Characterization of Birth Hospitalizations in the United States

We performed a cross-sectional analysis of the most recent year of the Kids’ Inpatient Database (KID), 2019. KID was developed by the Agency for Healthcare Research and Quality and is the only all-payer hospital administrative data set designed to assess use of hospital services by children.¹²  The KID is part of a family of databases and software tools developed for the Healthcare Cost and Utilization Project (HCUP), which brings together data collection efforts of state data organizations, hospital associations, private data organizations, and the Federal government to create a national information resource of encounter-level health care data.¹²  KID includes a systematic random sample of pediatric discharges from all hospitals in 48 participating states. Discharge records contain deidentified, patient-level, clinical and resource use data included in a typical discharge abstract. The database provides discharge weights to extrapolate from sampled discharges and produce national estimates of discharges from all US, nonrehabilitation hospitals. This study qualified for exempt status from the university’s institutional review board. All analyses were performed using SAS 9.4 (SAS Institute, Inc, Cary, NC).

We sought to identify all hospitalizations related to birth to better understand common diagnoses and analyze costs. We included all hospitalizations with the indicator “in-hospital birth” (complicated and uncomplicated) in the KID database record and/or any hospitalizations with the Pediatric Clinical Classification System (PECCS)¹³  category of “liveborn,” including hospitalizations of newborns transferred in after birth. When conducting analyses by diagnosis, it is necessary to group International Classification of Diseases-10 Clinical Modification (ICD-10-CM) to clinically meaningful categories, as a diagnosis may have several potential codes. The PECCS classifies all 72 446 ICD-10-CM diagnosis codes into 834 categories that reflect the diagnostic labels physicians use to describe and communicate the conditions they manage in children (eg, respiratory distress syndrome). We chose to use the PECCS because of its specificity in categorizing pediatric-specific conditions using ICD-10-CM administrative data. The commonly used alternate classification system, HCUP’s Clinical Classifications Software, was not used because it is missing several important pediatric conditions because of its broad categories and adult focus.¹³  Since cost was a central part of our primary analysis, records were excluded if they were missing cost data (n = 14 629; 0.4%) so a single cohort definition could be used throughout the analysis. The specifications of ICD-10-CM codes for each PECCS category are available at: https://www.childrenshospitals.org/content/analytics/toolkit/pediatric-clinical-classification-system-peccs-codes.

We applied discharge-level survey weights to extrapolate from sampled discharges and produce national estimates. We described characteristics of children including age, sex, race, ethnicity, medical complexity (using the Pediatric Medical Complexity Algorithm categories: no chronic conditions, noncomplex chronic conditions, complex chronic conditions),¹⁴  median household income by patient zip code (based on standardized national quartiles), and primary insurance or payer. Race and ethnicity were self-identified by parents and families using each hospital’s classification system and were included as characteristics to describe the population of children with birth hospitalizations. We also described visit characteristics, including if the visit was a transfer, disposition (eg, home, against medical advice), and length of hospital stay (eg, number of days). Lastly, we described hospital characteristics including hospital region (eg, Northeast), control or ownership (eg, government), size (number of beds per hospital type), and type (rural, urban nonteaching, urban teaching, free-standing children’s). Cost was calculated by applying a hospital-level ratio of cost-to-charge to total hospital charges (cost data from 2019).

We analyzed all diagnoses (primary and secondary) coded during birth hospitalizations and categorized them using the PECCS.^15,16  As 99.9% of all hospitalizations had a primary diagnosis code grouped to the PECCS category “Liveborn,” it was not included in the analysis. Additionally, immunizations are recommended preventive care and a routine part of the birth hospitalization, thus we removed diagnosis codes indicating immunizations from the PECCS category “Immunizations and screening for infectious disease” to create a new category, “Screening or risk for infectious disease.” The total prevalence of these PECCS categories were rank-ordered.

