2013 and December 15, 2013. Deliveries were included if both admission and 12 hour postpartum blood count (12hCBC) were available. Medians and distributions were calculated. Chi-Square, Mann-Whitney, and logistic regression were used. RESULTS: 592 of 626 (95%) women were analyzed. Twenty (3.4%) women were transfused. The overall group 5th percentile and median aHbg were 9.3 g/dL and 11.7 g/dL respectively. The 5th percentile (7.7 vs. 9.5 g/dL, p , 0.001) and median aHgb (9.7 vs. 11.7 g/dL, p , 0.001) were both significantly lower among those who were vs. were not trans- fused. Admission Hgb cut points are associated with different test char- acteristics in predicting transfusion. Progressively lower aHgb led to incrementally improved specificity with only minimally decreased NPV. CONCLUSION: Our data suggests that, despite poor sensitivity, an aHgb of 9.5 was associated with a 98% NPV and a , 5% false positive rate, making it an excellent antenatal tool to screen women at risk for receiving postpartum transfusion. Achieving this aHgb may serve as an antepartum target in anemic patients to potentially decrease transfusion risk and improve maternal outcomes. Financial Disclosure: Sindhu Srinivas disclosed the following—Penn: Expert witness. The other authors did not report any potential conflicts of interest. Predicting Perinatal Outcomes With an Obstructive Sleep Apnea Screening Tool [13N] Luis A. Bracero, MD West Virginia University-Charleston, Charleston, WV Samantha Chaffin, MD, Byron Calhoun, MD, FACS, FASAM, MBA, Dara Seybold, MAA, Peter Power, MD, and Stephen Bush, MD INTRODUCTION: The objective was to determine if a screening tool for obstructive sleep apnea (OSA) can be used to predict adverse perinatal outcomes. METHODS: Prospective observational study of patients receiving prenatal care (June 1 2013-May 31 2014) at a tertiary care center clinic. Patients were universally screened for OSA with the STOP question- naire, a concise tool with four questions related to snoring, tiredness during daytime, observed apnea, and high blood pressure. It has been validated in surgery patients. Positive and negative screens and several confounding variables were included in a backwards logistic regression model to predict the following adverse perinatal outcomes: cesarean delivery, preeclampsia, preterm delivery (PTD) (,37 weeks), low birth weight (LBW) (,2500 g), Intra Uterine Growth Restriction, and Neo- natal Intensive Care Unit (NICU) admission. RESULTS: Of the 680 patients who received prenatal care, 442 (65%) had singleton births and were included in analysis. Patient body mass indexes ranged from 12.8-61.4 with mean of 29.7. This study population had high tobacco (241; 54.5%) and illicit drug use (155; 35.1%). Positive STOP screens (64;14.5%) were associated with two outcomes: PTD and NICU admission. For PTD, history of pre-term labor was the strongest predictor, followed by STOP and nulliparity with odds ratios (OR) of 4.2 (95% CI 2.0-8.8; p , 0.001), 2.8 (95% CI 1.4-5.8; p50.004), and 2.3 (95% CI 1.2-4.4; p50.013). For NICU admission, a positive STOP was the only significant predictor with 2.5 OR (95% CI 1.1-5.7; p50.036). CONCLUSION: The STOP screening questionnaire may be useful in predicting PTD and NICU admission. Financial Disclosure: The authors did not report any potential conflicts of interest. Predicting Term Induction to Delivery Intervals Utilizing Machine Learning [14N] Corey Clifford, DO TriHealth, Cincinnati, OH Devin Namaky, MD, and Michael Holbert, MD INTRODUCTION: Machine learning methodologies, such as artifi- cial neural networks, are increasingly being used to improve clinical decision making. They can be used to discover complex relationships in clinical data. The goal of this study was to evaluate their ability to predict term induction to delivery intervals. METHODS: A retrospective cohort of 439 term inductions that resulted in an unassisted vaginal delivery was used to develop a predictive artificial neural network. This model utilized factors known at the time of