Variation in Length of Stay by Level of Neonatal Care Among Moderate and Late Preterm Infants Free

We performed a retrospective cohort study using data from the 2019 Kid’s Inpatient Database (KID).²³  The KID was developed as part of the Healthcare Cost and Utilization Project sponsored by the Agency for Healthcare Research and Quality. This database includes deidentified administrative data on pediatric discharges from 3998 hospitals from 48 states plus the District of Columbia and includes short-term, non-Federal hospitals. The data includes a 10% sample of normal newborns and an 80% sample of complicated newborns and other pediatric discharges. However, weights are provided to generate national estimates.²³  This study was exempt from the institutional review board at Children’s National Hospital.

Our study population was moderate and late preterm infants. Newborns were included in the study if they had an International Classification of Diseases 10th Revision (ICD-10) diagnosis code for “Liveborn” representing an in-hospital birth (inborn) as well as an ICD-10 diagnosis code for 32 to 36 weeks’ GA (Supplemental Table 4). Exclusions were made to remove newborns with significant comorbidities and those who may require higher-level specialized care. Specifically, newborns with complex chronic conditions (CCCs)²⁴  and/or very low birth weight (VLBW, defined as <1500 g) were excluded. Per Feudtner et al in 2014 who described the Pediatric Complex Chronic Conditions, CCCs are “any medical condition that can be reasonably expected to last at least 12 months (unless death intervenes) and to involve either several different organ systems or 1 organ system severely enough to require specialty pediatric care and probably some period of hospitalization in a tertiary care center.”²⁴  Using this definition, common diagnoses in moderate preterm infants including respiratory distress syndrome would not be excluded, but an infant with critical congenital heart disease would meet exclusion criteria. Infants with VLBW have increased risk of prolonged LOS compared with those with weights greater than 1500 g.²⁵  Dispositions other than discharge to home (ie, death and transfer) were also excluded. (Supplemental Fig 1).

The primary exposure was the hospital level of neonatal care, specifically level I, II, III, or IV. Since levels of neonatal care are not reported directly within the database, we developed criteria using the AAP definitions for levels of neonatal care.⁶  We classified each hospital within the database using their entire neonatal population (ie, all discharges ≤28 days old at admission) according to the highest level of neonatal care the hospital provided. As such, even if babies were admitted to the level I or II unit, if the hospital was classified as level III, those babies would be included in the level III data for this study. Level IV included hospitals that performed surgeries with cardiac bypass.⁶  Level III included hospitals that provide mechanical ventilation for greater than 24 hours, continuous positive airway pressure (CPAP) for greater than 96 hours, nonbypass surgery, and/or cared for babies born 23 to 31 weeks GA or <1500 g.⁶  To meet classification requirements for a level III, hospitals had to have 2 or more patients (whose disposition was not transferred or death) meet at least 2 of these characteristics. Level II included hospitals that had 2 or more patients (whose disposition was not transferred or death) with at least 1 of the following: received mechanical ventilation less than 24 hours, CPAP less than 24 hours, or were 32 to 34 weeks GA.⁶  Level I included hospitals that did not fit any of the above criteria (Supplemental Table 5).

Our primary outcome was LOS measured in days.

Demographic variables available in the KID include race and ethnicity, sex, health insurance payor, and census region. Race and ethnicity is a combined variable in the KID database, and if the data source supplies information for both race and ethnicity, ethnicity takes precedence over race in the database.²⁶  Clinical variables were derived from ICD-10 diagnoses codes and included delivery type and multiple gestation. We grouped ICD-10 diagnoses and procedures using the Agency for Healthcare Research and Quality’s Clinical Classification Software (CCS).²⁷ 

Demographic and clinical characteristics were compared across weeks of GA using χ-square tests. LOS was compared across neonatal levels of care within each GA using Kruskal–Wallis tests because of nonnormal data. To account for clinical severity differences when comparing LOS by GA across levels of neonatal care, we needed to determine which variables should be adjusted for in each model (Appendix 1). To accomplish this, we used Classification and Regression Trees (CART) for each GA group with log-transformed LOS as the outcome and included indicators for each CCS diagnosis and procedure, demographic variables, and clinical characteristics to identify variables with the greatest association with LOS. Generalized estimating equations for log-transformed LOS were used to adjust for variables identified by the CART models with a relative importance of >0.2 and accounted for patient clustering within hospitals. Using these models, we calculated and compared adjusted mean LOS with 95% confidence intervals for each level of neonatal care. The coefficient of variation (ie, SD or mean) was used to describe the variability in hospital-level adjusted mean LOS within and across levels of neonatal care. All analyses were conducted using SAS v.9.4 (SAS Institute, Cary, NC), and GraphPad Prism v.9.0 was used to generate violin plots. Statistical significance was set at P < .05.

