Causes and performing early identifying the preventable risk stratification of pregnant women are instrumental to develop strategies to prevent and reduce preterm birth (PTB). The study aims to combine different national level data sources to understand the main predictors of PTB and develop a machine-learning–based predictive model to conduct automated risk stratification at the point of care level, integrated with advanced data visualization for clinical decision support.
First, the database linkage was established, incorporating social determinants, prenatal and childbirth data, and individual information related to the risk of prematurity, which included data from various years. Next, data validation techniques were applied to ensure accuracy, addressing issues such as outliers, missing data, and multicollinearity, among others. In the third phase, predictive modeling using machine learning was conducted, where six different algorithms were tested. Finally, a decision support application was developed that integrates the machine learning algorithm with data visualization and geographic information systems, resulting in the creation of a preterm birth risk calculator named PRECOCE.
The calculator is a decision support tool designed for the clinical risk stratification of preterm birth. It assists healthcare providers in managing at-risk pregnant women to help prevent unfavorable outcomes. This tool is a web application that includes geolocation capabilities for each registered pregnant woman, enabling effective spatial risk stratification of prematurity.
The solution is groundbreaking in its use of population data, integrating information from thousands of individuals at various levels of aggregation, including municipalities, health units, and primary care teams, in a developing country of significant size. The algorithm can predict the week of delivery with a two-week margin of error, which is an improvement over previous attempts reported in the literature. The variables used for prediction are easy to collect, making this a low-cost technology with high scalability potential. Additionally, a key innovation is the use of geographic territory—specifically, the residence locations of pregnant women—as a data source for predictions, along with the integration of this tool into a geolocation platform.
Every year, more than 15 million children are born prematurely, defined as being born before 37 weeks of gestation. Preterm birth is the leading cause of death among children under one year old, with over 1.1 million preterm newborns still dying, particularly in low- and middle-income countries (LMICs). Along with India, China, and Nigeria, Brazil has one of the highest numbers of preterm births. Many of these preterm births could be prevented with proper prenatal care that is tailored to the specific risks of each pregnancy. However, stratifying risk for preterm birth is challenging, as existing predictive models often have significant margins of error or rely on predictors that are difficult to measure. This project aimed to tackle these issues by developing a prediction algorithm for risk stratification of preterm birth, using data visualization and analysis to support informed decision-making in public health policies.
This solution, which is based on artificial intelligence and implemented through a brief questionnaire administered to pregnant women—ideally during a home visit by a Community Health Agent—can be integrated into the primary care information system. This integration will support SUS workers in identifying women at risk for preterm birth. By identifying these women early in their pregnancies, they can be referred to specialized centers where prenatal care can be adjusted according to their specific risks. This approach aims to increase the number of intrauterine days for the baby and ultimately reduce SUS expenses related to hospitalizations, procedures, tests, and medications. Additionally, it could help decrease infant mortality rates.
The project team plans to validate the algorithm using external databases, including the BRISA cohorts from the Northeast and Southeast regions of Brazil, as well as the Moshi database from Africa. They also intend to make the solution available to primary care health professionals, allowing them to input data during prenatal care. This feedback will enable the algorithm to learn and improve, aiming to reduce the margin of error to less than seven days.
Furthermore, the team plans to implement the calculator in a non-randomized community pilot test in a municipality in northeastern Brazil. The objective is to analyze its potential effects on reducing preterm births, improving risk stratification, and ultimately decreasing infant mortality.
