Comparative Efficiency of Inpatient Care Settings for Pre-term Birth: An Instrumental Variables Analysis

BACKGROUND: Continued growth in the prevalence of cost of pre-term births underscores the need for improved management of cases across available care settings. Acutely premature and very low birthweight (VLBW) infants experience improved health outcomes when delivered in inpatient settings equipped with neonatal intensive care units (NICUs) and neonatologist coverage. The relative costs of pre-term delivery at NICU vs. non-NICU hospitals, however is less clear due to differences in case mix and survival rates across facilities. A better understanding of the economic impact of preterm delivery in NICU-equipped hospitals is needed to improve the quality and efficiency of care for preterm infants.

METHODOLOGY: We assembled birth records for all pre-term infants born in Arkansas during 2001-04, and linked them with corresponding hospital discharge records. Multivariate models were used estimate differences in LOS and costs between hospital settings while adjusting for observable covariates including birth weight, gestational age, congenital anomalies, maternal demographics, and maternal risks. Because complex and high-risk cases are more likely to be referred to NICU-equipped hospitals, we use instrumental variables (IV) methods to account for the nonrandom selection of cases across hospital settings. We use discrete factor approximation methods to control for censoring of LOS and cost measures due to in-hospital neonatal mortality, allowing this censoring process to be endogenously influenced by the hospital setting. Our analysis distinguishes between three types of hospital settings: (1) academic hospitals with NICUs; (2) community hospitals with NICUs; and (3) community hospitals without NICUs.

FINDINGS: Estimates from models without adjustment for endogenous selection and censoring suggest that LOS was 25.2% higher for deliveries in academic NICU settings and 32.1% higher in community NICU settings compared with non-NICU hospitals (p<0.01). When the IV selection and censoring adjustments were used, LOS was 9% lower in academic NICU hospitals and 17% higher in community NICU settings compared with non-NICU settings (p<0.05).

CONCLUSIONS Adjusted LOS differences imply that inpatient delivery costs could be reduced by 12-15% for each VLBW delivery shifted from non-NICU to academic NICU settings, and by 28-33% for each delivery shifted from community NICU to academic NICU settings. Community NICU hospitals appear to be the costliest settings for VLBW pre-term deliveries, due primarily to higher adjusted LOS. Policy strategies to increase the proportion of VLBW deliveries occurring in NICU-equipped hospitals may realize efficiencies by prioritizing academic perinatal centers, particularly for the most acutely pre-term deliveries.