There were 217 051 moderate and late preterm newborns from 2485 hospitals included (Supplemental Fig 1). Of these, 49.4% were 36 weeks’ gestation, 48.9% had government insurance, and 43.9% were in the South (Table 1). These newborns were cared for in hospitals that provided up to level II (19.7%), III (66.3%), or IV (11.1%) neonatal care.

Unadjusted analysis demonstrated that patient-level (race and ethnicity, primary payor, delivery type, multiple gestation, birth weight) and hospital-level (birth region, level of neonatal care) factors were significantly associated with LOS (Table 2). Notably, among 32 to 35 week GA infants, LOS was slightly shorter in level II centers (median LOS 3–21 [range 2–27]) compared with level III (median LOS 4–25 [range 3–31]) or IV (median LOS 4–25 [range 2–33]). These unadjusted findings were each statistically significant with a P value < .005.

After adjusting for clinical severity within each gestation, variation in LOS was seen between hospitals of the same level of neonatal care. Violin plots in Supplemental Fig 1 show the range of adjusted mean LOS among hospitals within each level of neonatal care broken down further by GA. For example, the adjusted mean LOS for a 35-week gestation newborn ranged from 2 days to 11.8 days between different level II hospitals. This wide range of adjusted mean LOS across hospitals was seen in all gestational ages and levels of neonatal care, with level II centers having the highest coefficient of variation (Supplemental Table 6). When comparing the variation, level II as compared with levels III and level IV showed the most significant variation, notably among 34- to 36- week GA neonates (P < .01).

The adjusted mean LOS also varied significantly between the levels of neonatal care. The greatest variation in LOS was 0.9 days (21 hours), seen in 32-week GA newborns between hospitals with level III and level IV neonatal care (Table 3). For 35- and 36-week gestation newborns, adjusted LOS was similar across level II–IV hospitals (4.33–4.35 days for 35-weekers in level II–IV neonatal care and 2.97 days for 36-weekers across all 3 levels of neonatal care). Despite the statistical difference, clinical significance is unlikely with mean LOS varying less than 1 day across all levels of neonatal care.

There was clinically and statistically significant variation in LOS of moderate and late preterm infants within the same level of neonatal care. Neonatal level II centers had the widest range of adjusted mean LOS, with a difference of over 2 weeks between centers caring for infants of the same GA after adjusting for severity. This variation represents a potential opportunity to standardize care. In previous studies, variation in outcomes of late preterm infants including LOS have been described between hospitals with different levels of neonatal care.^5,10,28  Our study also found a statistically significant variation in LOS between the different levels of neonatal care for newborns 32 to 36 weeks, however this variation was less than 1 day and may not be clinically meaningful.^29,30  The significant hospital to hospital variation in the care of moderate and late preterm infants is likely multifactorial.

First, the level of newborn care to which a moderate or late preterm infant may be admitted may vary based on state and hospital-specific regulations. The AAP outlines levels of neonatal care (I–IV) and specifies what each level of care should be capable of managing.⁶  However, there remains significant state variability of these level specifications and adherence to requirements. In June 2023, the AAP published “Standards for Levels of Neonatal Care: II, III, IV” as an implementation tool alongside the AAP policy and emphasized the need for a national verification program to improve the quality and consistency of neonatal care.¹⁵  In addition to level classification, specific admission criteria, including the GA below which babies should be admitted to a specific NICU level, are not standardized.^31,32  Thus, criteria vary significantly across states and hospitals,^4–9  and NICU admission may be influenced by factors unrelated to individual patient characteristics, including birth hospital characteristics, NICU census, regional NICU bed supply, and family-centered care priorities.^5,7–14  Although NICU care has clearly decreased mortality and morbidity in infants <32 weeks and VLBW,³³  clinical outcomes for moderate and late preterm infants and the role of NICU care within this population are less well described. In a study of one health system, newborns with similar conditions and GA who were admitted to the NICU for hospital-specific criteria or GA cutoff had a longer LOS and more medical interventions as compared with similar newborns remaining in the newborn nursery.³¹ Prospective studies that seek to define admission criteria in addition to age and weight cutoffs would help identify which population would most benefit from higher levels of neonatal care. Standardized NICU admission criteria have potential to reduce overutilization and decrease variation.