The SAS procedure “surveyreg” was used to generate model point estimates and variance, incorporating discharge weights and clustering at the hospital and sampling stratum-level.¹⁷  HCUP KID encounters not included in the cohort were retained in the analysis with missing outcome values to ensure the inclusion of all clusters for proper variance estimation. A detailed report on estimate and variance generation is found in the statistical report provided by HCUP.¹² 

We determined marginal cost of each condition (relative contribution of each condition to total hospitalization costs). Conditions were rank-ordered by total marginal costs, calculated by multiplying condition-specific marginal costs by the total number of hospitalizations with that condition. To determine the marginal cost, binary indicators were created for each PECCS condition. These binary indicators were then input into a design-adjusted lognormal regression with SAS’s procedure “surveyreg.”¹⁷  We used a smearing estimate to retransform the expected values to the original scale, then used a derivative approach (ie, an increase of 1 in the estimates of having these conditions versus not produced a given change in the retransformed values).¹⁸ 

PECCS categories may include many heterogenous ICD-10-CM codes, so we performed a secondary analysis to generate frequency tables of all secondary ICD-10-CM codes. To better understand the diagnoses associated with transfer to other facilities, we determined the most prevalent secondary ICD-10-CM codes for birth hospitalizations that were transferred out. To better understand the drivers of prolonged length of hospital stay, we also determined the most prevalent secondary ICD-10-CM codes for birth hospitalizations that lasted longer than 2 days (median length of stay). Lastly, knowing that prematurity and associated intensive care needs are big drivers of cost in this population, we conducted a stratified analysis examining prevalence and costs for those with prematurity indicated in any diagnosis field versus not.

Demographics of birth hospitalizations are presented in Table 1. In 2019, there were 5 299 557 pediatric hospitalizations (age <18 years) in the United States, and 67% (n = 3 551 253) were for births. Although the vast majority of hospital admissions involved infants with no chronic medical condition diagnosis codes (n = 3 378 525; 95.1%), we found that 4.9% (n = 172 727) of birth hospitalizations had some chronic condition diagnosis code. The most common chronic medical conditions were in the cardiac (n = 88 979; 2.5%), musculoskeletal (n = 21 988; 0.6%), genitourinary (n = 20 488; 0.6%), neurologic (n = 14 189; 0.4%), genetic (n = 11 938; 0.3%), and pulmonary systems (n = 11 401; 0.3%). The median length of stay was 2 days (interquartile range [IQR] 2–3 days). We found that most hospitalizations resulted in a routine discharge home (n = 3 425 924; 96.5%), but 3.2% (n = 114 615) of admissions transferred to other facilities and 0.3% (n = 9268) were deaths.

The majority of birth hospitalizations occurred in private, nonprofit hospitals (n = 2 646 685 hospitalizations; 74.5%) and in large hospitals (n = 1 956 341 hospitalizations; 55.1%) (Table 2). We found that 72.3% (n = 2 567 827) of patients were born at urban teaching hospitals. Only 0.9% (n = 30 477) of hospitalizations occurred in free-standing children’s hospitals.

The 2019 KID database includes more than 3 million pediatric hospital inpatient records from 48 states and the District of Columbia. The database included 3998 hospitals, of which 2678 hospitals (70.0%) provided inpatient newborn care. Most hospitals were private, nonprofit (n = 1883 hospitals; 70.3%), large (n = 951 hospitals; 35.5%), and rural (n = 891 hospitals; 33.3%) or urban teaching (n = 1138 hospitals; 42.5%). Only 1.2% of the hospitals providing care for birth-related admissions nationally were free-standing children’s hospitals (n = 32). The median annual volumes of birth-related admissions in each hospital type varied widely, with urban teaching hospitals having high annual volumes (median 1738 [IQR 939–2871]) and free-standing children’s hospitals having much lower annual volumes (median 4 [IQR 1–523]). Although rural hospitals accounted for 33.3% of hospitals providing birth care nationally, only 9.4% of birth hospitalizations occurred in rural hospitals.

The most prevalent conditions were specified conditions originating in perinatal period (n = 1 021 099; 28.8%), neonatal hyperbilirubinemia (n = 540 112; 15.2%), screening or risk for infectious disease (n = 417 421; 11.8%), and preterm newborn (n = 314 288; 8.9%) (Table 3). Subgroup analysis of the ICD-10-CM diagnoses that make up the PECCS category “specified conditions originating in perinatal period” showed that the top diagnoses included heavy for gestational age newborns and postterm newborns, in addition to complex births (eg, newborn affected by breech delivery and extraction, meconium passage during delivery), maternal substance use (eg, newborn affected by maternal use of tobacco, opiates, cannabis), and pregnancy complications (eg, newborn affected by maternal hypertensive disorders, newborn affected by maternal infectious diseases) (Supplemental Table 6).

