To develop and validate a support tool for healthcare providers, enabling them to make precise and critical decisions regarding intensive care unit (ICU) admissions for high-risk pregnant women, thus enhancing maternal outcomes. This retrospective study involves secondary data analysis of information gathered from 9550 pregnant women, who had severe maternal morbidity (any unexpected complication during labor and delivery that leads to substantial short-term or long-term health issues for the mother), collected between 2009 and 2010 from the Brazilian Network for Surveillance of Severe Maternal Morbidity, encompassing 27 obstetric reference centers in Brazil. Machine-learning models, including decision trees, Random Forest, Gradient Boosting Machine (GBM), and Extreme Gradient Boosting (XGBoost), were employed to create a risk prediction tool for ICU admission. Subsequently, sensitivity analysis was conducted to compare the accuracy, predictive power, sensitivity, and specificity of these models, with differences analyzed using the Wilcoxon test. The XGBoost algorithm demonstrated superior efficiency, achieving an accuracy rate of 85%, sensitivity of 42%, specificity of 97%, and an area under the receiver operating characteristic curve of 86.7%. Notably, the estimated prevalence of ICU utilization by the model (11.6%) differed from the prevalence of ICU use from the study (21.52%). The developed risk engine yielded positive results, emphasizing the need to optimize intensive care bed utilization and objectively identify high-risk pregnant women requiring these services. This approach promises to enhance the effective and efficient management of pregnant women, particularly in resource-constrained regions worldwide. By streamlining ICU admissions for high-risk cases, healthcare providers can better allocate critical resources, ultimately contributing to improved maternal health outcomes.
Read full abstract