Effects of a Birth Hospital’s Neonatal Intensive Care Unit Level and Annual Volume of Very Low-Birth-Weight Infant Deliveries on Morbidity and Mortality

Importance  The annual volume of deliveries of very low-birth-weight (VLBW) infants has a greater effect on mortality risk than does neonatal intensive care unit (NICU) level. The differential effect of these hospital factors on morbidity among VLBW infants is uncertain.

Objective  To assess the independent effects of a birth hospital’s annual volume of VLBW infant deliveries and NICU level on the risk of several neonatal morbidities and morbidity-mortality composite outcomes that are predictive of future neurocognitive development.

Design, Setting, and Participants  Retrospective, population-based cohort study (performed in 2014) of all VLBW infants without severe congenital anomalies delivered in all hospitals in California, Missouri, and Pennsylvania between January 1, 1999, and December 31, 2009 (N = 72 431). Risk-adjusted odds ratios and risk-adjusted probabilities were determined by logistic regression.

Main Outcomes and Measures  The primary study outcomes were the individual composites of death or bronchopulmonary dysplasia, necrotizing enterocolitis, retinopathy of prematurity, and severe intraventricular hemorrhage.

Results  Among the 72 431 VLBW infants in the present study, birth at a hospital with 10 or less deliveries of VLBW infants per year was associated with the highest risk-adjusted probability of death (15.3% [95% CI, 14.4%-16.3%]), death or severe intraventricular hemorrhage (17.5% [95% CI, 16.5%-18.6%]), and death or necrotizing enterocolitis (19.3% [95% CI, 18.1%-20.4%]). These complications were also more common among infants born at hospitals with a level I or II NICU compared with infants delivered at hospitals with a level IIIB/C NICU. The risk-adjusted probability of death or retinopathy of prematurity was highest among infants born at hospitals with a level IIIB/C NICU and lowest among infants born at hospitals with a level IIIA NICU. When the effects of NICU level and annual volume of VLBW infant deliveries were evaluated simultaneously, the annual volume of deliveries was the stronger contributor to the risk of death, death or severe intraventricular hemorrhage, and death or necrotizing enterocolitis.

Conclusions and Relevance  The risk of death or severe intraventricular hemorrhage and death or necrotizing enterocolitis was lowest among infants born in hospitals that had both a high volume of VLBW infant deliveries and a high-level NICU. Antenatal transfer of high-risk pregnancies to these hospitals may reduce mortality and improve outcomes.

Geographically regionalized systems of perinatal care seek to improve outcomes for preterm infants by antenatally directing high-risk pregnancies to high-level perinatal centers.¹ In the late 1970s and early 1980s, these systems were widely implemented throughout the United States.^2,3 During the subsequent decade, neonatal mortality rates steadily decreased as the number of pregnant women who were at risk of a preterm delivery and antenatally transferred to hospitals with high-level neonatal intensive care units (NICUs) increased.^4,5 Driven largely by changes in the economics of health care, however, the number of midlevel NICUs and the proportion of infants delivered at these hospitals proliferated during the 1990s.⁶ Studies conducted since that time consistently demonstrate that mortality rates are higher among preterm infants born in hospitals with a low-level or midlevel NICU than among those born in hospitals with a high-level NICU.^7-10 In contrast, there is conflicting evidence in the literature on whether a birth hospital’s NICU level is associated with differences in nonmortality outcomes that are predictive of long-term neurodevelopment.^8,11-16 Moreover, most of these studies also did not report morbidity-mortality composite outcomes, which may bias the estimated morbidity risk owing to the higher mortality rates at hospitals with a lower-level NICU.

Recently, investigators compared the effects of a birth hospital’s NICU level with the effects of a birth hospital’s annual volume of deliveries of very low-birth-weight (VLBW) infants and found that the annual volume of deliveries better predicts risk-adjusted mortality among preterm infants.^17-21 Whether a birth hospital’s NICU level and annual volume of VLBW infant deliveries differentially affect neonatal morbidity risk is not well established and requires further study.²⁰ Therefore, the objectives of the present study are (1) to assess the association between the birth hospital and several neonatal complications when evaluated as morbidity-mortality composites and as morbidities only, and (2) to compare the strength of the association between a birth hospital’s annual volume of VLBW infant deliveries, its NICU level, and its neonatal outcomes.

