Jessica R. Bogard
School of Public Health, The University of Queensland, Herston, Queensland, Australia and Agriculture Flagship, Commonwealth Scientific and Industrial Research Organisation (CSIRO), St Lucia, Queensland, Australia
Shakuntala H. Thilsted
WorldFish, Batu Maung, Bayan Lepas, Penang, Malaysia
Abdulla Mamun
School of Public Health, The University of Queensland, Herston, Queensland, Australia
Kazi Toufique
Bangladesh Institute of Development Studies, Sher-e- Bangla Nagar, Dhaka, Bangladesh
Masum Ali
Helen Keller International, House 10E, Gulshan 2, Dhaka, Bangladesh
Ben Belton
Department of Agricultural, Food & Resource Economics, Michigan State University, East Lansing, Michigan, United States of America
Jillian Waid
Helen Keller International, House 10E, Gulshan 2, Dhaka, Bangladesh and Institute of Public Health, Heidelberg University, Heidelberg, Germany
Geoffrey C. Marks
School of Public Health, The University of Queensland, Herston, Queensland, Australia
Sami Farook
Bangladesh Institute of Development Studies, Sher-e- Bangla Nagar, Dhaka, Bangladesh
Fish consumption, Capture fisheries, Aquaculture, Fish, Health care
Food Safety and Security
Dietary intake, Nutrition
Nationally representative household food consumption data from the Household Expenditure Survey 1991, and HIES 2000 and 2010 in Bangladesh were used to estimate apparent fish consumption. The three independent cross-sectional survey sample designs were based on a two-stage stratified sampling technique; primary sampling units were selected with probability proportional to size in the first stage, and households were selected by systematic random or circular sampling in the second stage. General characteristics of surveyed households are described. These are consistent with broader demographic trends in Bangladesh, including increasing urbanization, decreased poverty rates, reduced fertility rates and an aging population, along with improved access to water and sanitation, and health care. Of note is the considerable reduction in the proportion of extreme poor households between 1991 (40%) and 2010 (17%). Food consumption data were collected by trained interviewers using one-day recall for the 1991 survey over a period of 30 days, and using two-day recall for the 2000 and 2010 surveys to obtain food consumption over 14 consecutive days. Data collection was structured throughout the year, thereby controlling for seasonal effects. Poverty status was defined, using the cost of basic needs method as per survey reports, with households categorised as extreme poor, moderate poor and non-poor. The food poverty line (FPL) was estimated as the cost of a basic food basket that meets the energy needs of an adult, and the non-food poverty line (NFPL) was estimated as the cost of non-food expenditure by households close to the FPL. Thresholds for each line were set for each survey year for each district and for rural and urban areas by the Bangladesh Bureau of Statistics (BBS). Extreme poor households are those with total expenditures at or below the FPL, and the moderate poor households are those with total expenditures at or below the NFPL and above the FPL. Fish species recorded in the survey were grouped according to their dominant production sector; either capture fisheries (non-farmed) or aquaculture (farmed), for each survey year, allowing comparison of the relative contribution that each sector makes to fish consumption over time. Results are presented per Adult Male Equivalent (AME). AME reflects the energy requirements of individual households members, based on age and sex, as a proportion of an adult male, providing a more accurate estimate of the adequacy of household food consumption compared to per capita intake (see Detailed methods). Households with unrealistic levels of fish consumption in the local context (>500 g/AME/day, for comparison, mean fish consumption was 54 g/AME/day) were excluded (n = 15 in 1991, from a total of 5,745 households). Data on the quantity of each fish species consumed were then combined with species level nutrient composition data to estimate apparent nutrient intakes from fish, at each time point. Fish species consumption is recorded in the surveys according to common Bangla names, which may represent several distinct species. In these cases, the average nutrient composition of several applicable species was used. For a small number of fish species recorded in the surveys, data were not available for vitamin A content (4–6 species across the three surveys, or vitamin B12 content (7–13 species across the three surveys). A small number of households consuming only fish with these missing data were therefore excluded from analysis of those specific nutrients (vitamin A, n = 171; vitamin B12, n = 465) to minimise impacting the results. The proportion of households consuming some quantity of fish, eggs, poultry, meat or dairy within the survey period (compared to total households), is also reported. This data is used to reflect consumption patterns of other ASFs relative to fish, over time. All statistical analyses were conducted using STATA (version 12.1, StataCorp, College Station, TX, USA). Regression analyses were used to estimate mean fish consumption and mean nutrient intakes from fish, at each time point (P<0.01, using sample weights provided by BBS and adjusting for clustering of primary sampling units in survey design). All primary outcome variables (fish consumption and nutrient intakes from fish, per AME/day) were positively skewed in distribution and log transformation did not produce a Normal distribution. Non-parametric tests and equality of means were not appropriate, given the need to apply sample weights. However, a sensitivity analysis was conducted, using quantile regression which is suitable for non-parametric analyses and is also not sensitive to the presence of outliers. This analysis revealed similar trends and statistical significance; any deviations to main results are explained in footnotes to results. Unfortunately, quantile regression in STATA is not able to adjust for both clustering in survey design and survey weights simultaneously, and so, the sensitivity analyses presented here is based on quantile regression, adjusting for survey weights only. Overall strengths and limitations of the analysis are detailed.
https://www.researchgate.net/publication/315880907 Article in PLoS ONE ยท May 2017
Journal