The study was conducted in rural and semi-urban communities in the provinces of Pattani, Yala, Narathiwat, and Songkhla in southern Thailand.
Sample size calculation was based on the hypothesis that the proportion of delayed growth and development of children in an area with low intensity of armed conflict would be 0.40, while the proportion among children in an area with higher intensity of armed conflict would be 0.60. Using a formula for detection of difference between two proportions, a sample size of 107 cases per group was calculated using R software with Epicalc package. Since the study aimed to compare the effect of armed confict divided into 4 quartiles of intensity, a total of 428 subjects (4 groups with 107 subjects each) would be needed. Assuming a 40% non-response rate due to the nature of the study setting, a final sample size of 600 children (n = 600) was obtained.
Due to security concerns, data collection in home setting was deemed to be unsafe, and local community health centres (CHCs) became the study sites. The CHC is a sub-district-level primary healthcare facility normally staffed by 3–4 public health officers and/or nurses. Fifty out of 405 CHCs in the study area were randomly selected. Staff in each of the selected CHCs provided the researchers with a list of children aged 1 to 5 years who had been living within the service area of the CHC for the past 6 months. Twelve children were then randomly selected from the list, with three children for each age group stratum (12–23 months, 24–35 months, 36–47 months, 48–60 months), resulting in the total number of 12 children per CHC. Children aged 6–11 months were excluded from the sampling frame due to the reluctance of their parents to bring them to the CHC. Such reluctance is the normal child-rearing practice in the area: parents prefer not to take young children outside the home unless in case of necessity.
A list of participants selected by the above mentioned process was given to the health care providers and coordinating volunteers to verify eligibility. Exclusion criteria included physical disability, neurological disorder, psychiatric disorders and chronic medical conditions e.g. autism, HIV/AIDS or other chronic diseases. Those with direct experience of injury or death of family members due to the armed conflict were also excluded because of the imminent effect of such events on the study outcomes.
The primary caregivers of eligible children were informed about the purposes of the study and invited to participate in the study. Both the child and the caregiver were asked to visit the health center for data collection at the appointed time and date. Data were collected from February to May 2010, and most of the data were collected in the morning. Fortunately, no visit was postponed.
Anthropometric measurement for weight and height was taken by trained research assistants using standard instruments. The length of children who were too young to stand was measured in supine position on a length board with 0.1 cm precision. The height of children who were able to stand was recorded without shoes using a height-measuring board with 0.1 cm precision. Children who could not stand were measured for their height in supine position instead of the standing position. The weight of the child was measured by a beam balance scale with 0.1 kg precision. Shoes and clothing were removed to be as light as culturally appropriate before weighting.
The 125-items Denver Development Screening Test II (Thai version)  was used to assess personal-social function (25 items), fine motor ability (29 items), language development (39 items), and gross motor ability (32 items). The Denver Test was designed to be used in all children up to 6 years of age. Assessment criteria would vary by the age of the child being examined. The outcome of the test was categorized into normal and suspected delay groups based on the child's level of ability to perform a given task. A child would be considered as having suspected delay when the results showed two or more items under "caution" or one or more item under "delayed". Any child who refused to perform a given task would be categorized as being un-testable. A child classified as having suspected delay or being un-testable would be requested to come for a re-evaluation 1–2 weeks later.
The prominent covariates associated with child growth and development suggested by WHO  were taken into account during data analysis. These covariates included immediate factors (e.g. dietary intake and history of illness), underlying factors (e.g. food and health accessibility, sanitation, child-rearing practice), psycho-social factors, and basic factors (socio-economic status of the family and community). Missing covariates were not included in the final analysis. Potential confounders were identified based on the results of previous studies found during literature review.
Date of birth was assessed from the birth certificate. Birth weight was assessed from the child’s health records, the CHC's database, and cross-referenced with the mother's medical history. In case of discrepancy between the 3 sources, the weight that was agreed by 2 sources would be used as the final weight. If all 3 sources were discrepant, the weight from child’s health records was used since it would be the most accurate. History of illness within the previous 12 months was assessed from the child’s health records and through face-to-face interview with the primary caregiver. A 24-hour food recall was assessed by a qualified nutritionist to measure food intake. Food insecurity was measured by the Household Food Insecurity Access Scale (HFIAS) , while information on child-rearing practice, health service accessibility, and household sanitation was obtained using a structured questionnaire. The Depression Screening Scale of the Department of Mental Health, Thai Ministry of Public Health  was applied to measure depressive disorder among primary caregivers.
Data were coded and computerized (by double entry) using EpiData software version 3.0 . The INMUCAL-Nutrient software 4th edition database  was used to compute amount of nutrients from each raw food item that the child consumed. While all values of macro nutrients (carbohydrate, protein, fat) could be searched from the database, it was not possible to get all micronutrient values for local food items. Only micronutrients whose data were available in 80% or more of all food items in the INMUCAL database were included in statistical analysis. Adequacy of each nutrient intake was determined based on the 2003 Dietary Reference Intake for Thai People . Data analysis was carried out using R software.
Prevalences of growth failures were computed with descriptive statistics. Based on 2006 WHO growth standards , weight in kilogram, height in centimetre, and age in month of the children were converted to z-scores of weight-for-age (WAZ), weight-for-height (WHZ) and height-for-age (HAZ). Stunting, underweight and wasting were defined as having a z-score of more than 2 standard deviations below the reference for HAZ, WAZ and WHZ, respectively.
The average annual income of each sub-district was obtained from the database of the Thai Ministry of the Interior . The numbers of violent events per month in each sub-district over a 6-year period during January 2004-May 2010 were obtained from the database maintained by the Deep South Coordination Center (DSCC), an impartial non-state actor (NSA) which collects information of victims of violence in the deep south of Thailand . Based on the DSCC data, the intensity of violence in each sub-district was calculated as the average number of incidents per 100,000 population per year and was divided into quartiles.
Multi-level modeling (MLM) analysis using backward elimination procedure was employed to determine the association between the intensity of armed conflict and child growth and development. Adjustments were made for other variables in two levels, i.e. the sub-district level and the individual level. Sub-district level variable was the average annual income per household in the child’s sub-district. Individual-level variables included age group of the child, the child’s birthweight, caregiver’s education, annual household income and ethnicity. The variables which had p-value of less than 0.2 in univariate analysis were included in the full model. ANOVA test was performed to fit the final model.
The study was approved by the Ethical Committee of the Faculty of Medicine, Prince of Songkla University Hat Yai Campus.