Escherichia coli contamination of child complementary foods and association with domestic hygiene in rural Bangladesh

Authors


Abstract

Objective

To determine the frequency and concentration of Escherichia coli in child complementary food and its association with domestic hygiene practices in rural Bangladesh.

Method

A total of 608 households with children <2 years were enrolled. We collected stored complementary food samples, performed spot checks on domestic hygiene and measured ambient temperature in the food storage area. Food samples were analysed using the IDEXX most probable number (MPN) method with Colilert-18 media to enumerate E. coli. We calculated adjusted prevalence ratios (APR) to assess the relationship between E. coli and domestic hygiene practices using modified Poisson regression, adjusting for clustering and confounders.

Result

Fifty-eight percentage of stored complementary food was contaminated with E. coli, and high levels of contamination (≥100 MPN/dry g food) were found in 12% of samples. High levels of food contamination were more prevalent in compounds where the food was stored uncovered (APR: 2.0, 95% CI: 1.2–3.2), transferred from the storage pot to the serving dish using hands (APR: 2.0, 95% CI: 1.3–3.2) or stored for >4 h (APR: 2.5, 95% CI: 1.5, 4.2), in compounds where water was unavailable in the food preparation area (APR: 2.6, 95% CI: 1.6, 4.2), where ≥1 fly was captured in the food preparation area (APR: 1.6, 95% CI: 1.0, 2.6), or where the ambient temperature was high (>25–40 °C) in the food storage area (APR: 2.7, 95% CI: 1.5, 4.4).

Conclusion

Interventions to keep stored food covered and ensure water availability in the food preparation area would be expected to reduce faecal contamination of complementary foods.

Abstract

Objectif

Déterminer la fréquence et la concentration d’E. coli dans les aliments complémentaires pour enfants et leur association avec les pratiques d'hygiène domestique en zone rurale au Bangladesh.

Méthode

608 ménages avec des enfants < 2 ans ont été inscrits. Nous avons recueilli des échantillons d'aliments complémentaires stockés, avons effectué des contrôles sur place sur l'hygiène domestique et avons mesuré la température ambiante dans la zone de stockage des aliments. Les échantillons d'aliments ont été analysés en utilisant la méthode du nombre le plus probable d’IDEXX (MPN) et le milieu Colilert-18 pour énumérer E. coli. Nous avons calculé des rapports de prévalence ajustés (APR) pour évaluer l'association entre E. coli et les pratiques d'hygiène domestique en utilisant la régression de Poisson modifiée, en ajustant pour le regroupement et les facteurs de confusion.

Résultat

58% des aliments complémentaires étaient contaminés par E. coli et des niveaux élevés de contamination (≥100 MPN/g d'aliment sec) ont été trouvés dans 12% des échantillons. Les concentrations élevées de contaminants alimentaires étaient plus répandues dans les ménages où les aliments étaient stockés à découvert (APR: 2,0; IC95%: 1,2-3,2), transférés du pot de stockage vers le plat de service à la main (APR: 2,0; IC95%: 1,3-3,2) ou stockés pendant plus de 4 heures (APR: 2,5; IC95%: 1,5-4,2), dans les ménages où l'eau n’était pas disponible dans la zone de préparation des aliments (APR: 2,6; IC95%: 1,6, 4,2), là où ≥1 mouche ont été capturées dans la zone de préparation des aliments (APR: 1,6, IC95%: 1,0-2,6) ou là où la température ambiante était élevée (> 25 à 40 °C) dans la zone de stockage des aliments (APR: 2,7; IC95%: 1,5, 4,4).

Conclusion

Des interventions visant à maintenir couverts les aliments stockés et à assurer la disponibilité de l'eau dans la zone de préparation des aliments pourraient réduire la contamination fécale des aliments complémentaires.

Abstract

Objetivo

Determinar la frecuencia y la concentración de E. coli en alimentos infantiles complementarios y su asociación con las prácticas de higiene domésticas en zonas rurales de Bangladesh.

Método

Se incluyeron 608 hogares con niños <2 años. Recogimos muestras de alimentos complementarios, realizamos comprobaciones en el acto de higiene doméstica y medimos la temperatura ambiente en las áreas de almacenamiento de la comida. Las muestras de comida se analizaron utilizando el método IDEXX de número más probable (NMP) con medio Colilert-18 para enumerar E. coli. Calculamos las tasas de prevalencia ajustadas (TPA) para evaluar la relación entre E. coli y las prácticas de higiene doméstica mediante una regresión modificada de Poisson, ajustando para conglomerados y confusores.