Box Section Ref ID

A retrospective, population-based cohort of all infants with birth weights of 500 to 1499 g born in California, Missouri, and Pennsylvania between January 1, 1999, and December 31, 2009, was constructed. Each state’s department of health linked birth certificates and infant death certificates using name and date of birth and deidentified the records. More than 98% of these linked records were then matched to maternal and newborn hospital records using previously described methods.^20,22,23 More than 80% of the unmatched birth certificates were missing a hospital identifier, suggesting delivery at home or at a birthing center. The unmatched records had gestational age and racial/ethnic distributions similar to the matched records. Birth records were excluded if the birth weight was more than 5 SDs from the mean birth weight for the recorded gestational age because of potential recording errors in either variable.²⁴ Infants born weighing less than 500 g were excluded owing to potential variability by hospital with respect to resuscitation strategies for infants born at the limits of viability. Infants with severe congenital anomalies were also excluded (see eTable 1 in the Supplement for excluded diagnoses). The institutional review boards at the Children’s Hospital of Philadelphia and the departments of health in California, Missouri, and Pennsylvania approved our study.

The primary study outcomes were the individual morbidity-mortality composites of death or bronchopulmonary dysplasia (BPD), necrotizing enterocolitis (NEC), retinopathy of prematurity (ROP), and severe (grade 3 or 4) intraventricular hemorrhage (IVH). Death, chronological age at death, and the individual morbidities without mortality were evaluated as secondary outcomes. All maternal and infant sociodemographic and clinical data were obtained from infant birth and death certificates and the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis and procedure codes contained in the maternal and infant hospital discharge records (see eTable 1 in the Supplement for a list all codes used in the analysis). The death outcome included both (1) neonatal deaths that occurred prior to NICU discharge and (2) in-hospital fetal deaths with either a gestational age of 23 weeks or older or a birth weight of 500 g or more that met a previous definition of a potentially preventable fetal death during the birth hospitalization.²⁰ We included these fetal deaths to reduce bias against hospitals with high-level obstetrical services capable of rapid cesarean deliveries that result in live births of infants who may have otherwise died in utero.^19,20

Neonatal intensive care unit level was defined based on the 2004 American Academy of Pediatrics report on levels of neonatal care.²⁵ In brief, level I units provided basic well-newborn care. Level IIA units could resuscitate and stabilize preterm infants before transfer to a facility at which newborn intensive care is provided. Level IIB units had the added capability of providing mechanical ventilation for less than 24 hours or noninvasive positive airway pressure during the hospitalization. Data from level IIA and IIB units were combined. Level IIIA units provided comprehensive care for infants born more than 28 weeks’ gestation and weighing more than 1000 g, although respiratory support was limited to conventional mechanical ventilation. Level IIIB units provided comprehensive care for all extremely low-birth-weight infants and had advanced respiratory support including high-frequency ventilation and inhaled nitric oxide. Level IIIC units had the same capabilities as level IIIB units with the addition of extracorporeal membrane oxygenation. Data from level IIIB and IIIC units were combined. The classification of NICU level was retrospectively applied to deliveries prior to 2004 to maintain consistency over the study period. A birth hospital’s annual volume of VLBW infant deliveries was formulated into a categorical variable (≤10, 11-25, 26-50, or >50 deliveries per year) to facilitate comparison between our results and several previously published mortality studies.^18-20 The NICU level and the annual volume of VLBW infant deliveries were specified each year to allow for change in classification over time.

To compare the risk-adjusted association between a birth hospital’s annual volume of VLBW infant deliveries, NICU level, and the study outcomes, we developed a series of multivariable logistic regression models. The outcome of interest was defined as the dependent variable and attributed to the delivery hospital, regardless of where the outcome was diagnosed.^8,20 To assess the relationship between each hospital characteristic and the study outcomes, we first constructed separate risk-adjusted models for annual volume of VLBW infant deliveries and for NICU level. Next, we constructed models that included both annual volume of VLBW infant deliveries and NICU level as independent variables. This allowed us to compare the relative effect of one hospital characteristic on the risk of an outcome after adjustment for the effect of the other.