The total cost of birth hospitalizations in 2019 was $18.1 billion (95% confidence interval 16.7–19.6), of which $3.6 billion was associated with the “Liveborn” primary diagnosis and $14.5 billion was associated with marginal costs of secondary diagnoses. Conditions with the highest total marginal costs included specified conditions originating in the perinatal period ($957.6 million), neonatal hyperbilirubinemia ($813.8 million), preterm newborn ($674.9 million), neonatal jaundice with preterm delivery ($640.3 million), screening or risk for infectious disease ($619.0 million), and transient tachypnea of newborn ($577.4 million) (Table 3).

For the 114 615 birth hospitalizations transferred to other facilities, the most common conditions based on secondary ICD-10-CM codes included observation and evaluation of newborn for suspected infectious condition rule out (n = 22 122; 19%), respiratory distress syndrome of newborn (n = 18 361; 16%), neonatal jaundice associated with preterm delivery (n = 15 500; 14%), and apnea of newborn (n = 15 183; 13%) (Table 4).

Common conditions among the 1 000 560 birth hospitalizations with length of stay greater than the median 2 days similarly included neonatal jaundice (n = 240 175; 24%), observation and evaluation of newborn for suspected infectious condition rule out (n = 202 238; 20.2%), neonatal jaundice with preterm delivery (n = 129 407; 12.9%), neonatal hypoglycemia (n = 104 892; 10.5%), and transient tachypnea of newborn (n = 92 726; 9.3%) (Table 5). Infection condition rule out, neonatal jaundice, and respiratory conditions were top drivers of interfacility transfers, prolonged birth hospitalizations, and were also among the top 10 most costly conditions.

When stratifying by hospitalizations in which a diagnosis of prematurity was present versus not 342 244 (9.6%) of birth hospitalizations were for preterm infants. However, $7.6 billion (41.8%) of the $18.1 billion in total costs was attributed to these hospitalizations, making them over 13 times costlier ($30 866 for those with a diagnosis of prematurity versus $2361 for those without). Most conditions were more common in newborn hospitalizations (Supplemental Table 7). Respiratory distress and primary apnea were over 34 times more prevalent in preterm infants, and screening for infection diseases occurred over 3 times as often in newborn patients.

Hospitals are the most common place of birth in the United States, with 98.4% of births taking place in a hospital in 2017.¹⁹  We used the KID 2019, a nationally representative database, to describe birth hospitalizations, the types of hospitals in which they occur, and the most common and costly conditions managed during these hospitalizations. Roughly two-thirds of pediatric hospitalizations in 2019 in the United States were for births, totaling $18.1 billion in cost. Birth hospitalizations largely occurred in private, nonprofit and urban teaching hospitals, with small proportions occurring in free-standing children’s hospitals and rural hospitals. We found that specific conditions originating in the perinatal period, neonatal hyperbilirubinemia, and screening or risk for infectious disease were both common and costly. Thus, they represent the highest-impact targets for research and quality improvement efforts to improve infant health outcomes. We also identify the highest priority targets for reducing costs and improving health care value, including conditions that drive costly transfers: respiratory conditions (eg, respiratory distress syndrome), cardiac conditions, hyperbilirubinemia, and prematurity. Lastly, we quantify the differences in prevalent diagnoses and cost drivers in premature versus term infant hospitalizations. Thus, our work provides a global snapshot of birth-related health services nationally, compared with previous works that focus only on specific conditions and/or neonatal intensive care admissions.