Resultados

Un 58% de los alimentos complementarios almacenados estaban contaminados con E. coli y se encontraron unos altos niveles de contaminación (≥100 NMP/ g alimento seco) en un 12% de las muestras. Unos altos niveles de contaminación de los alimentos eran más prevalentes en lugares en los que la comida se guardaba sin tapar (TPA: 2.0, IC 95%:1.2-3.2), eran transferidos del contenedor donde estaban almacenados al plato utilizando las manos (TPA: 2.0, IC 95%:1.3-3.2) o guardados por >4 horas (TPA: 2.5, IC 95%:1.5, 4.2), en lugares en donde el agua no estaba disponible en el área de preparación de los alimentos (TPA:2.6, IC 95%:1.6, 4.2), donde ≥1 mosca fue capturada en el área de preparación de la comida (TPA: 1.6, IC 95%:1.0, 2.6), o donde la temperatura ambiente era alta (>25 a 40°C) en el área de almacenamiento de la comida (TPA:2.7, IC 95%:1.5, 4.4).

Conclusión

Se espera que las intervenciones para mantener la comida almacenada tapada y asegurar la disponibilidad de agua en las áreas de preparación de los alimentos reduzcan la contaminación fecal de los alimentos complementarios.

Introduction

Microbiologically contaminated food is particularly harmful for infants and children <2 years of age, who have immature immune systems and are vulnerable to infections with enteric pathogens [1]. Enteric pathogens can cause diarrhoea and contribute to malnutrition [2-8]. In low-income countries, substantial and irreversible growth faltering occurs in the first 2 years of children's lives [1], coinciding with the introduction of complementary foods and peak diarrhoea incidence [9, 10].

Escherichia coli, an indicator of faecal contamination, has been detected in various food items in Bangladesh [11-15], including complementary foods [16]. In a recent study in Bangladesh, 40% of complementary foods for 6- to 24-month-old children were contaminated with E. coli, and children exposed to these foods had more episodes of reported diarrhoea in the preceding 2 weeks [8].

Recently, the Hazard Analysis of Critical Control Point (HACCP) approach, conventionally used in industrial food manufacturing settings, was applied in the domestic setting to identify steps in the food preparation process where contamination likely occurs and can feasibly be controlled. In India, HACCP analysis identified high initial contamination of raw foods, poor environmental and personal hygiene (particularly unwashed hands before food preparation and child feeding), feeding wet foods stored overnight at ambient temperature and insufficient reheating of stored wet foods as risk factors for complementary food contamination [17]. Two recent studies among mothers from low-income settings closely observed and documented the food preparation steps and applied HACCP analysis to recommend handwashing with soap, washing utensils with treated water, cooking food until boiling, covering food with a lid during storage and reheating food before subsequent feeding of leftovers to children [5, 18]. These recommendations successfully reduced food contamination levels in the short run among a closely monitored group of respondents; however, it is not clear whether these recommendations are scalable or sustainable in the long run.

The aforementioned studies that employed the HACCP approach to identify all points at which contamination can occur and be controlled. However, HACCP does not provide guidance on the relative degree to which the steps impact contamination, which is important for prioritising feasible recommendations for behavioural interventions that account for the environmental and time constraints faced by a busy rural mother [19, 20]. While few studies conducted in low-income setting have identified that poor food hygiene by mothers can be associated with increased diarrhoea among their children [21, 22], no recent study has identified environmental and behavioural factors associated with food contamination in the context of low-income settings.

This study examined stored complementary food from rural Bangladeshi households to (i) assess the frequency and concentration of E. coli contamination and (ii) identify environmental and behavioural factors associated with E. coli contamination. This study adds to these findings by identifying further environmental and behavioural factors that contribute to food contamination. Additionally, we calculated attributable risks for each factor to gauge those that should be prioritised in future interventions.

Methods

Study setting and population

This study was conducted in rural villages in four districts (Gazipur, Kishorgonj, Mymensingh and Tangail) in central Bangladesh. The study period encompassed both rainy (July to October 2013) and dry season (November 2013 to March 2014). The study population comprised 720 households with young children aged 6–24 months, participating in the control arm of WASH Benefits, a large-scale randomised controlled trial of water, sanitation, hygiene and nutrition interventions [23].

