389 - Developing a Resiliency Score for Survival without Major Morbidity in Preterm Infants < 32 Weeks of Gestation with External Validation

Poster Number: 389
Publication Number: 389.134

Kelli K. Ryckman, University of Iowa College of Public Health, Iowa City, IA, United States; Martina Steurer, UCSF, San Francisco, CA, United States; Rebecca Baer, University of California, San Francisco, School of Medicine, La Jolla, CA, United States; Jean Costello, UCSF, San Francisco, GA, United States; Scott P. Oltman, University of California San Francisco, San Francisco, CA, United States; Charles McCulloch, University of California, San Francisco, School of Medicine, San Francisco, CA, United States; Laura Jelliffe-Pawlowski, University of California San Francisco, San Francisco, CA, United States; Elizabeth E. Rogers, University of California, San Francisco, School of Medicine, San Francisco, CA, United States

Background: Risk assessment and outcome prediction in extremely preterm neonates is important for multiple reasons. Understanding the range of outcomes for an infant can inform parental counseling and medical decision making, both antenatally when making shared decisions around the provision of intensive care at delivery as well as postnatally if complications develop. Another important goal of prediction models includes their use in benchmarking and comparing performance across medical centers, to guide continuous quality improvement in care practices. Finally, prediction models can help ensure that both care practices and outcomes are observed to be equitable to reduce the impact of discrimination and structural racism.

Objective: The objective of this study is to develop and validate a resiliency score to predict survival and survival without neonatal morbidity in preterm neonates < 32 weeks of gestation using machine learning.

Design/Methods: Models using maternal, perinatal, and neonatal variables were developed using the least absolute shrinkage and selection operator (LASSO) method in a population based Californian administrative dataset. Outcomes were survival and survival without severe neonatal morbidity. The resiliency score was calculated by summing the coefficients based on the value of the respective variables. Discrimination was assessed in the derivation dataset and in an external dataset from a single tertiary care center in Iowa. Performance of the resiliency score was assessed in different sociodemographic subpopulations.

Results: Discrimination of the model in the internal validation dataset was excellent with a c-statistic of 0.895 (95% CI 0.882-0.908) for survival and 0.867 (95% CI 0.857-0.877) for survival without severe neonatal morbidity, respectively. Discrimination remained high in the external validation dataset with a c-statistic of 0.817 (95% CI 0.741-0.893) for survival and 0.804 (95% CI 0.770-0.837) for survival without severe neonatal morbidity, respectively. The model performed well across racial/ethnic and sociodemographic subpopulations.Conclusion(s): Our model using a resiliency score successfully predicts survival and survival without major morbidity in preterm babies born at < 32 weeks. In addition, it works well across different epidemiological settings and racial/ethnic and sociodemographic subpopulations. Such a score can be used and be helpful in clinical settings and in antenatal and postnatal parent counseling with a focus on protection rather than risk.