Abstract
Objective(s):
To evaluate contemporary trends in postmenstrual age (PMA) at discharge among infants born 23–33 completed weeks’ gestation and examine what factors predict discharge PMA by birth gestational age (GA).
Study Design:
This retrospective observational cohort study used Epic Systems Cosmos data. Infants born 01/01/2017–12/31/2023, at 23+0 to 33+6 weeks’ gestation who were not transferred and survived to discharge were included. We examined discharge PMA annually stratified by birth GA categories (23–24, 25–26, 27–28, 29–30, and 31–33 weeks) and assessed trends using linear mixed models. Random forest models examined factors predicting discharge PMA.
Results:
Median discharge PMA increased over the study period by 3.4 days (95% CI 3.1–3.8). Trends varied by GA, with the highest increase of 6.4 days (95% CI 2.1–10.8) for infants born at 23–24 weeks and the lowest 2.3 days (95% CI 1.8–2.6) for infants born at 31–33 weeks. Among all infants, the three most important factors predicting discharge PMA were: GA, bronchopulmonary dysplasia (BPD) diagnosis, and mechanical ventilation. When examining factors by GA category, the three most important factors for infants 23–24 weeks were: BPD diagnosis, unit, and ventilation; for infants 31–33 weeks they were ventilation, caffeine administration, and BPD diagnosis. Unit was also predictive of discharge PMA in all GA categories.
Conclusion:
Median discharge PMA increased among all preterm infants. While indicators of illness were expected predictors of discharge PMA, the importance of unit on discharge PMA within GA strata suggests that practice variation plays a significant role in discharge timing.
Keywords: Neonatal Intensive Care Unit, Prematurity, Hospital Discharge, Length of Stay, Practice Variation
Introduction
Utilization of neonatal intensive care units (NICUs) continues to evolve, with increasing NICU admission rates and longer length of stay for preterm infants (1–3). Prior data from the Vermont Oxford Network (VON) show that between 2005 and 2018, the adjusted postmenstrual age (PMA) at discharge increased by eight days among infants born 24–29 weeks’ gestation (3). From 2012 to 2020 this trend was appreciated among infants born 24–29 weeks and cared for in NICUs in Austria, Italy, and the United States (US) (increasing by 3.6, 4.0, and 5.8 days, respectively), but not in Ireland, Switzerland, or the United Kingdom (4). Understanding if such trends are continuing among preterm infants and identifying clinical and non-clinical factors that predict discharge timing may highlight opportunities to improve neonatal care delivery (5).
Prolonged NICU hospitalization has developmental implications for infants, psychosocial effects on parents and caregivers (6–8), and resource allocation ramifications on hospitals and health systems (9). Many studies of discharge timing focus on infants born <30 weeks’ gestation who are at higher risk of complications and prolonged hospitalization (3,4). Yet, there is increasing study of infants <34 weeks’ gestation who are routinely admitted to the NICU, comprise a larger proportion of preterm births, and also experience the adverse effects of NICU hospitalizations (10,11). Data show that processes and outcomes for extremely and very preterm infants differ from those of moderate and late preterm infants cared for within the same hospital (12). These gestational age (GA)-based differences may carry over into clinical interventions or practices that affect the trajectory of care and ultimately discharge decisions. Patient characteristics, clinical care practices, and hospital policies affect discharge timing. Furthermore, factors affecting length of stay and discharge timing may vary across the GA spectrum.
Based on the current literature (3,4,11,13), increasing preterm infant survival (14), and changes in health care delivery through a global pandemic (15), our objective was two-fold: to 1) evaluate contemporary trends from 2017 through 2023 in PMA at discharge among infants born at 23 weeks 0 days to 33 weeks and 6 days gestation and 2) examine how sociodemographic, birth and infant characteristics, clinical interventions, clinical diagnoses, and the unit vary with respect to their influence on prediction of discharge PMA by GA at birth.
Methods
Study Population
We performed a retrospective, observational cohort study using Epic Systems Corporation’s Cosmos data from hospitals in the United States. This study was deemed exempt by the Children’s Hospital of Philadelphia Institutional Review Board as it was classified as non-human participant research and the requirement of informed consent waived. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
Health organizations voluntarily submit data to Cosmos, which is an Epic-supported application that provides a framework to aggregate electronic health record (EHR) data for research. For patients cared for in more than one Cosmos-participating organization, patient records are deduplicated. From the submitted EHR data, Cosmos creates a de-identified limited dataset for research, which includes demographic characteristics, International Classification of Diseases (ICD) codes, Current Procedural Terminology (CPT) codes, vital signs, laboratory values, and medications. At the time of this study, there were 1,274 hospitals contributing data capturing more than 220 million patients. Health organizations that share a single EHR can comprise multiple hospitals and departments. The department identifier in Cosmos was used as the birth hospital’s neonatal unit.
