ARTICLES 190 not intended to discuss all potentially relevant factors that influence breastfeeding behaviors. This study was conducted according to the PRISMA guidelines for systematic reviews and meta-analyses.23 Methods An initial search of the literature (PUBMED and CINAHL) was conducted to broadly identify characteristics that have been studied in relation to early breastfeeding (Search strings shown in Table I [available at www.jpeds.com]).We focused on factors identifiable during the perinatal period and attainable by history at birth or during the early lactation period. Postnatal changes in family, social, or work-related events were not included. The search was conducted from January to May 2015 with no lower date truncation and included over 2600 titles and abstracts and 183 full text papers. After data abstraction of the 183 papers (including study design, year and location of study, maternal and/or infant cohort details, predictor(s) of breastfeeding, breastfeeding outcome(s) assessed, and statistical techniques including model form and covariate adjustment), all independent variables predicting breastfeeding outcomes were grouped into distinct categories that were reviewed by practicing pediatricians and a Certified Lactation Counselor for clinical relevance and refined to contain the most homogeneous groupings. For all categories that had statistically significant predictors of breastfeeding outcomes in at least 5 publications, an evidence score from 1 to 9 was assigned for each factor-outcome relationship. This score was based on volume of supportive studies (up to 3 points), magnitude of association (up to 3 points), and relative consistency of associations (up to 3 points). Tables II and III (available at www.jpeds.com) include all initial factors considered as well as the associated evidence scores. After evaluation of the evidence scores in combination with clinical input, 6 factors were identified for full systematic quantitative evaluation: maternal smoking, mode of delivery (vaginal vs cesarean delivery), parity, infant-mother (hereafter called dyad) separation (including both early skin-toskin contact and rooming-in of infants in the hospital), maternal educational attainment, and breastfeeding education received by the mother or parents. Smoking and parity had lower evidence scores than other factors but were included based on perceived importance in clinical practice. Conversely, external/environmental characteristics and internal/ dyad characteristics of the mother as well as breastfeeding intent were not included despite relatively high evidence scores because the measures were too heterogeneous to be used in a meta-analysis. After careful consideration, SES was not included due to methodologic challenges. In US studies, the effects of the women, infants, and children (WIC) program on breastfeeding vs formula use are complex and the true relationship between breastfeeding and SES is difficult to tease out in meta-analysis; there is also lack of generalizability of WIC globally. Further, in accordance with recommended protocol for the conduct of systematic reviews and metaanalyses, such as the PRISMA guidelines, the measures of SES were deemed too heterogeneous to be combined in a meta-analysis (factors are described in Table IV [available at www.jpeds.com]). Maternal obesity initially did not emerge as a high-impact factor in our broad searches but after adding specific search terms related to maternal body size, a 2014 meta-analysis that reviewed 19 articles (6 obtained via literature search and 13 obtained via review of references or the snowball method)22 was identified; as few new articles on this topic were published since 2014, our review did not include an update of maternal obesity in relation to breastfeeding. Also, of note, factors including maternal age, race/ethnicity, gestational age, and pacifier use did not reach minimal evidence scores in terms of volume, strength, or consistency. Current Literature Review For the current systematic review, each of the 6 factors identified in the initial search was examined in a separate, comprehensive literature search using PUBMED limited to papers published between January 1, 2005 and March 12, 2016 (search terms shown in Table V). A 10-year search history was deemed a suitable representation of the current standard of care. Papers were included if they were written in English, included a study population in a developed nation, presented original data, and reported a relevant exposure (1 or more of the 6 factors of interest) and relevant outcome (either breastfeeding initiation, defined as in the birth hospitalization period or up to 1 week after birth, or breastfeeding continuation, defined as breastfeeding from 1 month of age up to the end of the first year). Relevant data from the papers selected for inclusion were abstracted by 1 epidemiologist and reviewed by a second. No data were sought from original investigators, and studies were not formally assessed for quality or potential sources of bias. Statistical Analyses For analysis, papers were included in 1 or both categories (initiation or continuation) for breastfeeding outcomes depending on available results. Exposures and outcomes were harmonized across studies so that all comparisons were in the same direction. When results for multiple time windows were provided, the earlier or