Design and participants
We used repeated cross-sectional data comprising two surveys conducted in six counties of northern and southern China in 2013 and 2016, respectively. The first survey was conducted prior to an early childhood development program in six countries of two provinces from July to September 2013: Songtao, Liping, and Pan Counties in Guizhou Province, and Fenxi, Lin, and Fangshan Counties in Shanxi Province [23]. In each county, a clustered random sampling method was used to select villages that met the following criteria: reachable by car from the county capital, with ≥ 50 resident children under 3 years of age, and with a sufficient number of caregivers willing to participate in the survey. Finally, totals of 40 intervention villages and 43 control villages were included in the study. All children under 3 years old in these selected villages were eligible for participation in the survey. The second survey was conducted in the same villages from July to September 2016 by the same method, and the participants were another group of children under 3 years old at this survey time who were not included in the first survey. Finally, totals of 2,953 and 2,745 children under 3 years old were enrolled in 2013 and 2016, respectively. According to the number of eligible children (n=4,288) in 2013, the response rate was 68.9% in 2013 and 64.0% in 2016.
According to the research objective, our analysis did not include children under 6 months old (375 in 2013 and 335 in 2016) because they were not measured for hemoglobin (Hb) concentration. Among children aged 6–35 months who were measured for at least one nutritional outcome (length, weight or hemoglobin concentration) (2,560 in 2013 and 2,398 in 2016), we also excluded twins, single-parent children, and those with serious diseases or disabilities (186 in 2013 and 125 in 2016). In addition, we did not include children with only migrant mothers because of the small number of such children (38 in 2013 and 63 in 2016). Finally, this present study only used data of children aged 6–35 months with available nutritional outcomes and with non-migrant parents (NLBC), migrant fathers (FLBC), or both migrant parents (PLBC).
Measure
Face-to-face interviews with caregivers were conducted by uniformly trained local health workers. Data were immediately input, saved, and transmitted to statisticians by means of an electronic questionnaire application, which had a basic logic and integrity-checking function to enable investigators to correct errors and supplement omissions in a timely manner.
Basic characteristics
The following sociodemographic characteristics of the children and their caregivers were collected: (a) child gender, age, preterm, and ethnicity; and (b) caregiver’s relationship to the child, gender, age, and educational attainment. Depression among caregivers was measured using Zung’s self-rating depression scale (ZSDS), which is validated and used worldwide and consists of 20 items representing depressive features with a total score ranging from 20 to 80 [24]. ZSDS was administered by trained interviewers, and a ZSDS score of ≥ 50 was defined as depression. Household economic status was measured by the number of the following household electrical appliances and vehicles owned: telephone, washing machine, refrigerator, and TV in 2013; and telephone, washing machine, refrigerator, TV, motorcycle, tricycle, and car in 2016. Low household economic status was defined as owning fewer than three and four of these items in 2013 and 2016, respectively.
Some children in Songtao, Liping, Fenxi, and Lin counties received the interventions of the early childhood development program from 2014. Therefore, two variables related to the interventions were measured and controlled for as confounding factors in the analysis of the 2016 survey data. The first variable is the frequency of consumption of Yingyangbao supplements (a daily intake package of soybean-based micronutrient fortified powders), which was classified as: never supplemented, ever supplemented but none in the past week, 0–6 packages in the past week, and ≥ 7 packages in the past week. The second is the frequency of nutritional consultations with health-care workers in the past 6 months, which was classified as: never, less than once per month, and one or more per month.
Child breastfeeding and dietary intakes
As recommended in Indicators for Assessing Infant and Young Child Feeding Practices by the World Health Organization (WHO) [25], a 24-hour reported food recall was performed to assess breastfeeding and the dietary intake of the following seven food groups: (a) grains, roots and tubers, (b) legumes and nuts, (c) dairy products, (d) flesh foods (meat or fish), (e) eggs, (f) vitamin-A rich fruits and vegetables, and (g) other fruits and vegetables. Then the following indicators were calculated: (1) ever being breastfed, (2) duration of breastfeeding (only for children weaned from breastfeeding), (3) meal frequency (frequency of dairy, solid, semi-solid, and soft food intake during the previous day), and (4) dietary diversity (number of food groups children consumed during the previous day).
Nutritional outcomes
The length, weight, and peripheral blood Hb concentration of the children were measured using standard procedures by uniformly trained health‑care workers who were blinded to the above interview information. Children were weighed twice in light clothes without shoes using an electronic weight scale with 0.01 kg accuracy. Their recumbent, barefooted, and bareheaded lengths were measured twice using a standard infant length scale with 0.1 cm accuracy. A third measurement was performed if the two measurements differed by 1.0 cm or more for length and 0.5 kg or more for weight. The Hb concentration of the children was measured using HemoCue201+ (HemoCue AB Inc.). The length-for-age Z score (LAZ), weight-for-age Z score (WAZ), and weight-for-length Z score (WLZ) were calculated according to the WHO Child Growth Standards and the corresponding score of < −2 was used to define stunting, underweight and wasting, respectively [26]. According to the WHO guidelines, Hb concentration was adjusted for altitude and children with an altitude-adjusted Hb concentration of < 11 g/dL were considered to be anemic [27].
Statistical analysis
To double validate any associations and take into account the effect of time, we analyzed the data of the two time points independently. Univariate analysis was first conducted to compare the basic characteristics and nutritional outcomes according to parental migration status by t-test and Mann-Whitney U test for continuous variables and the chi‑squared test for categorical variables.
Generalized linear regressions were performed to estimate the adjusted mean differences (aMDs) with 95% confidence intervals (CIs) in LAZ, WAZ, and WLZ scores and Hb concentration according to parental migration status at each time point, after controlling for resident county, child gender, age, preterm and ethnicity in 2013, as well as additional two intervention variables in 2016. Household economic status and caregiver’s relationship to the child, gender, age, and educational attainment were not controlled for because they were severely collinear with parental migration status and were strongly considered as mediators of the effects of parental migration on child nutrition. For categorical nutritional outcomes including stunting, underweight, wasting, and anemia, multivariate logistic regressions were employed to calculate the adjusted odds ratios (aORs) and 95% CIs after controlling for the same covariates as above.
Similar multivariable adjusted analyses were also performed on Hb concentration and anemia for children aged 6–17 and 18–35 months, respectively, based on the result of the preliminary analysis that the differences in Hb concentration and the risk of anemia between children with different parental migration statuses varied before and after the age of 18 months.
Statistical analyses were performed using Statistical Package for the Social Sciences (SPSS) software 20.0 (SPSS, Inc., Chicago, IL). A two-tailed p-value of < 0.05 was taken to reflect statistical significance.