Infants born between 23 weeks and 0 days and 33 weeks and 6 days gestation from January 1, 2017 to December 31, 2023 with linked maternal-infant records were included. Gestational age was determined based on the best obstetric estimate documented in the EHR. Infants born at 22 weeks completed GA were excluded given variation in resuscitation practices (16). Infants were excluded if there was an improbable birth or discharge weight (i.e. <210g or >10kg), a birth weight greater than 5 standard deviations from the mean for GA, were transferred to another organization, had a length of stay ≤7 days, or died prior to hospital discharge. Hospitals with <10 births annually (assumed to not have obstetric services) or those that did not contribute data in every year of the study period were excluded to ensure that changing organizations would not influence observed trends over time. Figure 1 depicts creation of the analytic cohort.
Figure 1.
Cohort Identification
Variable ascertainment
The primary outcome was PMA at hospital discharge. We examined several potentially predictive sociodemographic and clinical factors, previously shown to be associated with preterm infant hospital discharge timing, examined in prior work on this topic conducted by VON, and available in the Cosmos data (3). Maternal characteristics included race and ethnicity (categorized as Hispanic, non-Hispanic Asian, non-Hispanic Black or African American, non-Hispanic Other, non-Hispanic White, and Unknown as per maternal report in the electronic health record) (11,13,17); insurance (categorized as private/other or public) (18); the Centers for Disease Control and Prevention (CDC) social vulnerability index (SVI) (19), based on ZIP code reported by quartile with those in the top quartile being high risk; and Rural-Urban Commuting Area (RUCA) codes which assign rurality at the census tract (20). We included the following birth and infant characteristics: mode of birth (vaginal or cesarean) (13,21); infant sex (21,22); GA at birth (11,21); birth weight (21), small for gestational age (SGA; birthweight <10%ile for gestational age) (5,22), large for gestational age (LGA; birthweight >90%ile for gestational age) (17); and five minute Apgar score. The aforementioned sociodemographic and clinical factors are structured fields in the Cosmos EHR data, which in prior studies using Cosmos have been have been consistent with and reflective of national trends (15,23,24). We also examined the following clinical interventions and hospital morbidities that may predict prolonged hospitalization: receipt of any surfactant; receipt of any caffeine; receipt of any non-invasive continuous positive airway pressure (CPAP); receipt of any invasive ventilation; apnea of prematurity; sepsis; necrotizing enterocolitis (NEC; stage II or III); severe intraventricular hemorrhage (IVH; grade III or IV); retinopathy of prematurity (ROP; stage III or IV), and bronchopulmonary dysplasia (BPD) (5,22,25,26). Hospital interventions and morbidities were identified based on associated CPT and ICD-10 codes, consistent with health services research using other data sources (27,28). Hospital and birth year were also included given known practice variation and our focus on trends over time (12,29,30).
Statistical Analysis
We examined characteristics for differences across years and trends over time. Overall differences across years were assessed using Kruskal-Wallis testing. We repeated testing examining PMA at discharge annually and stratified by each completed GA week. To assess whether changes over time persisted after covariate adjustment, we used linear mixed-effects regression to model log-transformed PMA at discharge with three random effects: hospital for shared care practices, birthing parent for shared characteristics across pregnancies, and pregnancy for shared variance among multiple gestations. Fixed effects were included for mode of birth, infant sex, GA at birth (completed weeks), SGA, LGA, 5-minute Apgar score, receipt of any surfactant, receipt of any caffeine, diagnoses of apnea of prematurity, sepsis, NEC, severe IVH, ROP, and BPD. Models utilized complete case analysis (2.1% of infants were excluded from models based on missing covariate values). Relative changes in PMA at discharge by year were calculated through back-transformation and used to estimate absolute changes in median from the 2017 baseline. We applied random forest regression to generate variable importance plots (VIP) to identify the factors that best predicted PMA at discharge. All aforementioned sociodemographic, birth, and infant characteristics, clinical interventions and diagnoses, unit, and year were included in the random forest models. This approach was first completed for all eligible infants and subsequently stratified by GA birth category (23–24, 25–26, 27–28, 29–30, and 31–33 weeks).
Results
There were 75,678 eligible infants born in and cared for in 290 neonatal units, of which 2.8% (n=2,117) were born at 23–24 weeks, 6.1% (n=4,615) at 25–26 weeks, 10.4% (n=7,901) at 27–28 weeks, 17.0% (n=12,880) at 29–30 weeks, and 63.6% (n=48,165) at 31–33 weeks. Cohort characteristics by year are shown in Table 1. Several characteristics differed statistically over time, with notable increases in the proportion of infants with maternal comorbidity diagnoses (e.g., gestational hypertension), who were LGA, and those who received any caffeine or CPAP during their hospitalization.