To select the appropriate risk-adjustment covariates, all clinical and sociodemographic variables associated with a primary study outcome at the P ≤ .20 level in bivariate testing were evaluated in a stepwise manner for inclusion in the final multivariable model. A P ≤ .05 in a multivariable model, evidence of confounding of the association between the primary exposure and the primary study outcome, or agreement among the investigators of the covariates’ importance resulted in inclusion in the final model. State of birth and year or birth were included to account for potential differences across geographic regions or change in outcome incidence over time. To maintain consistency across models, we included the same risk-adjustment covariates in each of the final models. Gestational age was not used owing to colinearity with birth weight and missing data on gestational age for 6% of the cohort. To account for a possible within-hospital outcome correlation, we used a robust variance estimator for cluster-correlated data based on the birth hospital’s identifier.^26,27 Risk-adjusted probabilities were determined from the logistic regression models using marginal standardization.²⁸ All analyses were performed with Stata version 13.1 (StataCorp).

The study cohort consisted of 72 431 infants. Infant and maternal medical and sociodemographic data by a hospital’s level of neonatal intensive care and annual volume of VLBW infant deliveries are summarized in Table 1 and Table 2, respectively. The majority of infants were born at hospitals with a level IIIB/C NICU and at hospitals with more than 50 VLBW infant deliveries per year. The proportion of VLBW infants born at a level IIIB/C NICU increased during the study period from 68.8% to 72.6%, while the proportion of VLBW infants born at a level I NICU decreased from 9.5% to 7.9% (test of trend P < .001). The number of infants delivered at hospitals with a large volume of VLBW infant deliveries and a high-level NICU also increased. In 1999, 48.3% of VLBW infants were born at level IIIB/C NICUs with more than 50 VLBW infant deliveries per year compared with 54.2% in 2009.

Mean birth weight and mean gestational age were similar between the different NICU levels and annual volumes of VLBW infant deliveries. A higher proportion of the mothers who delivered at hospitals with the lowest-level NICUs and lowest-volumes of VLBW infant deliveries were white and had either public health insurance or no health insurance. Maternal chronic comorbid illnesses and most complications of pregnancy were more common among women who delivered at hospitals with a higher-level NICU and a higher volume of deliveries. Abnormal placentation and precipitous labor occurred more frequently among women who delivered at hospitals with a lower-level NICU and a lower volume of deliveries.

Delivery at lower-level NICUs and lower-volume hospitals was associated with earlier mortality (Figure 1). Among VLBW infants born at hospitals with 10 or less VLBW infant deliveries per year, 68% percent of deaths occurred within the first 24 hours of life. In contrast, at hospitals with more than 50 VLBW infant deliveries per year, less than 50% of deaths occurred within the first 24 hours. Similarly, among VLBW infants born at hospitals with a level I NICU, 66% of deaths occurred in the first 24 hours of life compared with 48% for hospitals with a level IIIB/C NICU. Most deaths (84%) occurred within the first 28 days of life.

Risk-adjusted probabilities of the composite morbidity-mortality outcomes are shown by the birth hospital’s NICU level and annual volume of VLBW infant deliveries in Figure 2 (see eTables 2 and 3 in the Supplement for the risk-adjusted odds ratios). Birth at a hospital with 10 or less VLBW infant deliveries per year was associated with the highest risk-adjusted probability of death (15.3% [95% CI, 14.4%-16.3%]), death or severe IVH (17.5% [95% CI, 16.5%-18.6%]), and death or NEC (19.3% [95% CI, 18.1%-20.4%]). The adjusted probability of these complications was also higher among infants born at hospitals with a level I or II NICU than among infants born at hospitals with a level IIIB/C NICU. Infants born at hospitals with a level II but not a level I NICU were also at increased risk of death or BPD. The risk-adjusted probability of death or ROP was highest among infants born at hospitals with a level IIIB/C NICU (32.6% [95% CI, 30.7%-33.7%]) and lowest among infants born at hospitals with a level IIIA NICU (25.5% [95% CI, 22.4%-28.7%]). Annual volume of VLBW infant deliveries was not associated with significant differences in the risk of death or BPD or the risk of death or ROP.

