One finding of this paper is that the differences in growth between children in the studied Vietnamese urban and rural areas previously reported during the first year of life , remain at two years of age. In addition, we found a positive association between weight growth and early initiation of breastfeeding and a negative association with reported illness symptoms.
Much evidence supports that breastfeeding provides good nutrition for children as it reduces the severity of e.g. respiratory and gastrointestinal infections in children [32–34]. Children with exclusive breastfeeding have been seen to grow better  and breastfeeding can be associated with reduced risk of obesity later in life compared with formula fed infants . One suggested reason is that breastfeeding protects through activity of specific components of breast milk such as hormones involved in appetite and energy balance .
Poor nutrition has been seen as the most important risk for poor growth  and differences in nutrition between urban and rural areas could be the main reason for the observed differences in this study. During the last decades, the Vietnamese dietary intake has improved in both quality and quantitive through consumption of food such as fish, meat, fat oils, etc. . However, differences in food consumption between urban and rural areas in Vietnam have been reported [38, 39]. Deficiency of iron, calcium, phosphorus, potassium, magnesium, beta- carotene, vitamin A and vitamin C has been found in Vietnamese rural girls 7–9 years old in spite of adequate consumption of all these elements except low carotene . The nutritional status of under five children is proposed as a sensitive indicator of household economic condition and parent’s education . Differences in nutrition between the urban and rural areas could be a strong reason for the observed growth differences.
According to WHO and UNICEF, the prevalence of stunting among children under five in Asia was 27% in 2011 . In Vietnam, one third of children under five were stunted . In the present study, the percentage of measurements indicating stunting two years after birth was over 20% indicating that the prevalence of stunted children is quite high.
The Vietnamese government has noted stunting as a public health problem. A plan to reduce the incidence of stunting to 23% by 2020 and underweight to 12.5% by the same year in children under five was launched in 2012 . A contributing factor for the high stunting in boys may be that boys are less breastfed than girls . The reason for this can be that mothers consider boys to be more important than girls and at the same time think that formula feeding is better than breastfeeding .
We observed some statistically significant simple (unadjusted) correlations between growth and ANC use. In, regression models where both underlying and immediate variables are included though, the education and assets variables turn out to be more important than the ANC indicators. The partial correlations between growth and ANC use, adjusting for education level of the mother and the household resources, are small and not statistically significant. The simple explanation can be that the ANC variables are themselves associated with education and economy. Socially and economically resourceful mothers possibly use, and benefit, more from ANC than others.
Child illness stands out among the studied immediate factors. A fairly strong association between growth of children and reported illness symptoms was found, particularly in the rural area.
Symptoms of illness were most commonly reported in the rural area. The risk for illness reporting at a particular visit was 0.40 compared to about 0.20 in the urban. The high incidence of illness could be important to explain the slow growth of rural infants. The possible negative influence on weight growth was also stronger in the rural area.
The most common causes of illness in children under five and especially during infancy are diarrhea and acute respiratory infection. This has been observed in several studies [42–44]. Diarrhoea was concluded to drastically reduce the growth velocity in weight and length e.g. in a Brazilian study  where diarrhoea during the first six months increased the risk of low BMI and weight for length later. Diarrhea after six months of age increased the risk for low weight for age and stunting in a Vietnamese study . Acute respiratory infection has also been seen to be significantly associated with incremental weight loss of infants e.g. in Indonesia .
A possible intervention may be that the Vietnamese Ministry of Health work to enhance the quality of the Integrated Management of Childhood Illness program aimed to help lay community health workers assess and treat sick children. Improvements of health staff skills as well as the health system itself seem also to be needed particularly in the rural area.
Early initiation of breastfeeding, within the first hour of life, can be a positive factor for growth and has been claimed to protect newborn from acquiring infections . The main reason why breastfeeding protects babies from infectious diseases is that it modulates the early exposure of neonate’s intestinal mucosa to microbes and limits bacterial translocation through the gut mucosa . Being more common in the urban than in the rural area , early initiation of breastfeeding can be another reason for the different growth of infants in rural and urban area.