Table 1.
Patient Sociodemographic and Clinical Characteristics, 2017–2023
| Year | P-value | |||||||
|---|---|---|---|---|---|---|---|---|
| 2017 | 2018 | 2019 | 2020 | 2021 | 2022 | 2023 | ||
| Infants (n) | 9871 | 11017 | 11199 | 10799 | 11161 | 11072 | 10559 | |
| Sociodemographic characteristics | ||||||||
| Maternal race and ethnicity, n (%) | <0.001 | |||||||
| Hispanic | 1389 (14.1) | 1632 (14.8) | 1639 (14.6) | 1614 (14.9) | 1773 (15.9) | 1908 (17.2) | 1864 (17.7) | |
| Non-Hispanic Asian | 449 (4.5) | 473 (4.3) | 510 (4.6) | 474 (4.4) | 488 (4.4) | 523 (4.7) | 532 (5.0) | |
| Non-Hispanic Black | 2848 (28.9) | 3129 (28.4) | 3296 (29.4) | 3266 (30.2) | 3261 (29.2) | 3079 (27.8) | 2849 (27.0) | |
| Non-Hispanic White | 4745 (48.1) | 5296 (48.1) | 5265 (47.0) | 4976 (46.1) | 5180 (46.4) | 5052 (45.6) | 4819 (45.6) | |
| Other | 356 (3.6) | 394 (3.6) | 385 (3.4) | 398 (3.7) | 368 (3.3) | 497 (4.5) | 375 (3.6) | |
| Unknown | 84 (0.9) | 93 (0.8) | 104 (0.9) | 71 (0.7) | 91 (0.8) | 103 (0.9) | 120 (1.1) | |
| Insurance, n (%) | <0.001 | |||||||
| Private/Other | 6778 (68.7) | 7442 (67.6) | 7672 (68.5) | 7471 (69.2) | 7835 (70.2) | 7819 (70.6) | 7478 (70.8) | |
| Public | 2997 (30.4) | 3430 (31.1) | 3401 (30.4) | 3168 (29.3) | 3178 (28.5) | 3097 (28.0) | 3030 (28.7) | |
| Unknown | 96 (1.0) | 145 (1.3) | 126 (1.1) | 160 (1.5) | 148 (1.3) | 156 (1.4) | 51 (0.5) | |
| High Risk SVI, n (%)a | 3847 (39.0) | 4236 (38.4) | 4311 (38.5) | 4155 (38.5) | 4251 (38.1) | 4337 (39.2) | 4139 (39.2) | 0.6 |
| Metropolitan Residing, n (%)b | 8220 (83.3) | 9250 (84.0) | 9395 (83.9) | 9154 (84.8) | 9467 (84.8) | 9432 (85.2) | 9009 (85.3) | <0.001 |
| Maternal clinical diagnoses | ||||||||
| Pre-existing Diabetes, n (%) | 269 (2.7) | 387 (3.5) | 377 (3.4) | 382 (3.5) | 422 (3.8) | 421 (3.8) | 422 (4.0) | <0.001 |
| Gestational Diabetes, n (%) | 1880 (19.0) | 2192 (19.9) | 2316 (20.7) | 2341 (21.7) | 2561 (22.9) | 2483 (22.4) | 2321 (22.0) | <0.001 |
| Gestational Hypertension, n (%) | 3741 (37.9) | 4498 (40.8) | 4572 (40.8) | 4595 (42.6) | 5108 (45.8) | 5202 (47.0) | 5047 (47.8) | <0.001 |
| Pregnancy & Birth Characteristics | ||||||||
| Multiple Gestation, n (%) | 2792 (28.3) | 3138 (28.5) | 3240 (28.9) | 2835 (26.3) | 2947 (26.4) | 3054 (27.6) | 2933 (27.8) | <0.001 |
| Cesarean Birth, n (%) | 6328 (64.1) | 6998 (63.5) | 7319 (65.4) | 7187 (66.6) | 7550 (67.6) | 7602 (68.7) | 6401 (60.6) | <0.001 |
| Infant characteristics | ||||||||
| Gestational age at birth (completed weeks), n (%) | <0.001 | |||||||
| 23 | 103 (1.0) | 115 (1.0) | 113 (1.0) | 111 (1.0) | 113 (1.0) | 109 (1.0) | 91 (0.9) | |
| 24 | 193 (2.0) | 207 (1.9) | 225 (2.0) | 194 (1.8) | 201 (1.8) | 190 (1.7) | 152 (1.4) | |
| 25 | 275 (2.8) | 331 (3.0) | 287 (2.6) | 283 (2.6) | 295 (2.6) | 277 (2.5) | 255 (2.4) | |
| 26 | 364 (3.7) | 391 (3.5) | 379 (3.4) | 401 (3.7) | 386 (3.5) | 356 (3.2) | 335 (3.2) | |
