A clustering model for monitoring preterm birth in Brazil to guide public health policies: A retrospective observational study

Background In low- and medium-income countries, the prevalence of preterm birth is increasing owing to persistent socioeconomic disparities, as well as increased medical indications and obstetrical procedures. This study analysed socioeconomic profiles, adequacy of antenatal care, and the occurrence of elective and spontaneous preterm births in healthcare units in Brazil in 2020. Methods This retrospective observational study used anonymised public data from the Ministry of Health and Instituto Brasileiro de Geografia e Estatística [IBGE], information from the Cadastro Nacional de Estabelecimentos de Saúde [CNES] and Sistema de Informações sobre Nascidos Vivos [SINASC] for the year 2020, and socioeconomic data from the 2010 Atlas Brazil. Herein, we created a unique database in which healthcare facilities were considered the unit of analysis. The analysed data included rates of elective and spontaneous preterm births, antenatal coverage, human development index [HDI], and Gini index of the counties, which were grouped using the K-mean clustering method. Kruskal–Wallis and Dunn’s tests were performed to assess differences between the clusters. Results Valid data from 2447 healthcare units were analysed and grouped into four clusters. In the Kruskal–Wallis test, the clusters showed significant statistical differences (p < 0.001) across all variables. Dunn's test revealed significant differences in spontaneous preterm birth between clusters 1 and 4 (p = 0.011) and in the Gini index between clusters 3 and 4 (p = 0.037). Clusters 1 and 2 showed no statistically significant difference (p = 0.757) in the number of antenatal visits, and clusters 1 and 4 did not differ in HDI (p = 0.198). However, comparisons of other variables revealed statistically significant differences (p < 0.001). Cluster 1 exhibited a higher rate of elective prematurity. Cluster 2 had adequate antenatal coverage, low socioeconomic inequality, high HDI, and a lower prevalence of spontaneous preterm births. Cluster 3 showed low antenatal coverage, low HDI, a higher prevalence of spontaneous prematurity, and lower elective prematurity. Characteristics of cluster 4 were comparable with those of cluster 2, except for a lower prenatal coverage. Conclusions Grouping healthcare facilities according to common characteristics can aid the development of targeted public health strategies aimed at reducing preterm birth rates in various contexts and help implement more successful and assertive actions.