Children studied were enrolled in the FIRST BREATH trial  and participated in the follow-on BRAIN-HIT (ClinicalTrials.gov: NCT00639184). BRAIN-HIT was a parallel-randomized controlled trial conducted in rural communities in India, Pakistan, and Zambia from 2007 to 2010 that aimed to determine whether a home-based, parent-provided early developmental intervention (EDI) plus WHO Enhanced Health Education Counseling (HC) would improve the BSID-II MDI scores at 36 months when compared to HC only in infants who have had birth asphyxia. The trial constituted two treatment populations: (1) babies who had mild-moderate birth asphyxia and were resuscitated via bag and mask ventilation and (2) non-resuscitated babies without perinatal complications who served as the healthy comparison group. Mild-moderate birth asphyxia was defined as insufficient breathing at birth and needing positive pressure ventilation. Infants were eligible for BRAIN-HIT if they met the following criteria: (1) weighed at least 1500 g at birth, (2) had a normal neurological examination (Stage I or II on the Ellis scale), and (3) were willing to participate in an intervention program for 36 months. Infants were ineligible if the mother was not contacted within 7 days of giving birth, younger than 15 years of age, unable/unwilling to participate, or not planning to stay in the study communities for the subsequent three years.
BRAIN implemented a modified version of the WHO Integrated Management of Childhood Illnesses Program (IMCI)  for the Health and Safety Counseling curriculum used in both arms of the trial. At enrollment and the 2-week visit, demographics and family resources were noted. At 12-, 24- and 36-months of age, information on family resources, health status and growth measurements were collected as well as neurodevelopmental assessments for BSID-II MDI and PDI scores. Child diet information was collected at 36 months of age. Full details of the BRAIN-HIT protocol have been published .
Among the 540 infants screened, 438 (81%) were eligible for participation with 407 (93%) of the eligible infants having mothers who consented to participate in the study . Among infants whose mothers consented, 371 (91%) had completed dietary forms and health evaluations at 36 months and are included in the current analysis (Fig. 1).
Dietary methods and indicators
Young child feeding practices were evaluated during the last visit at 36 months of age by use of a qualitative dietary questionnaire, including number of meals and food groups consumed on an average day. In 2008, WHO published a set of reliable and valid core indicators to assess feeding practices in children up to two years old. Since BRAIN-HIT collected data on dietary practices at 36 months of age only, we applied the WHO recommended diet up to 24 months to the 36 month intake assessment. The three core indicators of interest were modified versions of the Minimum Dietary Diversity (mMDD), Minimum Meal Frequency (mMMF), and Minimum Acceptable Diet (mMAD). Children met the mMDD if they received at least four of the following food groups: (1) grains, roots or tubers; (2) legumes or nuts; (3) dairy products; (4) flesh foods (meat, poultry, or insects); (5) eggs; and (6) fruits or vegetables. BRAIN-HIT did not distinguish between vitamin-A rich and non-vitamin-A rich fruits and vegetables and thus, all fruit and vegetables were combined in one food group. However, the mMDD was considered met if at least four food groups were being regularly consumed in a day, as per the WHO definition. To achieve mMMF, BRAIN-HIT applied the WHO Integrated Management of Childhood Illness (IMCI) program recommendation of three meals with two additional snacks a day for this age group. Finally, children met the mMAD if they met both the mMDD and mMMF.
Data were collected on demographic characteristics of the child (sex, premature birth, birth weight, resuscitation status, exclusively breast fed the first 6 months) and mother (socioeconomic status, age, educational level, marital status, parity).
Three anthropometric measures were taken at each visit: length/height, weight, and head circumference. Weight was measured using digital infant weighing scales in Pakistan and analog scales in India and Zambia. Head circumference was measured using a plasticized or fiberglass non-elastic tape measure. Standing height was measured in Zambia while recumbent length was measured in India and Pakistan. In India, children were made to lie supine on clean level floors with their heads resting against the wall and looking straight at the ceiling. Their legs were straightened at the knee with their feet perpendicular to the leg while ensuring their heads, backs, buttocks, and heels were in contact with the floor. Hard boards were then placed against the foot and lengths were measured by placing the measuring tape from the wall to inner edge of the board. Children were properly restrained during the procedure. In Pakistan, recumbent length was measured two times with a Seca 416 infantometer (Perspective Enterprises, Portage, MI). If the two initial measurements differed by more than 0.2 cm, a third measurement was undertaken. Anthropometric instruments were calibrated regularly.
Weight, length/height, and head circumference measurements taken at 36 months of age were used to determine weight-for-age (WAZ), height-for-age (HAZ), weight-for-height (WHZ) and head circumference-for-age Z-scores (HCZ) based on international growth standards developed by WHO for children up to 5 years of age . Lengths were converted to heights by subtracting 0.7 cm as recommended by WHO . Stunting and wasting were defined as HAZ < -2SD and WHZ < -2SD, respectively.
The BSID-II is a well-validated measure of development in infants aged 1–42 months . Both the BSID-II MDI- and PDI- scores, which measure cognitive development and motor skills, respectively, were used to assess development at 36 months. The BSID-II was used as a main measure for the BRAIN-HIT trial due to its extensive use in a number of LMIC. To verify the validity of BSID-II in the local context, it was pretested at each site and a few items were modified to ensure cultural appropriateness (e.g., image of a sandal in place of a shoe). The BSID-II was administered to each child in the appropriate language using standard material by certified neurodevelopmental evaluators (pediatricians and psychologists familiar with the local language and culture) who were masked to the birth history and intervention group. All evaluators received an intensive 4-day training in the purpose and correct administration of each item.
Descriptive statistics were calculated for child and maternal characteristics. Frequencies and percentages were reported for categorical variables with differences in characteristics between sites tested for by chi-square and Fisher’s exact test. Means, standard deviations, medians, minimums and maximums were reported for continuous variables with difference in means tested using the Kruskal Wallis test.
Linear regression models fitting each BSID-II Score index (i.e. MDI and PDI) and anthropometric measures (i.e. WAZ, HAZ, WHZ, and HCZ) were used to estimate adjusted mean scores in groups of children defined by dietary consumption. Models included one dietary indicator at a time as the primary independent variable with site, intervention group, resuscitation status, socioeconomic status, child sex, exclusively breastfed first 6 months, birth weight, preterm status, maternal age, and maternal education level as covariates. Results from the main trial suggested some heterogeneity between neurodevelopmental evaluator scoring. To account for potential confounding due to evaluator, evaluator within site was included as a nested effect in the BSID-II models. Statistical significance for a difference in adjusted means for those with and without each dietary indicator was determined by the F test. Tests of interaction were conducted to assess whether the relationship between the dietary indicator and outcome differed between the three study sites with means shown by site if interactions were significant.
Adjusted relative risks (RR) and 95% confidence intervals (CI) for stunting and wasting were estimated using Poisson regression models with robust variance estimators  assuming an independent correlation structure. Relative risks were adjusted for site, intervention group, resuscitation status, and child and maternal characteristics with statistical significance determined by Wald chi-square tests. All analyses were conducted in SAS version 9.4.