The most prevalent clinical conditions in birth hospitalizations highlighted in this study, specified conditions originating in the perinatal period (eg, large for gestational age, postterm, complex births, maternal substance use, pregnancy complications), neonatal hyperbilirubinemia, and screening or risk for infectious disease, represent practical targets for health care systems seeking to target quality improvement efforts for the greatest impact. There are examples of successful efforts in these areas that could be more broadly scaled nationally. These have included interventions aimed at improving the assessment and management of infants with neonatal abstinence syndrome,^20,21  as well as the implementation of risk calculators to better standardize care and decrease unnecessary tests and interventions for neonates at risk for sepsis.^22–24  Tools for standardizing treatment of neonatal hyperbilirubinemia have been developed by the American Academy of Pediatrics,^25–27  with updated guidelines²⁸  released in August 2022 that continue to decrease unnecessary tests and interventions. These new guidelines represent a potentially high-impact focus for future quality improvement work.

We found overlap between top conditions managed for birth hospitalizations, resulting in transfers and prolonged length of hospital stay, highlighting drivers of more complex and costly birth hospitalizations. These included respiratory conditions, cardiac conditions, hyperbilirubinemia, and prematurity. We also found all these conditions were substantially more costly in preterm infants. These conditions represent important targets for better standardizing and improving safety of transfers and decreasing health care costs. A study by Rosenthal et al found a neonatal principal diagnosis was associated with higher odds of a transfer.²⁹  Additionally, it is known that children with multiple complex chronic conditions are an especially vulnerable population at risk for interfacility transfer, longer length of stay, and in-hospital mortality.³⁰  Use of handoff scripts have been shown in multiple settings to standardize intrahospital handoffs,³¹  whereas more recent proposed interventions involve standardizing the transfer communication process and incorporating technology such as telemedicine during transfers.³²  Future work could focus on tailoring such interventions for neonatal patients to better support safe transfer of complex birth hospitalizations.

We found one-third of included hospitals were rural, and it is known that ∼14% of the US population lives in rural areas.³³  However, only 9.4% of births occurred in rural hospitals in our analysis. This suggests people in rural areas may be receiving birth-related care in urban hospitals, perhaps seeking higher quality care or specialized services. This finding reflects recent trends in decreasing availability of maternal and infant services in rural hospitals,³⁴  which has been associated with poorer outcomes for birth hospitalizations nationally.³⁵  There has been a steady decline in hospital based obstetric services, with obstetric service losses concentrated in rural US counties.³⁶  Such loss of hospital-based obstetric services in rural US counties not adjacent to urban areas has been associated with increases in preterm births.³⁵  Prematurity is a major driver of childhood morbidity and mortality.^37,38  We found it was also a driver of cost and complexity (the fourth most common and costly clinical category, a top condition in birth hospitalizations with longer length of stay and interfacility transfers). As maternal health and infant health are inextricably intertwined, potential interventions to mitigate barriers to care and improve access to obstetric services in rural settings may be beneficial to improving both maternal and birth outcomes, though low patient volumes in these rural hospitals pose a challenge to maintaining hospital services and care quality.

This analysis has several limitations. First, our analysis included all diagnoses in any field (primary or secondary), and although this provides a more complete assessment of the conditions addressed in the birth hospitalization, these codes do not allow us to understand the clinical importance or severity of these diagnoses relative to one another. Additionally, we used the PECCS grouping system to facilitate analysis (as it reduces the total number of diagnosis codes into manageable categories); however, the grouping can limit interpretation in some instances, for example, the PECCS “specified conditions originating in perinatal period” includes a variety of diagnoses (Supplemental Table 6). Additionally, KID is an administrative database focused on discharge-level primary diagnoses, which does not include details on the clinical course, precluding examination of such details. For instance, the maternal indications for method of delivery (vaginal versus cesarean section) may be an important factor that drives infant length of stay beyond 2 days. Also, diagnoses are not linked to specific procedures or treatments, thus they provide indirect measures of the actual cost burden. And, since the KID is completely deidentified, there is no way to link together a hospitalization that spans 2 locations. Finally, a recent National Vital Statistics Reportestimates annual births of 3.75 million in 2019.³⁹  Discrepancies from this and our total estimate of 3.55 million births in 2019 may be because of HCUP sampling methods and/or out- of-hospital births.

Our study provides a global snapshot of birth-related health services nationally. Our study highlights that neonatal hyperbilirubinemia, infectious disease screening, and perinatal complications represent the highest-impact targets for research and quality improvement efforts to improve infant health outcomes. We also identify the highest priority targets for reducing costs and improving health care value, including conditions that drive costly transfers and care in the preterm infant population.