In contrast to the observed association between birth hospital and either the morbidity-mortality composite outcomes or mortality alone, the risk-adjusted probabilities for BPD, NEC, and ROP were lowest among infants born at hospitals with a level I NICU or at hospitals with 10 or less VLBW infant deliveries per year (Figure 2). The risk-adjusted probability of severe IVH was similar across the different NICU levels and annual volumes of VLBW infant deliveries.

We developed risk-adjustment models that included variables for both the birth hospital’s NICU level and the birth hospital’s annual volume of VLBW infant deliveries to compare the strength of the association between the 2 hospital characteristics and the primary study outcomes. With this approach, lower volume of deliveries emerged as a stronger risk factor for death, death or severe IVH, and death or NEC (Table 3). The risk-adjusted odds ratios for death or severe IVH ranged from 1.36 (95% CI, 1.14-1.63) at hospitals with 10 or less VLBW infant deliveries per year to 1.15 (95%, CI 1.04-1.28) at hospitals with 26 to 50 deliveries per year, compared with preterm infants born at hospitals with more than 50 VLBW infant deliveries per year. After adjusting for NICU level, the odds of death or NEC were only increased for infants delivered at hospitals with 10 or less VLBW infant deliveries per year (odds ratio, 1.33 [95%, CI 1.12-1.57]). Level of NICU and annual volume of VLBW infant deliveries were associated with opposing effects on death or BPD.

We conducted the present study to address the ongoing uncertainty with respect to the association between delivery hospital and neonatal morbidities that are strongly predictive of neurodevelopment.^28,29 We also compared the risk-adjusted effects of a birth hospital’s NICU level and annual volume of VLBW infant deliveries to assess whether either hospital factor was the primary driver of any difference in outcomes. To account for the increased mortality risk among preterm infants delivered in hospitals with low-level or midlevel NICUs, we evaluated morbidity-mortality composites as primary outcomes.^17-20

When we examined the effects of a birth hospital’s NICU level and annual volume of VLBW infant deliveries individually, higher-level NICU and higher volume of VLBW infant deliveries were both associated with a lower risk-adjusted probability of death, death or severe IVH, and death or NEC. Among hospitals with a level III NICU specifically, hospitals with a level IIIA NICU had the lowest risk-adjusted probability of all morbidity-mortality composite outcomes except for death or BPD. Chung et al¹⁹ reported a similar association for mortality and suggested that this may reflect the ability of hospitals with level IIIA NICUs to triage and appropriately refer high-risk obstetrical cases. The largest benefit associated with delivery at a hospital with more than 50 VLBW infant deliveries per year was found for death or severe IVH. The risk-adjusted odds of this adverse outcome were 16% to 55% higher among infants born at hospitals with fewer than 50 VLBW infant deliveries per year.

In contrast to the morbidity-mortality composite outcomes, the risk-adjusted probability of each morbid condition, except severe IVH, was lowest among infants born at hospitals with a level I NICU or at hospitals with 10 or less VLBW infant deliveries per year. This apparent “protective” effect is likely explained by the higher and earlier mortality rates that we observed at these hospitals because most deaths occurred before the morbidities could develop. Watson et al¹³ reported a similar problem with competing risks in a cohort of extremely preterm infants born in the England. They found that delivery in a tertiary level hospital or one in the top volume quartile was associated with decreased risk of death but increased risk of BPD and treatment for ROP.¹³ This likely survival bias highlights the pitfalls that can arise when death strongly competes against morbidities diagnosed later in the neonatal period. Morbidity-mortality composite outcomes can help account for this bias, but they should be interpreted with caution when the effect size of the individual components is large and in the opposite direction.^29,30 We graphically presented risk-adjusted rates of all study outcomes to facilitate easy evaluation of how censoring by death may affect the observed outcome rates.