induction, such as maternal demographics, pregnancy history, cervical examination, and current pregnancy complications. A separate cohort of 233 patients was then used to evaluate the efficiency of the model. RESULTS: On the cohort of 233 patients, the model exhibited a predictive odds ratio for delivery within 12 hours of 5.76 (95% CI 3.17-10.43, P , .001). Furthermore, it exhibited a sensitivity, specific- ity, positive predictive value, and negative predictive value of 78%, 62%, 80%, and 59% respectively. The five elements that contributed the greatest were Bishop score-position, maternal race, parity, presence of diabetes in the current pregnancy, and presence of hypertensive disease in the current pregnancy. Of the remaining factors; gestational age, cervical dilation, gravidity, and maternal BMI demonstrated mod- erate weighting. The remainder were less important. CONCLUSION: A machine learning approach is moderately success- ful in predicting induction to delivery intervals in term patients. This technique could be used to guide the timing of the initiation of induction in these patients to maximize staffing availability and improve outcomes. Financial Disclosure: The authors did not report any potential conflicts of interest. Predictors of a Positive Illicit Drugs Screens for Pregnant Women in Colorado [15N] Shannon L. Son, MD University of Colorado Hospitals, Aurora, CO Kent Heyborne, MD, Jeanelle Sheeder, PhD, and Maryam Guiahi, MD, MSC INTRODUCTION: There are no clear guidelines to determine when illicit drug screening in pregnant women is warranted. We investigated which testing indications reliably predict a positive screen. METHODS: We performed a historical cohort study of all pregnant women who underwent drug screening on labor and delivery in 2014 at a safety-net hospital in Denver, Colorado. Testing was performed at provider discretion and with patient consent. Illicit drugs were included in the urine toxicology screen; tetrahydrocannabinol was not. Chart abstraction included testing indications, demographic, reproductive, social, and substance use characteristics. We performed descriptive frequencies and used Fisher’s exact tests to compare testing results. RESULTS: In 2014, 166 urine toxicology tests were performed; 14 (8.4%) were positive. Women with positive tests were more likely to have had a prior STI (83.3% vs 39.7%, p.0.01), previous child protective serv- ices case (35.7% vs 8.6% p50.009), child removed from custody (60.0% vs 7.3%, p , 0.001), prior drug use (92.9% vs 34.2%, p , 0.001) or treatment (64.3% vs 11.2%, p , 0.001), self-reported current drug use (71.4% vs. 27.6%, p , 0.01), current tobacco use (71.4% vs 19.7%, p , 0.001), less than 5 prenatal visits (58.3% vs 28.0%, p50.05), or outborn delivery (28.6% vs. 0.7%, p , 001). No obstetrical indications were associated. A composite variable including history of child protective services case, child removed from custody, history of drug use or treatment, and/or self- reported drug use had a PPV5100% and NPV565%. CONCLUSION: Using a composite variable based on historical factors to determine performance of drug screening may be a strong predictor of positive tests and reduce over-testing. Financial Disclosure: The authors did not report any potential conflicts of interest. Predictors of Quality of Care Amongst Gestational Diabetic Patients in Community Hospital [16N] Rukeme Ake-Uzoigwe, MD, MPH Icahn School of Medicine, Queens Hospital Center Program, New York, NY Ayodeji Sanusi, MD, MPH, Carolyn Salafia, MD, Dongping Zhang, MD, Kolawole Akinnawonu, MD, and Aleksandr Fuks, MD INTRODUCTION: The objective of our study was to determine factors that affect the quality of care received by a gestational diabetic patient. METHODS: A retrospective cohort study of gestational diabetic patients’ who delivered from January 2013 to December 2015. Patients 146S MONDAY POSTERS OBSTETRICS & GYNECOLOGY Copyright ª by The American College of Obstetricians and Gynecologists. Published by Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.