Neonatal hypoxic-ischemic encephalopathy (HIE) is a neurologic syndrome that results from reduced flow of oxygenated blood to the fetal or newborn brain. HIE occurs in 1-3 per 1,000 term births and may cause death or neurologic disabilities such as cerebral palsy. Electronic fetal monitoring (EFM) was developed in the 1970's to assess the adequacy of fetal oxygenation as a strategy to prevent HIE, and is now standard of care. Yet clinical trials report that EFM usage has not reduced the rate of CP, perinatal death or HIE, but is associated with a dramatic increase in cesarean deliveries. The currently used 3 Category fetal heart rate (FHR) classification system, based on simple rules designed to be easy to apply at the bedside, has some utility in predicting HIE. However, Category II FHR patterns that make up the vast majority of tracings are poorly predictive of HIE and confer “indeterminate” risk. Category III patterns are also of limited use in predicting HIE due to low sensitivity. There is an urgent need to develop better objective methods to assess EFM that would identify more fetuses at risk of HIE in time for corrective actions. Uterine tachysystole, or excessive frequency of uterine contractions, has been implicated as a preventable cause of HIE; yet studies report conflicting results. EFM research has been limited by an inability to access and manually analyze the large datasets needed to study HIE. We now have the ability to analyze digital EFM signals using automated methods to measure standard FHR patterns as well as to discover novel aspects of the tracing that may not be readily detectable by a clinician at the bedside. We hypothesize that modern signal processing and machine learning techniques can create highly predictive models of HIE by analyzing established and novel features of EFM tracings, in combination with demographic and pertinent clinical information from the mother and fetus. We propose a population-based retrospective cohort study of 350,000 infants born at ≥ 36 weeks gestation at Kaiser Permanente Northern California in 2010-19. Our specific aims are: 1) To create the MAESTRA Cohort dataset that links EFM recordings to HIE and neonatal acidosis among 350,000 infants born at ≥ 36 weeks gestation in 2010-19 at Kaiser Permanente Northern CA; 2) Using modern signal processing and machine learning techniques, to extract established and novel FHR and uterine contractility features from the EFM recordings, and to determine which of these features are most predictive of HIE and acidosis when combined with maternal and fetal clinical data; and 3) To perform external validation by applying the final predictive models to a historical dataset. We anticipate that machine learning techniques incorporating novel FHR and uterine contractility patterns over time, as well as pre- and perinatal clinical characteristics, will improve the predictive value of the EFM data that are already being collected as part of routine care. Our resu...