Data collection and food sample processing

Field workers visited household in enrolled compounds from July 2013 to March 2014. When members of eligible households were absent during the first enrolment attempt, the field team revisited them two additional times before excluding them. Immediately after taking the consents, field workers hung three horizontally 1.5-foot long strips of sticky fly tape in the food preparation area, away from smoke or rain, to measure fly density. Field workers administered structured questionnaires and performed spot checks on food and hand hygiene practices. The structured questionnaires elicited socio-demographic information, household assets, food preparation time and food storage duration. Behaviours and practices directly observed by field workers as part of spot checks included cooking and food storage container types and cover status, presence of flies and animals in the food storage area, and presence of faeces in the household compound and food storage area. They assessed water availability in the food preparation area, defined as presence of water within 10 steps from the food preparation area. They also inspected both hands of the respondent and the child to assess the cleanliness of fingernails, palms and fingers. Approximately after 3–4 h, when the field workers had completed the surveys, they revisited the households, counted and identified species of the captured flies. Field workers also measured the ambient temperature in the food storage area using an electronic thermometer (AcuRite 00325; Chaney Instrument Co., Lake Geneva).

During their visits, field workers, in discussion with child caregivers, identified stored food intended to be served to children <2 years of age and asked the caregiver to obtain the food from the storage pot to place into a child's feeding bowl in their usual manner. This demonstration allowed field workers to observe the usual food serving practices. They also observed the cleanliness of the plates and utensils used to serve the food. After completion of the survey including structured questionnaire and spot checks, field staff collected an aliquot of the food; when multiple foods were identified for consumption by the child, field workers prioritised collecting rice or rice porridge. If rice was not available, they collected the food type identified by the caregiver as the one most commonly fed to the child. Field workers wore clean gloves and filled a sterile plastic container with 50 ml of food from the child's feeding bowl using a sterile spoon.

Samples were transported to the laboratory on ice and processed on the same day, typically within 8 h of collection. Aliquots of 10 g wet food were placed into a sturdy blending bag with a filter (BagFilter P, 400 ml, Interscience, France) and homogenised with 100 ml of sterile distilled water in a laboratory-quality food processer (BagMixer C, Interscience, France) for 1 min at a specified mixing speed (speed 4). 10 ml of the homogenised mixture was further diluted with 90 ml of distilled water. Samples were analysed using the IDEXX most probable number (MPN) method with Colilert-18 media to enumerate E. coli. An additional aliquot (approximately 5 g) of the original wet food sample was placed in a drying oven overnight to determine the moisture content. The E. coli concentration was expressed as MPN/dry g food.

Data analysis

Our outcome measures were the percentage of samples with high E. coli contamination (≥100 MPN/dry g food) and the geometric mean E. coli concentration. In Bangladesh, there is no specific standard threshold for E. coli in complementary foods. Several countries have microbiological food quality guidelines that specify limits for ready-to-eat food. Levels exceeding 100 E. coli/dry g are considered unacceptable and indicate that pathogens have likely been introduced [24, 25]. In addition to the 100 MPN threshold, we also considered different cut points of E. coli contamination, such as presence/absence, ≥10 MPN/dry g and ≥50 MPN/dry g. The E. coli count data were right-skewed; therefore, we log-transformed the counts to calculate the geometric mean (GM) with 95% confidence intervals (CI). The value of 0.5 MPN/dry g was used to log-transform counts for samples below the detection limit of 1 MPN/dry g.

We calculated prevalence ratios (PR) to assess the relationship between high E. coli contamination (≥100 MPN/dry g food) and domestic hygiene and environmental conditions using generalised estimating equations (GEE) with modified Poisson distribution to account for clustering; the unit of clustering for this analysis was a group of eight enrolled households located within walking distance of each other that the WASH Benefits trial used as the unit of intervention assignment. We conducted unadjusted as well as multivariate analyses to control for potential confounders. We determined a set of a priori confounders including mother's education, household wealth (using a household wealth score developed from observations of household assets using principal component analysis), number of children <5 years in the household and number of households in the compound. The adjusted models included multiple variables that can affect food contamination; food storage time, food reheating history, food container covering status, presence of food waste in the courtyard, presence of domestic animals, the degree to which animals were free to roam and the presence of animal faeces observed in the food storage area were included in multivariate analyses. For each specific food hygiene exposure, we included all covariates in the model whose exclusion changed the coefficient for the exposure of interest at >5% level. We assessed effect modification by season by including an interaction term for dry vs. rainy season in the regression models. We estimated the geometric mean ratio (GM Ratio) for log-transformed E. coli counts using linear regression and used robust sandwich estimator for clustered households. Finally, we calculated the attributable risk (AR), using the formula described previously [26] using the adjusted prevalence ratio (APR) from the multivariate regression analysis for each of the food hygiene exposures that were significantly associated with increased risk of high E. coli contamination at the P < 0.05 level.

