The results of this study show that the intervention was associated with a lower risk of overweight. The approximation of the mean Z score of this index towards the value corresponding to the WHO reference population (Z = 0), with a decrease in the proportion of children with a score above the 85th percentile without causing an increase in wasting, points to a healthy decrease in the risk of overweight. The size of the adjusted effect (OR = 0.43) is equivalent to a decrease by nearly 48% in relation to the prevalence in the control group. This effect is important under the preventive point of view; children with body mass index over 85th percentile during early childhood have a high probability of being overweight in adolescence [37, 38].
Since the study aimed to analyze not only the effect on nutritional indicators but also on various health aspects that the intervention intended to modify (within the integrated approach of the IMCI), no data were collected on some intermediary factors that could have explained better the causal pathway of the effects on overweight. However, the results observed in the activities related to growth monitoring, coupled with the fact that the nutritional counseling protocols included clear instructions about what issues should be discussed with the mother when the weight-for-age curve had a tendency towards the overweight-for-age zone, support the plausibility of attributing these results to the intervention.
As expected in impoverished communities, the mean length-for-age Z-score (LAZ) in both groups was below the corresponding level at the WHO reference population. The intervention did not show statistically significant effects on length, but point estimates found were in the desired direction, with a mean difference of 0.21, which is similar to that found in other educational interventions. In a recent review study carried out by Imdad et al. , eight studies were selected that evaluated interventions based on maternal education regarding complementary feeding, without the provision of food, resulting in a positive effect in linear growth with a weighted mean difference of 0.21.
A limitation on the probabilistic analysis of the effect of the intervention on the prevalence of stunting was the sample size. It was calculated for a one-tailed test based on an expected prevalence of 20% in the control group, while the found value was 12%. For the magnitude of the intervention effect that was found, it would have required a larger sample size for a significance level of 95%. Furthermore, the sample size calculation did not take into account the clustering.
Our study found positive effect in several indicators of intermediary factors. Whereas in the control group there was a gradual decrease in action regarding growth monitoring, falling from 51% of children weighed in the past 4 months to 28% in cases where an explanation was given to the mother about the results of the curve, up to 18% receiving feeding advice, the intervention group showed another profile: 83% of the children had been weighed and approximately 70% of the mothers had received both explanations about the growth curve and child feeding advice. The intervention performed by the lay health volunteers showed also significant effects in micronutrient supplementation. The proportion of children supplemented with vitamin A capsules and with ferrous sulfate drops was higher in the intervention group, thus resulting in a lower prevalence of anemia, with adjusted odds ratio of 0.57 (P=0.011) for Hb < 11.0 g/dl and 0.46 (P<0.001) for Hb < 10.0 g/dl (article in preparation). As we expected, effects on food consumption were not as positive as those on micronutrients consumption. Nevertheless there was an increase in exclusive breastfeeding and in the proportion of children who had consumed 5 or more solid food portions in the last 24 hours.
Upon reviewing the eight studies analyzed by Imdad et al. , we observed that none of them evaluated the possible effect of the intervention with regard to overweight. In the context of nutritional transition in low-income and middle-income countries, we can consider that the principal contribution of our study is to integrate the prevention of children becoming overweight in an intervention program focused primarily on preventing malnutrition. Furthermore, in current scientific literature regarding interventions where the primary goal is the prevention of childhood obesity, there are very few intervention studies that focus on the first two years of age . In a recent systematic review about interventions to prevent obesity among children aged 0–5 years, seven of twenty-five studies had included younger than 2 years of age, but just two had reported anthropometric measurements, one of them finding a positive trend towards overweight prevention . Another contribution from this present study is that its results suggest that in situations in which body-mass-index is difficult to be periodically assessed early intervention to prevent overweight can be based on monitoring the weight-for-age.
The outcomes were assessed against the WHO Child Growth Standards, while during the intervention the mothers were given feeding advice based on plotting the child’s growth against NCHS weight-for-age charts due to the fact that until then the new standards had not been adopted. As the average weight of infants included in the WHO standards was above the NCHS median during their first six months of life, and thereafter continued below , we can deduce that during the intervention there was a tendency to be less demanding in controlling risk of underweight in the first six months (when the NCHS references identify lower prevalence of underweight than WHO standards) and be more demanding in this matter after seven months (when NCHS references identify higher prevalence of underweight than WHO standards). Opposite trends would be present in regards to control the risk of higher than expected weight for age. We could think that these tendencies due to differences between the WHO standards and NCHS references were diminished during the intervention by the fact that counseling protocols oriented to act not only when achieving the cutoffs level of underweight or high weight for age, but also by observing early trends of diversion of the curve.
Among the limitations of this study is included the fact that we did not use an intention to treat analysis, due to difficulties to gather the outcome data in the missing cases. This weakens its external validity in regards to public health policies, as their conclusions are applicable on the condition of the families remaining in the intervention process. Loss of follow-up after allocation is a frequent limitation in public health program studies. In this research, women who dropped out differed from those who remained in the study in 2 of 16 baseline variables analyzed in the intervention group, and in 2 variables in the control group. These differences did not negatively affect the homogeneity of the study groups among the remaining cases.
Furthermore, although important intervention performance indicators were used regarding the last four months of the intervention, it would have been important to evaluate some other longitudinal indicators besides the number of meetings attended by mothers, for example, the number of home visits received since the beginning of intervention. It would have also been desirable to use other indicators related to fidelity of volunteers to the intervention protocols. Another limitation of the study consists that the assessors could not be blind in regards to the allocation of the participants, since it was practically impossible not to recognize when a participant of the study was also a participant of the intervention group. To reduce bias, interviewers had not taken part in the intervention and were trained using proper instruction manual for their questionnaires. Anthropometry assessment followed standardized procedures.
In this study, the provision of the intervention and the adhesion of the beneficiaries have not been controlled to ensure the ideal compliance of the protocols. However, the participation of the team that designed the intervention and its evaluation implies a higher degree of stimulation in the provision of the intervention than there would be in routine conditions in a larger scale. Thus, this could be classified as a public health “program efficacy study” in the typology of evaluation studies proposed by Victora et al.  taking into account the level of control exercised on the dose in which the intervention would reach the beneficiaries. We thus have to place the findings of this research in an intermediary level between that of a “regime efficacy study” (in which the impact of the intervention tends to be greater) and that of a “program effectiveness study” in routine conditions (in which the impact of the intervention tends to be smaller). In the pair-matched design, in which a control area with similar socioeconomic conditions was assigned to each geographic area (branch) of intervention, the analysis of about twenty variables, following a hypothetical causal model, showed great homogeneity in the baseline characteristics of both studied groups. The variables in which there were differences were taken into account to build statistical models that allowed the adjustment of the analysis of the intervention effect with regard to possible confounding effects. This brings the design used closer to the benefits that would have a cluster randomized design.