Finally, we assessed the effects of a birth hospital’s NICU level and annual volume of VLBW infant deliveries simultaneously. In this analysis, lower volume of deliveries emerged as the stronger risk factor for death, death or severe IVH, and death or NEC. Among the infants born in hospitals with less than 50 deliveries per year, the odds of death or severe IVH, in particular, were 15% to 36% higher after adjusting for NICU level. These findings have important implications for the regionalization of perinatal care. The designation of NICU level is based on the availability of medical technology and subspecialty services and does not necessarily reflect the volume of deliveries. Although more than 70% of infants in this cohort were born at hospitals with level IIIB/C NICUs, more than half of these hospitals had fewer than 50 VLBW infant deliveries per year. Our findings suggest that regionalized systems that transfer pregnant women at risk of a preterm delivery to hospitals with high-level NICUs and large volumes of VLBW infant deliveries are likely to have the best outcomes. Fortunately, we found that the proportion of preterm infants delivered at hospitals with a level IIIB/C NICU and more than 50 VLBW infant deliveries per year increased during the 11-year study period. The breakdown of many regionalized perinatal systems and the health consequences associated with this trend have garnered considerable attention, especially in states where this trend has been viewed as most severe.^7,31-34 Our results indicate that the trend toward deregionalization might now be reversing.

Our study has several strengths. One is the large, population-based data set inclusive of infants from several different geographic regions. Many of the previous studies that evaluated the effect of birth hospital on neonatal morbidities used data limited to small geographic regions or only included infants transferred from lower-level units to academic institutions.^14,15,35-37 Our data set includes nearly all in-hospital deliveries in 3 states and contains infants cared for in both high-level community and academic hospitals. Moreover, California, Missouri, and Pennsylvania have a broad mix of urban and rural populations of varying ethnicities and sociodemographic backgrounds that are generally representative of the full US population. Another strength is the robustness of our risk adjustment. We observed, as others have reported,³⁶ that maternal characteristics and comorbidities of pregnancy vary based on the NICU level and volume of VLBW infant deliveries of the birth hospital. Because these differences may affect baseline risk for adverse outcomes, failure to account for them may bias the estimated association between birth hospital and outcome.

We also acknowledge several limitations. Maternal and infant comorbidities were classified using ICD-9-CM codes abstracted from hospital discharge records. This prevented a time-to-event analysis that may have allowed us to account for censoring of morbidities by earlier deaths at hospitals with a low-level NICU and a low volume of VLBW infant deliveries. There may be some heterogeneity among hospitals regarding how they code. The reliability of ICD-9-CM codes for several neonatal morbidities has been assessed, and, in general, these codes show high specificity and variable sensitivity.^38-41 If present, we would expect this information bias to result in an underestimation of morbidity rates. There are also no codes for administration of important medications such as antenatal corticosteroids, caffeine, or surfactant. Finally, we were only able to stratify risk based on the factors available in our data set. There may be additional unknown or unmeasured factors that resulted in residual confounding.

Our results demonstrate that the risk of mortality and some prognostically important morbidity-mortality composite outcomes is lower among VLBW infants delivered in hospitals with a high-level NICU and a large volume of VLBW infant deliveries. The higher and earlier mortality rates among preterm infants born at hospitals with a lower-level NICU and a lower volume of deliveries limited our ability to assess the full effect of birth hospital on morbidities diagnosed later in the neonatal period. At present, however, the available evidence indicates that neonatal outcomes are improved when pregnant women at risk of a preterm delivery are antenatally transferred to hospitals with high-level NICUs and large volumes of VLBW infant deliveries.

Accepted for Publication: June 8, 2015.

Corresponding Author: Erik A. Jensen, MD, Division of Neonatology, Department of Pediatrics, Children’s Hospital of Philadelphia, 2 Main, 34th and Civic Center Boulevard, Philadelphia, PA 19104 (jensene@email.chop.edu).

Published Online: August 3, 2015. doi:10.1001/jamapediatrics.2015.1906.

Author Contributions: Dr Jensen had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Both authors.

Acquisition, analysis, or interpretation of data: Both authors.

Drafting of the manuscript: Both authors.

Critical revision of the manuscript for important intellectual content: Jensen.

Statistical analysis: Both authors.

Obtained funding: Lorch.

Administrative, technical, or material support: Jensen.

Study supervision: Lorch.

Conflict of Interest Disclosures: None reported.

Funding/Support: This study was funded by the Agency for Healthcare Research and Quality (grant R01 HS 015696).

Role of the Funder/Sponsor: The Agency for Healthcare Research and Quality had no role in the design and conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.