Verbowski 2018.
| Study characteristics | ||
| Methods |
Study design: cRCT Study grouping: parallel group How were missing data handled? In 9 cases, the participant who had been enrolled at baseline was N/A at the time of data collection; so another adult female HH member of the appropriate age completed the 24‐hour recall. When nutrient values were N/A for protein, fat, iron, zinc, calcium, vitamin A, thiamine or riboflavin, values were imputed from USDA food composition equivalents, based on values per 100 g by weight (USDA 2016). Given the high attrition rates, instead of estimating missing data using multiple imputation, study authors employed the direct maximum likelihood method to account for the missing values at 22 months, which uses each respondent's available data to compute the likelihood function. The overall likelihood was the product of 2 factors: 1 computed for those respondents with missing data on some variables and 1 for those with complete data on all variables. Parameter estimation and SEs were derived from maximising the overall likelihood function. Randomisation ratio: 1:1:1 Recruitment method: among the eligible villages (190, which were those not already participating in other development programmes), random selection was used to identify 90 villages to participate in the study. Sample size justification and outcome used: number of clusters and HHs within each cluster was estimated based on the proportion of anaemic women and children, with 80% power and an a priori significant level of 0.025, to account for multiple comparisons. Assuming a 50% prevalence of anaemia and an interclass correlation of 0.05, a sample size of 300 for each group provided 80% power to detect a 15% absolute reduction in the prevalence of anaemia. Sampling method: with randomised villages, purposive sampling was used to identify 10 HHs per village to participate. Half of the participants (5/10 HHs within each cluster) were randomly selected to complete endline dietary assessment. Study aim or objective: to examine the effect of EHFP with or without aquaculture on dietary intake and prevalence of inadequate intake of select nutrients among women and children living in rural Cambodia, compared to controls. Study period: July 2012 to June 2014, a 22‐month period Unit of allocation or exposure: villages |
|
| Participants |
Baseline characteristics EHFP + aquaculture
EHFP
Control
Overall: NR Inclusion criteria: Within each village, 10 HHs were purposefully selected, according to specific criteria: HH home to a woman of childbearing age, considered poor based on local wealth rankings, had access to sufficient land and labour, had Ͱ 1 child aged < 5 years, and the woman was interested in participating in the FoF project. Exclusion criteria: NR Pretreatment: no significant differences between groups (P > 0.05), except for years of education and wealth quintiles were not equally distributed across groups; women on average had completed more years of schooling in the EHFP group than in the EHFP + aquaculture and control groups, and more HHs in the control group were in the bottom Wealth Index quintile as compared with HHs in the EHFP and EHFP + aquaculture groups. Therefore, these were included in the multivariable models as potential confounders. Attrition per relevant group: EHFP + aquaculture group: 7 women and 7 children LTFU; EHFP group: 4 women and 4 children LTFU; control group: 10 women and 10 children LTFU. The overall HH attrition rate was 16.2% (146), which did not differ across groups (P = 0.74), but attrition for women overall was higher (38.6%, 348), mainly due to employment‐related temporary migration. Primary outcome data were available for 179 (control), 185 (EHFP) and 188 (EHFP + aquaculture) women and 232 (control), 255 (EHFP) and 245 (EHFP + aquaculture) children. Venous blood samples were successfully obtained from 88% of the subset of 450 women at 22 months. LTFU for the venous blood draw was higher among women in control group (22.0%, 33) than in the EHFP (6.7%, 10) and EHFP + aquaculture (6.0%, 9) groups. Description of subgroups measured and reported: only a subgroup of participants measured and analysed per group: of the 10 HHs per village, only 5 per group were randomly selected to be assessed for dietary intake outcomes and analysed as such. Of these, further subgroups (43 woman–child pairs for EHFP + aquaculture group; 45 woman–child pairs for EHFP group; 46 woman–child pairs for the control group) were selected to do a repeat dietary intake assessment on a non‐consecutive day. Total number completed and analysed per relevant group: EHFP + aquaculture group: 143 women and 142 children analysed; EHFP group: 146 women and 144 children analysed; control group: 140 women and 135 children analysed. Total number enrolled per relevant group: each group had 30 villages randomly assigned to them, and from each village 10 HHs were enrolled, which provided per group: 300 HHs, 300 women (of which 150 women's venous blood samples were taken), 300 children Total number randomised per relevant group: see above |
