Doocy 2017.

Study characteristics
Methods	Study design: PCS Study grouping: N/A How were missing data handled? With exception of imputation procedures for child anthropometric data, analysis did not consider interim measures of each indicator because after exploratory analysis, it became clear that inclusion of interim data points did not change results and conclusions, and thus eliminating analysis of interim measures would facilitate the interpretation of findings. Used a multiple imputation approach for anthropometric outcomes, where missing values were replaced by values sampled from a distribution defined by the fit of a linear regression model at a given follow‐up as a function of previous outcomes, as well as of child age and sex. Missing values for maternal age was assigned the mean and education the mode of those variables, so they could be included in the analysis. Randomisation ratio: N/A Recruitment method: at enrolment, a full description of study was read to prospective participants and they had the opportunity to ask questions about participation; those who agreed to participate gave oral consent due to high levels of illiteracy. At each subsequent survey, the respondent was read an abbreviated consent statement and asked to re‐affirm their willingness to participate prior to the interview (Doocy 2018). Study enrolment occurred between August and October 2012 following identification of beneficiaries for each intervention. 1820 beneficiaries and their HHs were enrolled and followed over 3.5 years during February/March 2016; study HHs were followed for the entire period, regardless of whether they graduated or dropped out of intervention. Sample size justification and outcome used: primary outcome measure was reduction in HH food insecurity, and authors conducted calculations for varying levels of reduction, assuming 80% power and a significance level of 0.05. With a minimum sample size of 325 HHs per group (or 1625 HHs in total for the parent study), study was powered to detect a 10% or greater reduction in prevalence of food insecurity indicators within each comparison group compared to baseline levels. Sampling method: The Jenga Jamaa II parent study used a quasi‐experimental matched design in which communities planned to receive 1 intervention (vs multiple interventions) selected for participation so that the effect of individual interventions could be assessed. Authors analysed 2 of the 5 comparison groups recruited for the parent study of Jenga Jamaa II: the FFS intervention group (388 participants) and the control group (324 participants). Participating communities within each territory (Fizi and Uvira) were matched by livelihood zone (mountains, plains or lakeside) and proximity into sets of villages with each type of intervention. The final sample had 13 sets of 3 villages; within each set of villages, 1 village received agricultural interventions, 1 received PM2A, and 1 received WEGs. In each set of villages, intervention groups were formed (i.e. 1 intervention per village) and all beneficiaries in the group were enrolled in the study. In agricultural intervention villages, the entire FFS group of approximately 30 beneficiaries was enrolled in the study. Controls were selected from WEG villages, where each beneficiary was matched with a female neighbour not participating in Jenga Jamaa II interventions, and that woman's HH was enrolled as a control. Villages were assigned to an intervention based on agreement with local leaders and availability programme resources; intervention participants were identified based on programme targeting criteria and community selection processes. Once intervention groups were formed, all group members were invited to participate in the research. In agriculture villages, 1 FFS with approximately 30 participants was enrolled; 1 of 3 farmer‐to‐farmer trainees of each FFS participant was randomly selected to comprise the F2F group. In WEG villages, 1 WEG group of 25 participants was selected. Members of the control group were also selected from WEG villages because the WEG intervention had lower coverage than FFS/farmer‐to‐farmer and PM2A interventions (i.e. there was only 1 WEG group per village compared to multiple PM2A and agricultural intervention groups), which lessened the likelihood of spillover effects on control HHs. Study aim or objective: to examine the changes in agricultural production practices, HH food security and child nutritional status that are associated with participation in FFS programmes. Study derived from a subset of data from the parent study of Jenga Jamaa II which sought to address HH food insecurity and child undernutrition through 1. increasing incomes among farming HHs through FFS and farmer‐to‐farmer training interventions, 2. improving the health and nutritional status of children aged < 5 years through the PM2A, and 3. empowering women via WEGs. Study period: HHs were followed for 3.5 years, from enrolment in autumn of 2012 (baseline) to February or March 2016 (endline). Unit of allocation or exposure: villages
Participants	Baseline characteristics FFSs Age: n, mean, years: 37.9 (SD 13.4); children aged < 2 years, n, mean: 0.5 (SD 0.5); children aged 2–4 years, n, mean: 1.5 (SD 1.1) Place of residence, %: proportion in Fizi territory: 46.2; proportion in Uvira territory: 53.8 Sex: female, %: 69.4 Ethnicity and language: NR Occupation: NR Education: mean years of education: 3.4 (SD 3.6) SES: number of income sources, mean: 2.0 (SD 1.0); proportion with land ownership, %: 69.5; HH size, mean: 6.2 (SD 2.4) Social capital: NR Nutritional status: HDDS, mean: 3.4 (SD 1.4); proportion moderately and severely food insecure (HFIAS), %: 97.8 Morbidities: NR Concomitant or previous care: NR Control Age: children aged < 2 years, n, mean: 0.7 (SD 0.5); children aged 2–4 years, n, mean: 1.8 (SD 1.1) Place of residence: proportion in Fizi territory, %: 46.2; proportion in Uvira territory, %: 53.8 Sex: NR Ethnicity and language: NR Occupation: NR Education: