers, training hall and meals, and per diems), supervision and monitoring (fuel costs and per diems), personnel costs (staff salaries and benefits, and incentives for CHWs and supervisors), consumables (RUTF purchase and transportation, photocopying and binding, drugs, bicycle maintenance and spare parts) and capital costs (weighing scales, thermometers, MUAC tapes, clinic furniture, and room rent). The quantity of RUTF dispensed was as reported in the child enrolment and follow-up register (from admission to discharge). Personnel costs were adjusted for time spent on the project. All costs were expressed in 2019 US dollars (1 TZ = 0.0004 US$). Capital items (any item that can be used for more than one year), were annualized using a 3% interest rate and corresponding useful life. The same strategy was used in estimating the cost of sensitization/mobilization and trainings. We computed the unit cost i.e. cost per child treated and cost per child cured. In addition, we calculated incremental cost-effectiveness ratio (ICER) by dividing the difference in costs incurred in the intervention and control areas by the difference in the number of children treated or cured in the intervention and control areas (i.e. C1 − C0/E1 − E0). We analysed the data using Microsoft Excel. The National Health Research Ethics Committee at the National Institute of Medical Research, Tanzania (NIMR/HQ/R.8a/Vol.IX/2532) approved the study protocol. This study complied with the ethical standards set by the National Health Research Ethics Committee on research regarding human subjects and with the Helsinki Declaration. Written informed consent was obtained from caretakers of all participating children before recruitment. This study was registered in the Pan African Clinical Trial Registry (Trial number PACTR201901856648139) on 21/12/2018. Views expressed in this study are solely those of the authors and do not necessarily represent the official position of Doctors with Africa CUAMM or Children’s Investment Fund Foundation.