|
| Interventions |
Intervention characteristics EHFP + aquaculture
EHFP
Control: no intervention |
|
| Outcomes | Dietary intake: prevalence of inadequacy of food intake in women/children Anthropometry: underweight (mothers/children); stunting Biochemical: Hb concentration (women/children) Morbidity: anaemia (women/children) |
|
| Identification |
Sponsorship source: International Development Research Centre (IDRC, grant number 106928) and Global Affairs Canada (GAC); HKI; University of British Columbia (UBC) Country: Cambodia Setting: villages in the rural Prey Veng Province, 1 of the poorest provinces with 27% of homes classified as poor, and located on the east bank of the Mekong river. Comments: trial registry number: NCT01593423 Authors' names: Susan Barr and Tim Green Email: susan.barr@ubc.ca; tim.green@sahmri.com Declarations of interest: yes; no conflicts of interest Study or programme name and acronym: Fish on Farms (FoF) project using the Enhanced Homestead Food Production (EHFP) programme Type of record: journal article |
|
| Notes | ||
| Risk of bias | ||
| Bias | Authors' judgement | Support for judgement |
| Random sequence generation (Selection bias) | Low risk | Quote: "Random allocation was done by the study coordinator in Cambodia using a computer generated random number sequence in Excel." |
| Allocation concealment (Selection bias) | Unclear risk | NR |
| Baseline characteristics similar (Selection bias) | Low risk | Baseline characteristics of participants per group were reported and mostly similar, except for it appeared that years of education and wealth quintiles were not equally distributed across groups; women on average had completed more years of schooling in the EHFP group than in the EHFP + aquaculture and control groups, and more HHs in the control group were in the bottom Wealth Index quintile as compared with HHs in the EHFP and EHFP + aquaculture groups. Because these were included in the multivariable models as potential confounders, we judged this domain at low risk of bias. |
| Baseline outcome measurements similar (Selection bias) | Unclear risk | The study authors reported it a limitation of their study, that baseline dietary intake data were not collected. Although most baseline characteristics were similar across groups, the years of education and bottom Wealth Index (which were included in the multivariable models as possible confounders) were not, and it is also not certain that dietary intake data were similar across groups at baseline. |
| Blinding of participants and personnel (Performance bias) | Low risk | No blinding, but it was unlikely that the performance were influenced by lack of blinding. |
| Blinding of outcome assessment (Detection bias) | High risk | No blinding done. It is possible (but unknown) whether outcome assessors behaved differently when interviewing women from different groups (e.g. prompting women from different groups differently during facilitating the 24‐hour recall). The dietary intake of women and children was self‐reported, thus there was also a possibility that a lack of blinding of participants could have influenced their recall and outcome reporting. |
| Protection against contamination (Performance bias) | Low risk | Allocation was by village and it was unlikely that the control group received the intervention, or that the group with only the EHFP also received the aquaculture. |
| Incomplete outcome data (Attrition bias) | Unclear risk | Quote: "At the end of the study, there were no missed clusters (n = 90). The overall HH attrition rate at 22 months was 16.2% (n = 146) and did not differ across groups (P = 0.74…). Attrition was higher for women only (38.6%; n = 348) than for households …" Comment: because the total attrition was high, study authors used the direct maximum likelihood method to account for the missing values at 22 months. However, no sensitivity analysis was done and we are unclear as to how this method influenced the findings. |
| Selective outcome reporting (Reporting bias) | Low risk | The trial was prospectively registered on a trial registry website (NCT01593423). All important outcomes pre‐specified in this registry entry have reported in either Verbowski 2018, Michaux 2018 or Karakochuk 2015. |
| Other bias | Low risk | Misclassification bias: low risk. Incorrect analysis: low risk as clustering was taken into account adequately during analysis. Recruitment bias: low risk because participants in relevant villages were recruited before randomisation took place. |