NR SES: mean number of income sources: 1.8 (SD 0.9); proportion with land ownership, %: 68.6; HH size, mean: 6.3 (SD 2.4) Social capital: NR Nutritional status: mean HDDS: 3.4 (SD 1.5); proportion moderately and severely food insecure (HFIAS), %: 98.4 Morbidities: NR Concomitant or previous care: NR WEGs Age: mean, years: 34.0 (SD 8.8) Place of residence: proportion in Fizi territory, %: 46.2; proportion in Uvira territory, %: 53.8 Sex: female, %: 100 Ethnicity and language: NR Occupation: NR Education: mean years of education: 2.8 (SD 3.1) SES: mean number of income sources: 1.9 (SD 0.9); proportion with land ownership, %: 70.2; HH size, mean: 6.6 (SD 2.3) Social capital: NR Nutritional status: mean HDDS: 3.4 (SD 1.7); proportion moderately and severely food insecure (HFIAS), %: 99 Morbidities: NR Concomitant or previous care: NR Overall (all intervention groups; including those irrelevant to the review) Age: mean, years: 33.4 (SD 11.4) Place of residence: proportion in Fizi territory, %: 46.2; proportion in Uvira territory, %: 53.8 Sex: female, %: 86.1 Ethnicity and language: NR Occupation: NR Education: mean years of education: 3.1 (SD 3.5) SES: mean number of income sources: 1.9 (SD 0.9); proportion with land ownership, %: 69.7 Social capital: NR Nutritional status: mean HDSS: 3.4 (SD 1.5); proportion moderately and severely food insecure (HFIAS), %: 98.5 Morbidities: NR Concomitant or previous care: NR Inclusion criteria: intervention participants identified based on programme targeting criteria and community selection processes. For the control group, enrolled primary carer of children, and 100% were women. No other details. Exclusion criteria: NR Pretreatment: more women in control group (as control group selected was women only – from the village where women empowerment intervention was implemented, whereas the interventions included all FFS beneficiaries (men and women). Intervention group participants were significantly older; control group participants had higher proportion completing any formal education control; group had higher mean number of younger children (aged 0–4 years); higher proportion of farmers in intervention group; mean HH size differed significantly between the groups; smaller for FFS groups and larger for WEG groups; number of income sources (highest in FFS group and lowest in control group). Attrition per relevant group: results included only 82% of participants who were present for both baseline and endline surveys (both conducted in February/March). Village of Kibirizi, which included 1 FFS group, was not included in endline survey (and thus was excluded from the final evaluation) due to security concerns. 1820 HHs were enrolled in study and 1481 (81%) participated in the endline survey; follow‐up ranged from 74% to 90% in different intervention groups and was lowest among the control group: FFS baseline 390; endline participation rate 81%; WEG baseline 325; endline participation rate 90%; control baseline 325; endline participation rate 78%. Description of subgroups measured and reported: NR Total number completed and analysed per relevant group: for food security outcomes: intervention group: 317 FFS beneficiaries; control group: 254 non‐FFS participants. For child anthropometric outcomes: intervention group: 265 children of FFS beneficiaries; control group: 206 children of non‐FFS participants. Total number enrolled per relevant group: FFS programme: study enrolled 388 FFS beneficiaries and their HHs in the intervention group and 324 non‐FFS adults and their HHs in the control group. WEG group: 390 HHs. Total number randomised per relevant group: N/A
Interventions	Intervention characteristics FFSs Food access intervention category: increase buying power Intervention type: income generation: FFSs Description: FFS intervention provided farmers with experience‐based education on farming practices, postharvest handling, and business and natural resource management skills. Each FFS group received semimonthly training from ADRA field agents for 2 years. Each FFS group had a community demonstration plot, and group members also received starter packages of seeds and tools for use on individual farms. The FFS programmes focused on a variety of common crops in the region, including cassava, maize, rice, beans, banana and peanuts. The first year of training focused on knowledge of production systems and technologies; second year focused on adoption of techniques and technologies, and behaviour change. Content was designed to be crop‐specific and seasonally appropriate. After completing the FFS intervention, many beneficiaries transitioned to farmer business associations, which were intended to improve access to credit and marketing opportunities. Duration of intervention period: 3.5 years Frequency: every 2 weeks Number of study contacts: 8 (August/September 2012; February/March 2013; August/September 2013; February/March 2014; August/September 2014; February/March 2015; August/September 2015; February/March 2016) Providers: implemented by ADRA. Delivery: training sessions on agriculture techniques and other content by field agents, with a community demonstration plot; provision of starter packages of seeds and tools; some reports of delayed seed arrival and inefficient tools that delayed processes. Insecurity complicated both programme delivery and data collection in some communities. Possible that spillover from the intervention areas affected the control areas. Co‐interventions: some FFS participants trained 3 farmers from their community in FFS techniques (F2F programme). This was supposed to be a scalable and less resource intensive intervention to increase agricultural input. After FFS intervention, some transitioned to farmer business associations, which were intended to improve access to credit and marketing opportunities. Resource requirements: small incentive – which was most often soap and worth approximately USD 1 – for participation in each survey. Economic indicators: NR Control Food access