Study participants
All data were recorded by trained community health volunteers in March 2014 in ward
22 of Lalitpur district, an urban area near Kathmandu, Nepal. This ward was selected
because it’s one of the least developed wards and requires particularly high input
of health services and welfare benefits from the public health unit of LSMC. There
are 22 Wards in LSMC and Ward 22 contains 13 Toles (districts). In total, data from 294 mother-child dyads were recorded but 5 were
excluded due to missing information. All mothers from 13 Toles of Ward 22, who had at least one child younger than 5 years of age, were interviewed
for the survey by door-to-door contact. The applied questionnaire was pretested in
an urban setting, revised and then pilot-retested. A group of 10 local community health
volunteers (Tole Health Promoters, THPs, and Ward Health-In-Charges, WHIC) from ward 22 of the Lalitpur
district were recruited by the local Public Health Office. These recruits attended
an intensive 3 day training on survey methodology, including administration of the
questionnaire and how to carry out anthropometric measurements of mothers and their
children. The ten trained recruits were divided into groups of a minimum of three,
which were accompanied by 3 staff members of the Nutrition Promotion and Consultancy
Services (NPCS) to supervise data collection during the entire period, All groups
had both male and female enumerators and most of the interviews were conducted in
Nepali but a few were conducted in Newari language, the language of the majority of people living in that Ward.
The questionnaire contained questions on: self-assessed food security; food safety;
supplement intake; health check-ups; infant feeding practices; along with anthropometric
measurements of the mother and her child, if under 60 months of age. Demographic information
such as the age of the mother at birth of her firstborn, the number of family members,
the number of children in the family (segregated by sex), and the time since the birth
of her youngest child (3-level scale) were recorded. Antenatal iron supplementation
(8-level scale) for the child in question was also documented.
Food security was assessed based on having enough money to cover food expenses (4-level
scale), availability of desired food (2-level scale), and land ownership (2-level
scale). To avoid bias of using only one question to assess food security, the consistency
of answers to these three questions was checked before analysis. The following food
safety indicators were recorded: self-reported habit of washing hands in the kitchen
and the toilet for the child and the mother, with or without soap (2?×?3 levels);
water and soap availability at interview (3 levels); the type of water source (bought/tap/well/other);
the method used to decontaminate water (boiling/filter/sunlight-UV/untreated); storage
of food (in the fridge, covered or uncovered); and cleaning procedures for fruits
before consumption (washing, or not, with water of different safety levels). A possible
bias when recording the food safety parameters was that some women could misreport
that they washed their hands with soap and water after toilet use. To overcome this
bias, data collectors were asked to check toilets or hand washing basins in each household
to see if soap was present or not. The data collectors sought permission from the
mothers before checking their toilets or hand washing basins.
The socio-economic status was assessed based on the duration of the women’s education,
their employment situation, and the ownership of land. The mothers’ education level
was grouped into the following categories: illiterate, attending literate classes
for adults, elementary school, secondary school, and any education going beyond secondary
school. The professional activity of the women was assigned to one of the following
categories: housewife (without profession), employee with a rather sedentary lifestyle,
running own business (usually maintaining small selling booths), and workers doing
physically demanding labour.
Anthropometric indices were recorded using a calibrated balance (Seca 874 U, Hamburg,
Germany) and a stadiometer (UNICEF, S0114400 Height measuring instrument). The weight
and the height, the z-scores of the height-for age (HAZ), the weight-for-age (WAZ)
and the body-mass-for age (BAZ) were calculated using WHO AnthroPlus 19] where the age of the child was calculated as the difference between the date of recording
and their date of birth. Moderate or severe stunting of the children were defined
according to WHO standards 20]. Accordingly, severe and moderate underweight were classified using the WAZ, while
overweight and obesity were defined on the basis of the BAZ 21].
A data collector from each group had to review the completed questionnaires to ensure
accuracy of data collection and recording at end of each day.
Statistics
The dependence of HAZ and BAZ of the children and the BMI of the mothers on the recorded
(independent) data was evaluated with multivariable linear regression using a step-wise
backward removal procedure (threshold P value: 0.05). First, a complete multivariable linear regression model was built including
all recorded variables considered to be linked to stunting, overweight, or obesity.
Then, in subsequent steps, the variable that was the least significant was recursively
removed until the multivariable regression model contained independent variables with
significant betas only (= minimum adequate model, MAM). Ordinal variables (e.g. education
level, food safety level of drinking water) were included into the regression analysis
due to their rank. Two-level categorical predictor variables (e.g. male/female, yes/no)
were subjected to sigma-parametrization, which means that they were re-coded to values
of 1 or ?1 and then implemented into the multivariable linear regression analysis.
Correlations with a P value below 0.05 were considered as significant. For multi-group comparisons, ANOVA
was applied with Fisher’s LSD post-hoc test. Simple correlations (BMI of the mothers
vs. HAZ and BAZ of the children) were analysed with Pearson’s test. P values of distribution inhomogeneity (crosstabulation testing) were calculated with
Fisher’s exact test. Software packages used were STATISTICA® V. 12.1 (StatSoft Inc.,
Tulsa, OK) and [R] V. 3.1.1 (The R Foundation, Vienna, Austria) along with RStudio
V0.98.987 (RStudio Inc., Boston, MA).