Study area
The four study sites are located in the districts of Surkhet, Achham, and Dailekh in the Karnali province of Nepal. These districts were selected for their combination of conditions: the region is mountainous, incomes are low, infrastructure is lacking, access is difficult and WASH conditions are challenging.
Study design, sample population, sample size and sampling methods
The study sites were located within Helvetas Swiss Intercooperation Nepal’s Water Resources Management Programme (WARM-P) sites. The sites were selected according to the following criteria: (a) extremely remote location, (b) the availability of a piped water supply scheme that has not yet received the WARM-P training interventions, and (c) the population not having access to products for household water treatment. The cross-sectional study was conducted from March to May 2018 and involved 1427 households with children aged 6 months up to 10 years. Sample size and statistical power were calculated using G*Power 3.1. A sample size of 300 households was required at each of the four sites to detect an effect in Cohen’s f2 at one-tailed alpha of 0.05 and a statistical power of 90% with multiple logistic regression and 15 predictor variables adjusting for clustering effect [19, 20]. We therefore randomly sampled a minimum of 345 households at each of the four sites (Surkhet A, Surkhet B, Dailekh, and Achham). If a household declined to participate, then the neighbouring household was included.
Questionnaire survey
A quantitative structured questionnaire was administered to the caregivers, whether male or female, in the households. When both caregivers were available, the interviews were conducted with the mother. The questionnaire contained both closed and open-ended multiple-choice questions with Likert-scale answer categories. The interviews took around 15 minutes on average and were complemented by structured observations. The questionnaire was coded in Open Data Kit software [21] on tablets (Samsung Galaxy note 10.1 N8010) and contained questions on the use of water sources, psychological factors concerning water handling and hygiene practices, observations of WASH infrastructure, information on WASH promotion activities received, nutrition provided to children, children’s history of waterborne illnesses in the past 7 days, and potential confounders, such as sociodemographic and socioeconomic factors. The questionnaire was developed from standardized questions following international guidelines and incorporated amendments necessary to meet the conditions in our study areas [22, 23]. The questionnaire was developed in English and translated into Nepali with a back-translation. Interviewers were intensively trained for two days on data-collection procedures. The questionnaires were pretested and adapted to ascertain the reliability of the questions used in the final survey instrument [24].
Child diet and household food security
Child dietary information was assessed following the guidelines of the Food and Agricultural Organisation (FAO) [23]. The caregivers were asked to recall whether nine different food groups were consumed within the past 7 days and, if consumed, the frequency of consumption. The supervisors randomly re-interviewed a subset of 10% of the surveyed households to assess reproducibility. Household food security was assessed with questions relating to the availability of food during the entire year.
Anthropometric measurements
Certified medical assistants collected anthropometric measurements adhering to standard procedures [25]. Supine lengths were obtained for children younger than 2 years old using a Seca BabyMat 210. For children aged between 24 months and 10 years, a height-measuring board and a digital scale (Seca 877; Hamburg, Germany) were used with precisions of 0.1 cm and 0.1 kg, respectively. Before each measurement, scales were calibrated with a standardized weight, and the children were measured barefoot [24]. Anthropometric indices were calculated using AnthroPlus (WHO; Geneva, Switzerland) in accordance with the World Health Organisation (WHO) guidelines [25, 26]. Age in total months was recorded for each child and confirmed by an inspection of the birth certificate. Three anthropometric indices were expressed as z-scores (i.e. differences from the median in standard deviations): (a) weight for age (WAZ, underweight); (b) height for age (HAZ, stunting); and (c) body mass index for age (BMIZ, thinness) [4]. Z-scores of ≥ -2 were regarded as normal, those between < -2 and ≥ -3 as moderate undernutrition and those below < -3 as severe undernutrition [31,32]. Children were considered to be undernourished if at least one of the anthropometric indices indicated undernutrition.
Parasitological survey
On the day following the household survey visit, caregivers were asked to provide a fresh morning stool sample without urine contamination from the participating child. The samples were processed on the same day by two experienced laboratory technicians. Each stool sample was analysed using direct wet-mount and formalin-ether concentration techniques to detect intestinal protozoa and helminths following standard guidelines [27–29]. In addition, a duplicate Kato-Katz thick smear was prepared for the diagnosis of helminths [30]. The presence of infection by any worm species was defined by the detection of one or more eggs on either slide [7]. Children were considered positive if at least one of the diagnostic techniques revealed an infection. The infection intensity of helminths was calculated according to criteria defined by the WHO and multiplied by 24 to reach the total number of eggs per gram (EPG) of stool. Infection intensities were then classified as light, moderate, and heavy as per the thresholds set by the WHO [5, 31].