Ethical considerations

Respondents in the participating households (children's caregivers) provided written, informed consent before interviews and sample collection. The study protocol was reviewed and approved by the Ethical Review Committee of the icddr,b (previously known as International Centre for Diarrhoeal Disease Research, Bangladesh) and by the Institutional Review Boards of University of California, Berkeley and Stanford University.

Results

Among the initial 720 households approached for the WASH Benefits study, 112 (15%) were ineligible for this study due to lack of a child in the eligible age range; thus, a total of 608 households were enrolled. The field team collected domestic hygiene data in 608 households and food from 572 (36 households had no food available during the data collection visit).

The average number of family members per household was five. The mean age of mothers was 24 years and approximately 15% lacked any formal education. The primary source of drinking water was a tube well (97%); 19% of compounds had their own improved latrine. More than half (57%) had electricity while only 9% had a refrigerator (Table 1).

Table 1. Socio-demographic characteristics of participants and households, rural Bangladesh, 2013
CharacteristicsN = 608
  1. SD, standard deviation.

  2. a

    Households in rural Bangladesh are clustered in multiple-family compounds (baris) shared by members of extended families.

  3. b

    Functional water seal, presence of slab, pour flash that drain into piped water system, septic tank and flush to pit latrine. WHO/UNICEF Joint Monitoring Programme (JMP) used for definition for latrine characterisation.

  4. c

    Items used to construct the wealth index.

General
Households/compounda mean (SD)2.5 (1.5)
People/compound, mean (SD)11.5 (6.4)
People/household, mean (SD)4.7 (2.2)
<5-year-old children/compound, mean (SD)1.9 (1.0)
<5-year-old children/household, mean (SD)1.3 (0.5)
Mother's age, mean (SD)23.8 (5.0)
Mother lacked formal education15%
Latrine facility
Improved ownedb19%
Improved shared7%
Unimproved74%
Drinking water source
Tube well97%
Piped water3%
Electrical connectionc57%
Proportion who owned
Housec99%
Wardrobec19%
Bicyclec29%
Mobile phonec86%
Black and white televisionc7%
Colour televisionc25%
Sewing machinec6%
Refrigeratorc9%
Motor cyclec8%
Radioc5%
Number of items owned
Tables, mean (SD)c1.20 (1.0)
Chairs, mean (SD)c2.60 (2.5)
Beds, mean (SD)c1.10 (1.2)
Inexpensive sleeping cots, mean (SD)c1.20 (0.9)
Acres of homestead land, mean (SD)0.14 (0.2)
Acres of non-homestead land, mean (SD)0.97 (1.2)
House construction
Tin roof98%
Cement floor12%
Brick walls13%
Number of rooms, mean (SD)2.0 (1.2)
Cooking fuel
Wood21%
Crop residue/grass74%
Dung cakes5%
Biogas1%

Seventy-seven percentage of stored food samples collected were rice; 20% were porridge, and 3% were khichuri, a mixture of rice, lentils and vegetables (Table 2). More than 90% of households had water available near the food preparation area (<10 steps). About 85% of the food storage containers were fully covered. Mothers served complementary food with bare hands in 37% of instances, and approximately half used utensils. A quarter of serving plates were unclean. The average reported food storage duration was 4.4 h. Nearly all (98%) respondents used wide-mouth containers for food storage and 85% of the containers were fully covered. During spot checks, field workers observed flies in 11% of household food storage areas and animals in 94% of compounds.

