Maternal Risk Modeling in Critical Care-Development of a Multivariable Risk Prediction Model for Death and Prolonged Intensive Care.

OBJECTIVES: We aimed to develop and validate an accurate risk prediction model for both mortality and a combined outcome of mortality and morbidity for maternal admissions to critical care. DESIGN: We used data from a high-quality prospectively collected national database, supported with literature review and expert opinion. We tested univariable associations between each risk factor and outcome. We then developed two separate multivariable logistic regression models for the outcomes of acute hospital mortality and death or prolonged ICU length of stay. We validated two parsimonious risk prediction models specific for a maternal population. SETTING: The Intensive Care National Audit and Research Centre Case Mix Programme is the national clinical audit for adult critical care in England, Wales, and Northern Ireland. PATIENTS: All female admissions to adult general critical care units, for the period January 1, 2007-December 31, 2016, 16-50 years old, and admitted either while pregnant or within 42 days of delivery-a cohort of 15,480 women. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We aimed to develop and validate an accurate risk prediction model for both mortality and a combined outcome of mortality and morbidity for maternal admissions to critical care. For the primary outcome of acute hospital mortality, our parsimonious risk model consisting of eight variables had an area under the receiver operating characteristic of 0.96 (95% CI, 0.91-1.00); these variables are commonly available for all maternal admissions. For the secondary composite outcome of death or ICU length of stay greater than 48 hours, the risk model consisting of 17 variables had an area under the receiver operating characteristic of 0.80 (95% CI, 0.78-0.83). CONCLUSIONS: We developed risk prediction models specific to the maternal critical care population. The models compare favorably against general adult ICU risk prediction models in current use within this population.