Exclusively breastfed infants have been seen to grow faster during the first 6 months of life compared to groups of weaned and partially breastfed children [15, 47]. Exclusive breastfeeding can decrease the number of diarrhea and acute respiratory infection episodes . The duration of exclusive breastfeeding though, does not relate to growth in the present study. A reason may be that the duration is short in both areas; less than two months for most children.
Decisions taken by mothers about use of ANC, breastfeeding, nutrition and child health care utilization are related to the educational level of mothers and the household resources. Likewise, the risks for illness are associated with education and economy. In the present study about 19% of the variation in weight growth and 12% for length growth are explained by the variation in education of the mother and household wealth. Adding the ANC indicators, early initiation of breastfeeding and illness symptoms as independent variables in the regression model increased the determination coefficient by about two percent units. However, more than 80% of the total growth variation is left unexplained.
The associations between growth and the immediate factors in the conceptual framework for this study are to large extents reflections of the associations between growth and the underlying socio-economic factors. Thus these turn out to be the most important to explain growth variation. Interventions aimed at improved provision and use of antenatal care, promoting good breastfeeding practices and preventing child morbidity will have their effects. These risk to be limited as long as the underlying social and economic conditions are not strong and equitable. In the end the basic factors with its political, social, economic, cultural and other contexts will determine the conditions for child growth.
A child with a complete weight and length measurement set has been measured 12 times during the first year of life and four times during the second. As can be seen in Table 1, complete sets were not received in 34.2% of the included children. The most common reasons for dropout was that the visit to the household could not take place for practical reasons or the mother declined to cooperate.
The dropout rates were clearly different between the urban and the rural area and to some extent between boys and girls in the rural area. The first mentioned difference could be expected since many mothers in the urban area work outside the household and visits could be more difficult to arrange.
The dropouts cannot be expected to be random but systematic and possibly creating bias. To investigate we compared the growth curves fitted using all available observations with other curves using only the data from children with complete sets. The former curves came out systematically lower than the latter but the differences were small, 30 to 50 gram after two years of age, largest in DodaLab. Another approach used to investigate possible bias was to correlate the means of relative residuals for the first half-year to the number of visits. Very weak positive correlations were found. Both approaches thus indicate that the risk to dropout is higher for children with slower growth. Correlations between birth weight and number of measurements however, did not support that conclusion.
The study is to a large extent dependent on information reported by the mothers, e.g. birth weight. . Possible sources of errors are both how the measured birth weight was reported to the mother and the recall of mothers. It could be suspected that the hospital or health center staff tends to report a higher weight to please the mother. The proportion of low birth weight newborn (birth weight below 2,500 gram) is lower than expected. On the other hand there is no heaping e.g. at 2,500 gram in the birth weight distribution. The precision of birth weight reporting is 100 gram. Systematically and incorrectly rounding upwards would create a bias of that size. Recall biases are likely to be small as it is considered important for a mother to remember the birth weight of a child in the Vietnamese tradition. Another important piece of information provided by the mothers was the date of the last menstruation before pregnancy. This is needed to calculate the gestational age at birth. The adequacy of this information turned out to be somewhat problematic. The information is missing for quite many women and it seems that by using the available information we underestimate the gestational age. Too many women appear to be classified as giving birth prematurely. However, using or not using the gestational age as an explanatory variable for growth does influence the results only marginally.
Household assets have been used as an indicator of household economical resources. The available alternatives could be the reported household incomes or reported expenditures. A third alternative could be the “Wealth index” which is a wider combination of housing characteristics and assets [49, 50]. All these indicators have been tried in the present analysis, one by one and in combinations, although they are strongly correlated. The asset index was concluded to have the strongest correlation to growth. We also tried to use both assets and household income in the same model. Then the correlations with assets variable come out statistically significant whereas correlations with income are smaller and non-significant.