Table 2. Association of complementary food contamination* (≥100 MPN E. coli/dry g food) with domestic hygiene in rural Bangladesh, 2013
IndicatorsN (%)High Escherichia coli contamination (≥100 MPN/dry g), n/N (%)Unadjusted prevalence ratio (PR) (95% CI)P-value**Adjusted prevalence ratio (APR) (95% CI)P-value
  1. *From 572 households providing food samples; **APR: Adjusted prevalence ratio, adjusted for mother's education, wealth index, number of households in compounds, number of people living in compounds, number of children <5 years of age in the compound, food storage time, food reheat history, food container covering status, food waste in the courtyard, animal presence, animal roaming and animal faeces observed in the food storage area.

Observed water availability in food preparation area
Present504 (93)49/457 (11)1.0 1.0 
Absent40 (7)10/36 (28)2.6 (1.6, 4.1)<0.0012.6 (1.6, 4.2)<0.001
Observed food container covering status
Fully covered487 (85)48/466 (10)1.0 1.0 
Uncovered85 (15)18/81 (22)2.0 (1.3, 3.3)0.0032.0 (1.2, 3.2)<0.001
Observed hand contact while food serving
No357 (63)29/340 (9)1.0 1.0 
Yes211 (37)33/203 (16)2.0 (1.3, 3.1)0.0012.0 (1.3, 3.2)<0.001
Reported food storage time
≤4.0 h396 (69)29/379 (8)1.0 1.0 
>4.0 h176 (31)37/168 (22)2.4 (1.4, 4.2)0.0012.5 (1.5, 4.2)<0.001
Temperature of food storage area
14–25 °C210 (37)10/198 (5)1.0 1.0 
26–40 °C360 (63)55/347 (16)3.1 (1.8, 5.4)<0.0012.7 (1.5, 4.7)<0.001
Fly captured in food preparation area (at least 1)
No414 (68)35/365 (10)1.0 1.0 
Yes194 (32)31/184 (17)1.5 (0.90, 2.6)0.091.60 (1.0, 2.6)0.05
Observed food container storage location
Elevated surface365 (64)41/347 (12)1.0 1.0 
On the ground160 (28)18/154 (12)1.0 (0.60, 1.8)0.901.0 (0.60, 1.8)0.90
Observed animal faeces in the food storage area
No557 (98)64/532 (12)1.0 1.0 
Yes14 (2)1/14 (7)0.7 (0.2, 2.6)0.600.7 (0.2, 2.6)0.60
Flies observed in food storage area
No510 (89)57/487 (11)1.0 1.0 
Yes62 (11)9/60 (15)1.2 (0.6, 2.2)0.601.2 (0.6, 2.2)0.60
Observed food waste in the courtyard
No396 (65)45/361 (12)1.0 1.0 
Yes212 (35)21/188 (11)0.8 (0.5, 1.4)0.401.2 (0.7, 2.1)0.40
Observed animal roaming in the compound
No229 (40)30/206 (15)1.0 1.0 
Yes342 (60)36/310 (12)0.9 (0.6, 1.3)0.400.8 (0.6, 1.3)0.50
Observed mother's hands clean
Yes504 (83)51/452 (28)1.0 1.0 
No104 (17)15/97 (15)1.3 (0.7, 2.3)0.401.3 (0.7, 2.3)0.35
Reported food reheated in last 4 h
No553 (97)63/529 (12)1.0 1.0 
Yes19 (3)3/18 (17)1.1 (0.4, 2.9)0.801.1 (0.40, 2.7)0.80
Food storage pot hot to touch
No371 (65)53/357 (15)1.0 1.0 
Yes201 (35)13/190 (7)0.5 (0.3, 0.9)0.010.7 (0.40, 1.20)0.20
Observed utensils appearance
Unclean145 (47)10/138 (7)1.0 1.0 
Clean161 (53)12/152 (8)1.0 (0.5, 2.1)0.950.9 (0.40, 2.1)0.90
Observed plates appearance
Unclean157 (28)20/149 (13)1.0   
Clean401 (72)41/384 (11)0.7 (0.4, 1.2)0.250.8 (0.45, 1.3)0.30