Second, standardized management approaches have been shown to reduce unnecessary LOS of infants in the NICU, however, practice guidelines for common diagnoses affecting moderate and late preterm infants are limited.^17–22  Specifically, there are no clinical practice guidelines or standardized protocols for apnea of prematurity, thermoregulation, and feeding issues, all of which have been associated with variation in NICU LOS.^20,21  One study used a multisite collaborative to successfully decrease LOS of preterm infants by 3 days by standardizing their approach to feeding, managing apnea, bradycardia, or desaturation events, and standardizing discharge protocols.²²  However, this study did not address the moderate and late preterm population. Quality improvement collaboratives such as the Vermont Oxford Network and California Perinatal Quality Care Collaborative have made meaningful impact in reducing healthcare associated infections and standardizing delivery room interventions but have focused less on the moderate to late preterm populations.^34,35  A more detailed investigation into variation in care of specific postbirth conditions among moderate and late preterm infants may help highlight which diagnoses would most benefit from standardized practice guidelines.

Lastly, another potential cause for variation in care between hospitals is a growing and varied provider workforce that cares for moderate and late preterm infants. These may include neonatologists, pediatric and neonatal hospitalists, and APPs.^16,36,37  These providers often have different practice styles, training, and backgrounds that may lead to variation in care.^9,16  The number of neonatologists per live birth decreased by 43% from 1981 to 2013.³⁸  This change in the neonatology workforce may be supporting a trend where non-neonatologists work in hospitals providing up to level II neonatal care, allowing neonatologists to focus on the care of more preterm infants in hospitals with levels III and IV neonatal care.⁵  Specifically in levels I and II neonatal care centers, pediatric hospitalists have played a role in decreasing LOS, facilitating transitions, and improving family centered care for newborns.^39–41  Further investigation is needed into both the workforce caring for newborns and provider influence on patient outcomes and LOS.

Variation in care, longer LOS, and overutilization of higher levels of neonatal care come at both financial and psychosocial cost.²²  Not only are there potential adverse financial effects to families and the entire healthcare system when caring for otherwise well preterm newborns in higher levels of care, but there is also a psychosocial impact on the families of these newborns, especially when the mother-infant dyad is separated.^{11–13,42–44}  With an estimated cost of 1 night in the NICU around $3000, even a small decrease in LOS or avoided admissions can have a large financial impact, especially given that many neonates’ healthcare is federally funded.²² 

Our study was limited by several factors. First, in KID, the specific level of neonatal care provided at each hospital is not available. Consequently, we developed criteria based on the AAP definitions⁶  to classify each hospital according to the highest level of neonatal care they provide. It is possible that some hospitals were misclassified. Hospitals with level I nurseries are most likely to be misclassified as level II for babies who require brief CPAP at delivery but stabilize shortly thereafter. Second, the level of neonatal care to which each patient was admitted is not available in KID, and neonates born at a hospital may not have required its highest level of care. Therefore, our data reflects the level of care of the hospital, not each individual neonate at that hospital, which is a level of granularity the database does not allow and could affect our results. Third, the KID database does not contain comprehensive clinical variables, including birth weight. We used exclusions such as CCCs and VLBW to attempt to create a more unified study population and adjusted our analyses for factors aligned with clinical severity utilizing CART modeling, which controlled for diagnoses such as neonatal opioid withdrawal syndrome, which is known to have longer LOS. However, there may still be residual factors confounding our results. Further, our study used length of stay as a marker of variation, but this falls short of identifying the specific causes and potential solutions for this variation in care. Also, there may be other factors we could not account for that could significantly affect LOS. Despite these limitations, this work is an effort to study variation of care in a growing population of moderate to late preterm infants within the constraints of available data. Additional research into the factors that impact variation in admission criteria and the variation in management of specific diagnoses may help identify more targeted areas for standardization and improvement. Future work via prospective studies or enhanced databases that include more comprehensive information regarding levels of neonatal care and clinical details for this population will promote further quality studies on neonates.

Our data adds to the growing literature on moderate and late preterm infants by demonstrating significant variation in LOS within the same level of neonatal care. By describing that variation in care exists primarily between hospitals within the same level of neonatal care, our study takes the first step toward identifying major contributing factors for future interventions to standardize and optimize care for this population. Further research is needed to identify factors influencing variable admission criteria to specific levels of neonatal care and to trend this data over time, which may allow for more targeted research and standardization to achieve higher value care for this population.

We thank neonatologist, Dr Brian Stone, who provided valuable insight and feedback during the planning of this study.