| 27 | 463 (4.7) | 463 (4.2) | 558 (5.0) | 467 (4.3) | 496 (4.4) | 441 (4.0) | 487 (4.6) | |
| 28 | 607 (6.1) | 664 (6.0) | 663 (5.9) | 688 (6.4) | 638 (5.7) | 660 (6.0) | 606 (5.7) | |
| 29 | 791 (8.0) | 729 (6.6) | 832 (7.4) | 797 (7.4) | 775 (6.9) | 787 (7.1) | 804 (7.6) | |
| 30 | 961 (9.7) | 1047 (9.5) | 1073 (9.6) | 1075 (10.0) | 1105 (9.9) | 1056 (9.5) | 1048 (9.9) | |
| 31 | 1295 (13.1) | 1491 (13.5) | 1461 (13.0) | 1380 (12.8) | 1423 (12.7) | 1390 (12.6) | 1339 (12.7) | |
| 32 | 1937 (19.6) | 2332 (21.2) | 2276 (20.3) | 2256 (20.9) | 2301 (20.6) | 2358 (21.3) | 2307 (21.8) | |
| 33 | 2882 (29.2) | 3247 (29.5) | 3332 (29.8) | 3147 (29.1) | 3428 (30.7) | 3448 (31.1) | 3135 (29.7) | |
| Infant female sex, n (%) | 4702 (47.6) | 5327 (48.4) | 5378 (48.0) | 5103 (47.3) | 5241 (47.0) | 5162 (46.6) | 5154 (49.8) | 0.054 |
| Birth weight, median (IQR) | 1609 (1207, 1950) | 1620 (1240, 1950) | 1613 (1220, 1945) | 1616 (1223, 1958) | 1630 (1240, 1970) | 1630 (1250, 1955) | 1625 (1250, 1954) | 0.01 |
| SGA, n (%)c | 817 (8.3) | 896 (8.1) | 949 (8.5) | 959 (8.9) | 1005 (9.0) | 993 (9.0) | 943 (8.9) | 0.09 |
| LGA, n (%)d | 868 (8.8) | 985 (8.9) | 991 (8.8) | 1012 (9.4) | 1027 (9.2) | 1012 (9.1) | 1061 (10.0) | 0.03 |
| Medications or Interventions Received | ||||||||
| Surfactant, n (%) | 3165 (32.1) | 3346 (30.4) | 3499 (31.2) | 3311 (30.7) | 3517 (31.5) | 3431 (31.0) | 3342 (31.7) | 0.12 |
| Caffeine, n (%) | 5683 (57.6) | 6467 (58.7) | 6675 (59.6) | 6627 (61.4) | 6880 (61.6) | 6895 (62.3) | 6670 (63.2) | <0.001 |
| CPAP, n (%) | 5870 (59.5) | 7045 (63.9) | 7474 (66.7) | 7552 (69.9) | 8015 (71.8) | 8229 (74.3) | 8268 (78.3) | <0.001 |
| Invasive ventilation, n (%) | 4391 (44.5) | 4943 (44.9) | 5169 (46.2) | 4830 (44.7) | 5031 (45.1) | 4749 (42.9) | 4612 (43.7) | <0.001 |
| Infant clinical diagnoses | ||||||||
| Apnea of prematurity, n (%) | 6472 (65.6) | 7463 (67.7) | 7615 (68.0) | 7485 (69.3) | 7703 (69.0) | 7423 (67.0) | 6755 (64.0) | <0.001 |
| Sepsis, n (%) | 1434 (14.5) | 1604 (14.6) | 1367 (12.2) | 1249 (11.6) | 1283 (11.5) | 1174 (10.6) | 1041 (9.9) | <0.001 |
| NEC, n (%)e | 109 (1.1) | 143 (1.3) | 125 (1.1) | 141 (1.3) | 129 (1.2) | 138 (1.2) | 147 (1.4) | 0.4 |
| Severe IVH, n (%)f | 162 (1.6) | 198 (1.8) | 222 (2.0) | 139 (1.3) | 201 (1.8) | 184 (1.7) | 184 (1.7) | 0.006 |
| ROP, n (%)g | 226 (2.3) | 242 (2.2) | 247 (2.2) | 247 (2.3) | 270 (2.4) | 225 (2.0) | 204 (1.9) | 0.2 |
| BPD, n (%) | 1020 (10.3) | 1178 (10.7) | 1038 (9.3) | 1093 (10.1) | 1077 (9.6) | 1092 (9.9) | 1139 (10.8) | 0.001 |
Abbreviations: BPD – Bronchopulmonary dysplasia, CPAP – Continuous Positive Airway Pressure, IQR – Interquartile Range, IVH – Intraventricular Hemorrhage, LGA – Large for Gestational Age, NEC – Necrotizing Enterocolitis, ROP – Retinopathy of Prematurity, SGA – Small for Gestational Age, SVI - Social Vulnerability Index
High risk is defined as census tracts in the top quartile of SVI scores.