We detected E. coli in 58% of stored complementary food samples; the geometric mean among all samples was 1.4 MPN/dry g, and 12% of samples were highly contaminated (≥100 MPN/dry g). In multivariable analyses, high E. coli contamination was more prevalent in food stored in an uncovered container (APR: 2.0, 95% CI: 1.2–3.2), transferred from the storage pot to the serving dish using hands (APR: 2.0, 95% CI: 1.3–3.2) or stored for >4 h (APR: 2.5, 95% CI: 1.5, 4.2), in households where water was not present in the food preparation area (APR: 2.6, 95% CI: 1.6, 4.2), at least one fly was captured in the food preparation area (APR: 1.6, 95% CI: 1.0, 2.6), or where there were higher ambient temperatures (25–40 °C) in the food storage area (APR: 2.7, 95% CI: 1.5, 4.4) (Table 3). The prevalence of high E. coli contamination appeared to be greater among mothers with unclean hands and lower when the food storage pot was hot or if households used clean plates and utensils for serving food, but the associations were not statistically significant (Table 2). The proportion of food samples with high contamination was greater in the rainy season. However, the factors significantly associated with high E. coli contamination did not differ between dry and rainy season (Table S1).

Table 3. Association of complementary food contamination* with Escherichia coli count per dry gram with domestic hygiene in rural Bangladesh, 2013
IndicatorsGeometric mean (GM)Unadjusted GM ratioP-valueAdjusted GM ratio**P-value
  1. *From 572 households providing food samples; **GM Ratio adjusted for mother's education, wealth index, HHs in compound, number of people live in compound, number of children <5 years in the compound, food storage time, food reheat history, food container covering status, food waste in the courtyard, animal presence, animal roaming and animal faeces observed in the food storage area.

Observed water availability in food preparation area
Present1.30    
Absent2.001.6 (1.0, 2.6)0.051.6 (0.9, 2.5)0.07
Observed food container covering status:
Fully covered1.24    
Uncovered1.921.6 (1.4, 2.7)0.011.6 (1.0, 2.2)0.01
Observed hand contact while food serving
No1.25    
Yes1.431.2 (0.9, 1.4)0.171.2 (0.9, 1.4)0.20
Reported food storage time
≤4.0 h1.10    
>4.0 h2.101.9 (1.5, 2.5)0.002.0 (1.5, 2.5)<0.001
Temperature of food storage area
14–25 °C1.00    
26–40 °C1.551.6 (1.2, 2.0)0.0011.5 (1.2, 1.9)0.002
Fly captured in food preparation area (at least 1)
No1.20    
Yes1.651.4 (1.0, 1.8)0.021.4 (1.0, 1.8)0.03
Observed animal presence in the compound
No0.75    
Yes1.381.8 (1.4, 2.5)0.001.9 (1.4, 2.7)<0.001
Observed food containers storage location
Elevated surface1.30    
On the ground1.341.0 (0.8, 1.3)0.701.0 (0.9, 1.3)0.65
Observed animal faeces in the food storage area
No1.32    
Yes1.301.0 (0.5, 1.8)0.950.9 (0.5, 1.8)0.90
Flies observed in the food storage area
No1.30    
Yes1.621.3 (0.9, 1.8)0.181.2 (0.90, 1.8)0.19
Observed food waste in the courtyard
No1.33    
Yes1.301.0 (0.80, 1.2)0.911.0 (0.8, 1.2)0.85
Observed animal roaming in the compound
No1.39    
Yes1.371.0 (0.8, 1.2)0.900.9 (0.80, 1.2)0.90
Observed mother's hand clean
Yes1.31    
No1.421.1 (0.8, 1.5)0.501.1 (0.9, 1.5)0.45
Reported food reheated in last 4 h
No1.30    
Yes2.281.7 (0.9, 3.4)0.101.8 (0.9, 3.3)0.09
Food storage pot hot to touch
No1.50    
Yes1.000.7 (0.6, 0.9)0.0020.9 (0.7, 1.1)0.38
Observed utensils appearance
Unclean1.20    
Clean1.331.1 (0.9, 1.5)0.371.1 (0.8, 1.5)0.40
Observed plates appearance
Unclean1.38    
Clean1.250.9 (0.7, 1.1)0.440.9 (0.70, 1.2)0.50

Among significantly associated risk factors, storing food uncovered, temperature >25 °C in the food storage area and storage time >4 h remained associated when using various cut points of E. coli contamination (i.e. presence/absence, ≥10 MPN/dry g, and ≥50 MPN/dry g). Hand contact while serving food was associated with ≥50 MPN/dry g. Water unavailability in the food preparation areas was associated with all E. coli presence levels examined but not with presence/absence. Capturing at least one fly in the food preparation areas was associated with high E. coli contamination (≥100 MPN/dry g) and E. coli presence/absence but not with the ≥10 MPN/dry g or ≥50 MPN/dry g cut points (Tables S2S4).