Signs and symptoms of nutritional deficiencies
During the household survey, the children were screened by the certified medical assistants using a standard checklist for clinical signs and symptoms of nutritional deficiencies: wasted appearance (protein deficiency), loss of hair pigment and easy pluckability (protein deficiency), Bitot’s spots (vitamin A deficiency), dry and infected cornea (vitamin A deficiency), oedema (thiamine deficiency), several types of dermatitis (niacin deficiency), spongy bleeding gums (vitamin C deficiency), pale conjunctiva (iron-deficiency anaemia), red inflamed tongue (riboflavin deficiency), sub-dermal haemorrhage (vitamin C deficiency), bowed legs (vitamin D deficiency), and goitre (iodine deficiency) [32].
Drinking water quality examination
Water samples were collected from the household’s main drinking water source and from the container used for drinking water transport and storage. The sample at the source was taken after letting the water run for 60 seconds from the tap, and then filling a sterile Nasco Whirl-Pak bag. The caregivers were requested to bring fresh drinking water from the point of collection to the household in the same container they usually use to fetch their drinking water [33]. A second water sample was taken by pouring water from the transport container into another sterile Nasco Whirl-Pak bag. All water samples were stored inside cooler bags and analysed at the field site using the membrane filtration technique: 100 ml water samples were passed through sterile 0.45 µm millipore cellulose membrane filters with sterilized filtration equipment. The filter pads were plated on Nissui Compact Dry Coli-scan plates and incubated for 24 hours at 35 +/- 2°C [15]. A solar-powered electrical system was set up to run a low-power incubator at the field site. Colony-forming units of total coliforms and Escherichia coli (E. coli) were counted after 24 hours of incubation.
Data management and statistical analysis
Data cleaning was performed daily during the survey, and if any values were missing or inconsistent, the respective household was consulted the following day. Readings of intestinal parasite and nutritional deficiency screenings were double entered into an Excel 2010 spread sheet (Microsoft; Redmond, USA) and cross-checked. Numerical variables were described by means and standard deviations if normally distributed and by medians and interquartile range otherwise. Categorical variables were described by absolute and relative frequencies. We employed χ2 statistics to assess the differences in distribution for categorical variables between the study areas. Household socioeconomic status was characterized by factor analysis of reported household assets, such as the availability of electricity, radio, television, solar panel, mobile phone, bicycle, motorbike, fridge, watch, own house, and land ownership. Three factors reflecting three socioeconomic domains were retained and divided using the k-means procedure into three categories: (a) low, (b) medium, and (c) high [34]. The same procedure was applied to create one variable for the hygiene of containers used for transport and storage of drinking water, latrine hygiene, cleanliness of the household environment and kitchen, and personal hygiene. For each of these variables, three factors were retained and categorized, indicating (a) low, (b) medium, and (c) high status.
We assessed four health-related outcome variables: (a) intestinal parasitic infection; (b) diarrhoea; (c) undernutrition including stunting, underweight, and BMIZ (thinness); and (d) clinical signs and symptoms of nutritional deficiencies. The associations between the outcome variables and risk factors were first assessed by conducting univariate mixed logistic regression analyses with random intercepts of the study areas. Since only a few undernutrition cases were severe, the cases were pooled into a binary variable of stunted/non-stunted, and underweight/non-underweight for the subsequent analysis. Similarly, there was a low prevalence of parasites such as T. trichiura, E. vermicularis and Ancylostoma duodenale, so all reported intestinal parasitic infections were pooled into a binary variable of parasite infection/no infection to maximize statistical power. The clinical signs of nutritional deficiencies and diarrhoea outcomes were coded into binary variables for the subsequent comparative analysis.
We assessed associations between the binary outcome variables and hypothesized risk factors using logistic regression models with random study site intercepts and controlling for potential confounders with the demographic variables of age, sex, and socioeconomic status. First, the associations between outcome variables and risk factors were assessed using univariate models. Variables with p-values < 0.2 were retained for the maximal model. The final model was then obtained using backward selection with the same level of 0.2 [35]. Odds ratios were reported and the associations were considered as statistically significant if p-values were < 0.05. The statistical analysis was performed with STATA version 14 (STATA Corporation, College Station, TX, USA).