intervention category: no intervention Intervention type: no intervention Description: no intervention Duration of intervention period: no intervention Frequency: no intervention Number of study contacts: study HHs followed for 3.5 years, from enrolment autumn 2012 (baseline) to February or March 2016 (endline), graduated or dropped out of intervention. Data collected in 8 semiannual surveys (August/September and February/March) to account for seasonal variations in food security. Both data collection periods were at beginning of local rainy seasons. Providers: no intervention Delivery: no intervention Co‐interventions: no intervention Resource requirements: participants received a small incentive – which was most often soap and worth approximately USD 1 – for participation in each survey. Economic indicators: no intervention WEGs Food access intervention category: increase buying power Intervention type: income generation: WEGs Description: WEGs met weekly and served as a delivery mechanism for a variety of interventions including literacy and numeracy, business and marketing training and income‐generation activities, primarily soap‐making, bread‐making and fish‐drying. Beneficiaries were provided with a starter kit of basic materials for their income‐generation activity and savings and credit groups were started in each WEG; many WEG participants also received goats and energy‐efficient stoves. Duration of intervention period: 3.5 years Frequency: weekly meetings; one‐off starter kit of materials for income‐generation activities. Number of study contacts: study HHs were followed for 3.5 years, from enrolment in autumn 2012 (baseline) to February or March 2016 (endline), graduated or dropped out of intervention. Data collected in 8 semiannual surveys (August/September and February/March) to account for seasonal variations in food security. Both data collection periods were at beginning of local rainy seasons. Providers: implemented by ADRA. Delivery: beneficiaries were provided with a starter kit of basic materials for their income‐generation activity and savings and credit groups were started in each WEG; many WEG participants also received goats and energy‐efficient stoves. Co‐interventions: NR Resource requirements: NR Economic indicators: NR
Outcomes	Food security: HFIAS; proportion of HHs improving in a HFIAS category from baseline to endline; proportion of HHs food secure and mildly/moderately/severely food insecure Dietary diversity: HDDS; HHs achieving target dietary diversity Anthropometry: stunting, underweight
Identification	Sponsorship source: USAID Office of Food for Peace Cooperative Agreement (AID‐FFP‐A‐11‐00006). Country: Democratic Republic of the Congo Setting: 2 farming villages in South Kivu province in the Congo Author's name: Shannon Doocy Email: doocy1@jhu.edu Declarations of interest: none declared. Study or programme name and acronym: Farmer Field Schools programme, subset of interventions implemented as part of the Jenga Jamaa II project Type of record: journal article
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (Selection bias)	High risk	PCS
Allocation concealment (Selection bias)	High risk	PCS; therefore, no allocation concealment carried out.
Baseline characteristics similar (Selection bias)	Low risk	Some imbalances in gender, age, number of younger children, proportion who were farmers between the intervention and control groups. These characteristics were adjusted for in the analysis and authors also carried out PSM.
Baseline outcome measurements similar (Selection bias)	Unclear risk	Study authors reported differences in food security outcomes at baseline, although the statistical analysis of these outcome variables was not provided. Analyses regarding food security outcomes was adjusted for baseline HDDS values. However, for child anthropometric outcomes, there were no baseline values reported, thus it is unclear whether there were imbalances or not.
Blinding of participants and personnel (Performance bias)	Low risk	No blinding of participants and personnel was done. However, in this type of intervention, it is unlikely that lack of blinding would have affected experience or treatment of participants.
Blinding of outcome assessment (Detection bias)	High risk	Although blinding was not possible, interviews were carried out by staff that did not regularly interact with the participants. However, outcomes such as dietary diversity and food insecurity, that are based on respondents recall, may be susceptible to lack of blinding.
Protection against contamination (Performance bias)	High risk	Control HHs were sampled from villages that received 1 of the intervention types (WEG), thus the risk for contamination was high. Authors also reported that, "It is also possible that spillover from the intervention areas affected the control areas."
Incomplete outcome data (Attrition bias)	High risk	Differential attrition observed between groups: 10% for WEGs, 19% for FFS and 22% for control group. Reasons for missing/excluded data not provided. 1 entire village (Kibirizi) was excluded from the analyses (due to conflicts in the country). For child anthropometric data, missing values were inputted using a multiple imputation approach, where missing values were replaced by values sampled from a distribution defined by the fit of a linear regression model at a given follow‐up as a function of previous outcomes, as well as of child age and sex.
Selective outcome reporting (Reporting bias)	Unclear risk	Study protocol N/A; however, the primary outcome, namely HH food security (as prespecified in the Methods section of the citation by Doocy 2017), was reported in the text.
Other bias	Low risk	Misclassification bias of exposure: low risk. Measurement bias: low risk; scales used were validated and widely used, as was anthropometric measurement; and staff were trained in data collection. Seasonality: low risk; study collected data over different seasons over 3.5 years. Random sequence generation (selection bias).