In multivariate models of log E. coli concentration, higher concentrations were significantly associated with uncovered food containers (GM Ratio: 1.6, 95% CI: 1.0, 2.2), storing food for >4 h (GM Ratio: 2.0, 95% CI: 1.5, 2.5), temperatures >25 °C in the food storage area (GM Ratio: 1.5, 95% CI: 1.2, 1.9) and at least one fly captured in the food preparation area (GM Ratio: 1.4, 95% CI: 1.0, 1.8). The presence of animals in the compound was associated with E. coli concentration (GM Ratio: 1.9; 95% CI: 1.4, 2.7) but not with any of the binary outcome measures. Conversely, hand contact while serving food and water presence in the food preparation area were significantly associated with the binary outcome, but not with the continuous outcome (Table 3).

Among factors significantly associated with high E. coli contamination, the attributable risk was 17% for unavailability of water in the food preparation area, 13% for storage time >4 h, 11% for food remaining uncovered, 10% for temperatures >25 °C in the food storage area, 10% for at least one fly captured in food preparation area and 8% for hand contact while serving food (Table 4).

Table 4. Attributable risk using identified risk factors from multivariate analysis for high level Escherichia coli contamination (≥100 MPN E. coli/dry g food) in rural Bangladesh, 2013
IndicatorsAPR95% CIP valueAttributable
Risk
Absence of water in food preparation area2.61.6, 4.2<0.0010.17
Food stored >4 h2.51.5, 4.2<0.0010.13
Uncovered food container2.01.2, 3.2<0.0010.11
Higher temperature in food storage area (26–40 °C)2.71.5, 4.7<0.0010.10
At least one fly captured in food preparation area1.61.0, 2.60.050.10
Hand contact while serving2.01.3, 3.2<0.0010.08

Discussion

We found high E. coli contamination (≥100 MPN/dry g) in 12% of complementary food samples in rural Bangladeshi households, which is similar to studies from other low-income settings [27, 28]. Lack of water near the food preparation area, longer storage duration, storing food uncovered, temperatures >25 °C in the food storage area, flies captured in food preparation area and hand contact with food while serving were all factors that significantly contributed to high levels (≥100 MPN/dry g) of E. coli contamination throughout the year, independent of season. The presence of animals in the compound was associated with an increase in E. coli counts. These findings provide guidance for designing targeted food hygiene interventions.

In the quantitative analysis, food contamination was most commonly attributed to lack of water in the food preparation area. It is plausible that the presence of water reflects more frequent hand washing and overall hygienic food preparation. We did not specifically collect data on whether participants washed their hands before food preparation, but food contamination in the current study was associated with mothers serving food with hands, suggesting that hand hygiene contributes to the microbiological safety of weaning food [29, 30]. Multiple studies from low-income countries have reported that mothers’ hands are heavily contaminated with faecal organisms [27, 31, 32]. In Bangladesh, mothers are primarily responsible for food-related activities such as food preparation, cooking, serving and child feeding. Recommending that mothers use utensils might not be feasible as it is the cultural norm in Bangladesh to serve and eat food with hands. However, there is ample evidence suggesting that caregivers washing hands with soap before food preparation can prevent diarrhoea among their children [33-35]. Conveniently locating soap and water near the food preparation area can promote handwashing before and during food preparation and reduce contamination [36]. Handwashing station prototypes that have been tested for acceptability in the community can serve as enabling hardware that increases convenience to facilitate handwashing [37].

In the current study, as well as several epidemiologic studies in other low-income countries, food storage at ambient temperature has been associated with food contamination [[3, 5, 8-10, 38, 39]]. However, temperature control using refrigerators is not feasible for many households in low-income countries due to financial constraints and lack of reliable power supply. It is not feasible to promote shorter storage in Bangladesh because lighting stoves several times during the day is costly and inconvenient; therefore, caregivers prepare complementary food in the morning and feed it to their children on a number of occasions until it is finished [8].