Metropolitan residence is based on Rural-Urban Commuting Area (RUCA) codes 1–3, which are assigned at the census tract level.
SGA is defined as a birth weight <10% tile for gestational age.
LGA is defined as a birth weight >90% tile for gestational age.
NEC was defined as stage II or III.
Severe IVH was defined as grade III or IV.
ROP was defined as stage III or IV.
Among all infants in the cohort, the crude median PMA at discharge increased by 4 days between 2017 and 2023. The crude median in 2017 was 36 weeks and 4 days (interquartile range [IQR] 35 weeks and 5 days to 38 weeks and 1 day) and in 2023 was 37 weeks and 1 day (IQR 36 weeks and 0 days to 38 weeks and 6 days). Our analyses of trends over time demonstrated significant effects of year (p <0.001) for the cohort as a whole and when stratified by GA week (p <0.001 for all). The PMA at discharge increased for all GA weeks, with those born more preterm with later PMA at discharge over time (Figure 2).
Figure 2.
Discharge Postmenstrual Age by Gestational Age at Birth, 2017–2023
In adjusted analyses, the median PMA at discharge among all eligible infants increased by 1.3% (95% confidence interval [CI] 1.2%, 1.4%) during the study period, or 3.4 days (95% confidence interval [CI] 3.1, 3.8) from 36 weeks and 4 days to 37 weeks and 0 days. Among infants born at 23–24 weeks’ gestation, the median PMA at discharge had an adjusted increase of 6.4 days (95% CI 2.1–10.8) from 290 (41 weeks and 3 days) to an estimated 296 days (42 weeks and 2 days), while among infants born at 33–34 weeks’ gestation, the median PMA at discharge increased by 2.3 days (95% CI 1.8–2.6) from 254 days (36 weeks and 2 days) to an estimated 256 days (36 weeks and 4 days).
When including all infants in the analytic cohort, the random forest importance analysis demonstrated that the five factors most associated with discharge PMA in descending importance were GA at birth (completed week), a diagnosis of BPD, invasive ventilation, receipt of any caffeine, and unit (Figure 3). Earlier GA at birth, BPD, invasive ventilation, and caffeine all predicted older PMA at discharge. When stratified by GA birth category, the variable importance plots revealed some differences in key factors by category (Figure 4). The five most important variables for infants 23–24 weeks’ gestation in descending importance were BPD, unit, invasive ventilation, a diagnosis of sepsis, and SGA (Figure 4). In contrast, for infants born at 31–33 weeks’ gestation, the most important variables were invasive ventilation, receipt of any caffeine, BPD, SGA, and unit (Figure 4). The only factors that were among the five most important variables to predict discharge across all GA categories were BPD, SGA, and unit.
Figure 3.
Variable of importance plot for PMA at discharge among infants born at 23 to 33 completed weeks’ gestation
Abbreviations: GA – Gestational Age, BPD – Bronchopulmonary Dysplasia, SGA – Small for Gestational Age, CPAP – Continuous Positive Airway Pressure
Variable importance plots were generated using patient characteristics (e.g., birth year, maternal characteristics, mode of birth, and sociodemographic characteristics), clinical diagnoses (e.g., BPD, SGA), clinical interventions (e.g., exposure to invasive ventilation or CPAP, receipt of medications), and the birth hospital’s neonatal unit where they were cared for and discharged from (e.g., indicated by Unit).
Figure 4.
Variable of importance plot for PMA at discharge stratified by categories of completed weeks gestational age at birth
Abbreviations: BPD – Bronchopulmonary Dysplasia, CPAP – Continuous Positive Airway Pressure, IVH – Intraventricular Hemorrhage, NEC – Necrotizing Enterocolitis, ROP – Retinopathy of Prematurity, RUCA – Rural-Urban Commuting Area, SGA – Small for Gestational Age.
Variable importance plots were generated using patient characteristics (e.g., birth year, maternal characteristics, mode of birth, and sociodemographic characteristics), clinical diagnoses (e.g., BPD, NEC, ROP, SGA), clinical interventions (e.g., exposure to invasive ventilation or CPAP, receipt of medications), and the birth hospital’s neonatal unit where they were cared for and discharged from (e.g., indicated by Unit).
