AbstractObjectiveThe current study aims to demonstrate independent associations between social, educational, and health practice interventions as determinants of exclusive breastfeeding in an urban Ecuadorian population.DesignProspective survival analyses.SettingEcuadorian mother-child dyads in urban settings.ParticipantsWe followed-up 363 mother-baby dyads who were attended in health care centres in Portoviejo, province of Manabi, for up to 150 days.Main outcome measuresWe performed a survival analysis, by setting the time-to-abandonment of exclusive breastfeeding measured in days of life, periodically assessed by phone, as the primary outcome. Crude and adjusted mixed-effects Cox proportional hazards model were performed to estimate hazard ratios (HR) for each explanatory variable.ResultsThe median time of follow-up (P25 to P75) was 125 (121 to 130) days, and the incidence rate of abandonment of breastfeeding was 8.9 per 1000 persons-days in the whole sample. The three more significant protective determinants of exclusive breastfeeding were (i) sessions of prenatal breastfeeding education, with a reduced risk of abandoning the practice of exclusive breastfeeding of 30% (95%CI: 50% to 10%) per each extra session, (ii) self-perception of milk production, with a reduced risk of abandoning the practice of exclusive breastfeeding of 57% (95%CI: 50% to 10%) per each increase in the perceived quantity of milk production; and (iii) receiving early skin-to-skin contact represented a 90% risk reduction of interrupting breastfeeding (95%CI: 94% to 70%) vs. not receiving.ConclusionsPrenatal education on breastfeeding, self-perception of milk production, and early skin-to-skin contact appears to be strong determinants of exclusive breastfeeding in urban Ecuadorian mother-baby dyads; further, educational level of the mother, excreta management conditions are important determinants too.Article summaryStrengths and limitations of the studyOne of the most important strengths is that the study, as far as we know, is the first to address the topic in this specific region of the country. Also, this design allows the identification of patterns and elements; helping determine risk factors over time and cause and effect relationships. As we collected data in different intervals, we aimed to minimize recall bias and identify different changes at individual and group level. Our estimates are calculated by proper adjustment of potential confounders, reducing potential sources of confounding bias. This ensured a clear focus and increased validity.This study focused on urban populations as considering that there is a lack of research directed towards this group. However, one limitation can be the lack of representativeness of rural population, in which determinants would be different.32 As our research analysed hospital data, another limitation is related to the population in the area that did not go or decided to not deliver in a health centre; considering accessibility and use of the health services as relevant factors. Further research is needed. Also, other potential confounders were not considered, like nutritional factors or blood tests results. However, to our knowledge, there is no reason to consider that such variables could bias the estimates.This is a longitudinal prospective study developed under real conditions, in an urban Ecuadorian population, in the Coast region, Manabí province.A total number of 363 dyads were followed up for 150 days, having collected data at three separated times in three different phone surveys.Log-rank tests for equality of survivor functions for assessing differences of actual time in days to abandonment of exclusive breastfeeding were performed.A Cox proportional hazards models directed to evaluate the independent association between each explanatory variable and actual time in days to abandonment of exclusive breastfeeding was built, including the estimation of crude and adjusted hazard ratios per each explanatory outcome.Several secondary analyses to assess the sensitivity of our estimates with our assumptions regarding biases were conducted, as well as to test for model misspecifications.