While reducing food storage temperature and duration may not be feasible, storing food in covered containers is. Reduced risk for food contamination was attributable to this simple food storage practice in our study. Covering food is also a strong recommendation from HACCP analyses of domestic food preparation in both rural Bangladesh and Mali [5, 18]. Uncovered food is vulnerable to contamination from flies and animals; we found food contamination attributable to the presence of flies in the food preparation area. The evidence linking flies with food contamination has been found elsewhere in both high- and low-income countries [10, 40-42], and several pathogens and faecal indicator bacteria have been detected on flies including E. coli [43, 44]; flies can contaminate food when they land and regurgitate or deposit excreta [45].

Animal presence was associated with food contamination when E. coli contamination was measured as a continuous outcome, a finding consistent with studies conducted in low-income settings elsewhere [9, 46, 47]. Using microbial source tracking, ruminants have been attributed to child hand and household environmental contamination in Bangladesh [48].

While food covering by itself can decrease animals and flies from coming into contact with stored foods, a qualitative study on complementary food safety in rural Bangladesh found that mothers are reluctant to cover foods immediately after cooking because of concerns that trapped steam can make foods spoil quickly [49]. To protect food from contamination, the study's authors recommended evaluating enabling hardware such as storage cabinets and containers with tight-fitting lids with vents that allow steam release [49]. These options are feasible because they can be incorporated into a busy mother's routine.

One limitation of the current study is that we measured food contamination using the indicator organism E. coli, which does not necessarily reflect levels of enteric pathogens [50]. However, E. coli is a reliable and commonly used indicator of faecal contamination in water [51, 52] and its presence in drinking water is associated with diarrhoea [52]. In this study, we primarily sampled rice, the contamination of which may not be representative of the range of foods consumed by children due to the processes involved in the preparation of rice or the characteristics of cooked rice. However, the majority of children's diet in rural Bangladesh consists of rice [53].

Another limitation is that we tested food samples for E.coli at single time point that may have been before the demonstrated food handling event. Future studies that link food contamination levels with specific observed and reported food handling events at different intervals could provide additional insight on the most dangerous contamination opportunities and the growth of bacteria over time. Handwashing can interrupt pathogen transmission and is a cornerstone of food safety. It is however quite difficult to measure both reported and observed indicators of handwashing are at high risk for bias [54, 55]. Within this study we used a proxy measure, whether the child's and caregivers’ fingerpads, palms and fingernails had visible dirt during spot check observations [56]. The absence of an association with this handwashing indicator may be due to misclassification of handwashing behaviour, although as handwashing with soap around food handling is rare in rural Bangladesh [57], we had limited power to detect associations.

A further limitation is that food storage time was estimated from maternal recall, an imperfect measure of exposure. We interviewed the mother within 12 h of cooking to enhance her ability to remember what she fed her child to minimise inaccuracies in recalled storage time. In the absence of an intervention, we would expect no bias from this imperfect recall. Rather, we would expect a non-differential effect that would weaken the association between elapsed time and contamination [58]. As we found a similar relationship as has been observed in the literature between storage time and contamination [8], we consider the relationship likely causal, although the cut point of 4 h that we used in our analysis is arbitrary. Finally, although 37% of mothers had hand contact with food when asked to demonstrate how they served food for their child; the demonstrated methods may differ from usual practice. For example, some respondents may not have served food using their hands assuming that this would be considered unhygienic (courtesy bias), potentially underestimating the practice.

A strength of this study is that most of the food hygiene indicators were directly observed. Direct observation is more reliable than self-reported practice [59-61], as socially desirable practices are commonly over-reported [55, 62].

We identified key risk factors contributing to food contamination, based on which we recommend enabling hardware. These include hardware that can be easily incorporated into caregivers’ daily routines without adding burdensome activities. They include handwashing stations in the food preparation area, storage cabinets and lidded containers that allow for steam venting. We recommend community trials to measure the impact of these enabling hardware components, in conjunction with food safety messages that resonate with the community that can be integrated into water, sanitation, and hygiene and/or nutrition programs.

Acknowledgements

We are indebted to our study participants for their time and invaluable information. We would also like to thank our field teams, field supervisors and laboratory team. This study was funded by the World Bank; the WASH Benefits study, in which it was embedded, was funded by Bill & Melinda Gates Foundation. icddr,b acknowledges with gratitude the commitment of the World Bank to its research efforts. icddr,b is also grateful to the Governments of Bangladesh, Canada, Sweden and the UK for providing core/unrestricted support.

Ancillary