Discussion
From 2017 to 2023, we found that the PMA at discharge among preterm infants born between 23 and 33 completed weeks’ gestation continued to rise. While these trends were experienced by all infants, the change in PMA at discharge was most notable among infants born at the lowest GA. Though respiratory illness factors were most predictive of PMA at discharge, the neonatal unit also emerged as an important factor. Collectively, these findings shed light on opportunities to optimize care delivery and identify process variation among hospitals that may contribute to discharge timing.
Further validating the Cosmos dataset, our findings confirm similar trends reported by VON, which showed that between 2005 and 2018 the adjusted PMA at discharge among infants 24 to 29 weeks GA increased by 8 days over 13 years (3), and we estimate an increase of 3.4 days over 6 years. These data also extend prior VON findings as we demonstrate discharge at later PMA across a wider GA range (23 to 33 weeks). Our study also includes years before and after the coronavirus 2019 (COVID-19) pandemic, during which there were changes in discharge timing and hospital length of stay among term infants (15). Similar changes in trends were not explicitly seen in this preterm population, which is not surprising given the need for preterm infants to reach clinical stability and physiologic maturity prior to discharge, regardless of the pandemic (31).
Our use of variable importance plots to examine how important factors are with respect to predicting discharge timing overall and by GA group is novel. Advantages to this approach include examining multiple types of factors and highlighting the relative predictive strength of baseline characteristics, respiratory diagnoses and associated interventions, and unit. A prior study using California data from 2008 to 2010 examined different risk factors for longer length of stay by birth weight group (<1,000g, 1,000–1,500g, 1,500g-2,500g and >2,500g) and noted that risk factors varied across groups, findings that conceptually parallel the current study demonstrating differences across GA categories (17). The California study focused largely on sociodemographic and baseline clinical characteristics (e.g., race and ethnicity, maternal hypertensive disorders of pregnancy, birth weight, SGA, LGA, infant sex, etc.), which our current study similarly identified as important factors with respect to discharge timing, although neither race and ethnicity, nor hypertensive disorders were in the top five factors in our overall data. Our study builds on prior literature by also including clinical interventions and diagnosed comorbidities identifying the most influential factors. It is not surprising that infants with respiratory immaturity, requiring more interventions (e.g., caffeine, CPAP, ventilation), the use of which has been increasing over time (32), or with a BPD diagnosis would be discharged at a later PMA (33). While these infants may be sicker, this analysis highlights the importance of these factors and potential value of examining associated variation in management and practice patterns. Weight thresholds for discharge, which are used in many units and associated studies, are another likely source of variation and their relationship is reflected in the importance of SGA (34).
Neonatal unit was a notable variable that emerged in this analysis. Prior studies have reported marked variation in discharge timing between hospitals, with some studies reporting as much as 2 weeks (5). Our unit variable may be capturing a variety of hospital characteristics, such as level of care, admission volume, hospital ownership, and geographic region, some of which have been shown to influence discharge timing (11). However, this variable likely also captures unmeasured hospital-based characteristics: for example, variation in preterm care practices, from varied CPAP use in the delivery room to uneven adoption of neuro-centric preterm infant care practices between hospitals (29,35). These and other potential varied practices, especially those related to respiratory care based on other important covariates identified in this analysis, may contribute to differences in PMA at discharge. Discharge decisions for each individual infant are complex, and many units have policies regarding discharge such as weight thresholds, apnea/bradycardia/desaturation event monitoring, and potential for discharge with a nasogastric tube. Quality improvement is a valuable tool for creating significant change in adherence to unit practices and reducing practice variation for conditions that are closely tied to discharge (36). Notably, the literature shows that specific to discharge timing, compared to individual unit quality improvement work, participation in a quality improvement collaborative is associated with a 3-day decrease in length of stay among infants 27–31 weeks completed GA (37). It is also possible that case-mix of infant illness severity could explain part of the hospital-level variation observed. However, given that hospital remained important even when incorporating GA into the analysis and stratifying by GA groups, we suspect that hospital practice variation is responsible for much of the variation observed. Given that length of stay is often cited as a hospital- or unit-level measure of care quality, our results highlight potential populations and areas of care for targeted review of clinical practices and policies that may optimize discharge timing for preterm infants.
Length of stay is both an important and nuanced measure, whether considering implications at the hospital-level or individual patient-level. For hospitals and their NICUs, longer length of stay may decrease NICU bed availability. This combination may create strain on NICU capacity, which when high as been associated with adverse outcomes for preterm infants (38). Prolonged length of stay is also challenging for patients and their families, with parental quality of life scores lowest while their infants remain hospitalized (39). Hospitals and clinicians must balance unit constraints, parental well-being, and the individual infant’s clinical needs to achieve a safe discharge, while minimizing the risk of infant readmission. As illustrated by prior quality improvement work, decreasing length of stay is feasible (37). Successfully quality improvement work has largely focused on standardizing practices, such as those pertaining to nutritional support (e.g., timing of initiating feeds, transitioning to oral feeding), the approach to apnea, bradycardia, and desaturation events (e.g., definition of such an event, duration between an event and discharge), as well as discharge criteria (37). Attention to and sharing of clinical practice and policies is well suited for quality improvement projects or implementation science studies, which are designed to evaluate of a new or updated policy. Conducting such work within health systems, state-based collaboratives, or larger networks of NICUs, has the potential to positively affect infants and their families as well as the units and health systems caring for them.
Study limitations
Although Cosmos is a large dataset, currently capturing ~25% of US births annually, it is not designed to be population-representative. However, its size and inclusion of a wide variety of health systems (e.g., county, community, and academic) across the US improve generalizability. While variable importance analysis may be helpful in identifying factors that best predict an outcome, importance rankings can be difficult to interpret with highly correlated variables, such as those related to respiratory immaturity and its comorbidities. Nevertheless, these analyses reveal other predictors that may highlight opportunities to review care practices. Details regarding hospital characteristics and practices are not currently available in these data. We conducted our analysis at the level of the birth hospital neonatal unit. This allowed for examination of variation between hospital units given that health systems may have multiple departments present in Cosmos. However, birth hospital neonatal units (departments) in Cosmos cannot be identified nor linked to other data sources that contain additional hospital characteristics (e.g., the American Hospital Association Annual Hospital Survey). Similarly, specific hospital policies and practices, such as those around feeding (including discharge with a nasogastric tube), apnea/bradycardia/desaturation events, discharge or outpatient follow up criteria are not available, which is the case in many observational studies. As data become available, future studies may have more nuanced analyses, such as capturing discharge timing relative to the last nasogastric feed or event associated with apnea, bradycardia, or desaturation. In addition, it is possible that patient factors (e.g., prenatal opioid exposure or maternal medication use) that we did not include in our analysis, could have changed over time and contributed to our findings. Finally, datasets that rely on EHR data can reflect the accuracy of data entered by clinicians and staff and may include erroneous entries or missing diagnoses (e.g., BPD). However, we would expect these errors to be stable over time thus unlikely to threaten study findings.
Conclusion
We found that the PMA at discharge among preterm infants born 23–33 weeks continues to increase. Important variables associated with discharge vary by gestational age and efforts to understand and systematically address variation in preterm infant care practices and policies within and between hospitals may illuminate opportunities to improve neonatal care and optimize length of stay.
Funding:
This work was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development [grant number K23 HD109426 to SCH].
Abbreviations:
- GA
Gestational Age
- NICU
Neonatal Intensive Care Unit
- PMA
Postmenstrual Age
Footnotes
Conflicts of Interest: None
Author CRediT statement:
Sara C. Handley: Conceptualization, Methodology, Writing – Original draft. Melissa Schmatz: Conceptualization, Writing – Review & Editing, Evan Miller: Software, Data curation. Heather H. Burris: Conceptualization, Methodology, Writing – Review & Editing, Supervision. Justin Y. Lo: Conceptualization, Methodology, Writing – Review & Editing. Lindsay Knake: Software, Writing – Review & Editing. Moeun Son: Methodology, Writing – Review & Editing. Kathryn McKenney: Methodology, Writing – Review & Editing. Jennifer Culhane: Methodology, Writing – Review & Editing. Jay Greenspan: Conceptualization, Writing – Review & Editing, Supervision. Kevin Dysart: Conceptualization, Methodology, Software, Formal analysis, Resources, Data curation, Writing – Review & Editing, Visualization, Supervision.
References
- 1.Harrison W, Goodman D. Epidemiologic Trends in Neonatal Intensive Care, 2007–2012. JAMA Pediatr. 2015. Sep;169(9):855–62. doi: 10.1001/jamapediatrics.2015.1305. [DOI] [PubMed] [Google Scholar]
- 2.Hughes CS, Schmitt S, Passarella M, Lorch SA, Phibbs CS. Who’s in the NICU? A population-level analysis. J Perinatol. 2024;44(10):1416–1423. doi: 10.1038/s41372-024-02039-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Edwards EM, Greenberg LT, Ehret DEY, Lorch SA, Horbar JD. Discharge age and weight for very preterm infants: 2005⇓2018. Pediatrics. 2021;147(2):e20200016006. doi: 10.1542/peds.2020-016006. [DOI] [PubMed] [Google Scholar]
- 4.Edwards EM, Greenberg LT, Horbar JD, Gagliardi L, Adams M, Berger A, et al. Discharge Age and Weight for Very Preterm Infants in Six Countries: 2012–2020. Neonatology. 2023. Mar 28;120(2):208–16. doi: 10.1159/000528013 [DOI] [PubMed] [Google Scholar]
- 5.Altman M, Vanpée M, Cnattingius S, Norman M. Moderately preterm infants and determinants of length of hospital stay. Arch Dis Child Fetal Neonatal Ed. 2009. Nov 1;94(6):F414–F418. doi: 10.1136/adc.2008.153668 [DOI] [PubMed] [Google Scholar]
- 6.Whitfield MF. Psychosocial effects of intensive care on infants and families after discharge. Semin Neonatol. 2003. Apr 1;Community Neonatology: The continuing care of NICU graduates after hospital discharge; 8(2):185–93. doi: 10.1016/S1084-2756(02)00218-X [DOI] [PubMed] [Google Scholar]
- 7.Treyvaud K, Spittle A, Anderson PJ, O’Brien K. A multilayered approach is needed in the NICU to support parents after the preterm birth of their infant. Early Hum Dev. 2019. Dec 1;139:104838. doi: 10.1016/j.earlhumdev.2019.104838 [DOI] [PubMed] [Google Scholar]
- 8.McAndrew S, Acharya K, Westerdahl J, Brousseau DC, Panepinto JA, Simpson P, et al. A Prospective Study of Parent Health-Related Quality of Life before and after Discharge from the Neonatal Intensive Care Unit. J Pediatr. 2019. Oct 1;213:38–45.e3. doi: 10.1016/j.jpeds.2019.05.067 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Boss RD, Henderson CM, Weiss EM, Falck A, Madrigal V, Shapiro MC, et al. The Changing Landscape in Pediatric Hospitals: A Multicenter Study of How Pediatric Chronic Critical Illness Impacts NICU Throughput. Am J Perinatol. 2022. Apr;39(6):646–51. doi: 10.1055/s-0040-1718572. [DOI] [PubMed] [Google Scholar]
- 10.Ohuma EO, Moller AB, Bradley E, Chakwera S, Hussain-Alkhateeb L, Lewin A, et al. National, regional, and global estimates of preterm birth in 2020, with trends from 2010: a systematic analysis. The Lancet. 2023. Oct 7;402(10409):1261–71. doi: 10.1016/S0140-6736(23)00878-4. [DOI] [PubMed] [Google Scholar]
- 11.Qattea I, Ejaz S, Kattea MO, Kteish R, Chaudry S, Aly H. The effect of demographic, financial and hospital factors on the length of stay of preterm infants: a US national database study. J Perinatol. 2024. Feb;44(2):173–8. doi: 10.1038/s41372-023-01836-9. [DOI] [PubMed] [Google Scholar]
- 12.Salazar EG, Handley SC, Greenberg LT, Edwards EM, Lorch SA. Measuring quality of care in moderate and late preterm infants. J Perinatol. 2022. Oct;42(10):1294–300. doi: 10.1038/s41372-022-01377-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Ismail L, Markowsky A, Adusei-Baah C, Gallizzi G, Hall M, Kalburgi S, et al. Variation in Length of Stay by Level of Neonatal Care Among Moderate and Late Preterm Infants. Hosp Pediatr. 2024. Jan 1;14(1):37–44. doi: 10.1542/hpeds.2023-007252. [DOI] [PubMed] [Google Scholar]
- 14.Bell EF, Hintz SR, Hansen NI, Bann CM, Wyckoff MH, DeMauro SB, et al. Mortality, In-Hospital Morbidity, Care Practices, and 2-Year Outcomes for Extremely Preterm Infants in the US, 2013–2018. JAMA. 2022. Jan 18;327(3):248–63. doi: 10.1001/JAMA.2021.23580. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Handley SC, Gallagher K, Breden A, Lindgren E, Lo JY, Son M, et al. Birth Hospital Length of Stay and Rehospitalization During COVID-19. Pediatrics. 2022. Jan 1;149(1). doi: 10.1542/PEDS.2021-053498. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Rysavy MA, Li L, Bell EF, Das A, Hintz SR, Stoll BJ, et al. Between-hospital variation in treatment and outcomes in extremely preterm infants. N Engl J Med. 2015;372(19):1801–11. doi: 10.1056/NEJMoa1410689. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Lee HC, Bennett MV, Schulman J, Gould JB, Profit J. Estimating Length of Stay by Patient Type in the Neonatal Intensive Care Unit. Am J Perinatol. 2016. Feb 18;33:751–7. doi: 10.1055/s-0036-1572433 [DOI] [PubMed] [Google Scholar]
- 18.Brandon GD, Adeniyi-Jones S, Kirkby S, Webb D, Culhane JF, Greenspan JS. Are outcomes and care processes for preterm neonates influenced by health insurance status? Pediatrics. 2009. Jul 1;124(1):122–7. doi: 10.1542/peds.2008-1318 [DOI] [PubMed] [Google Scholar]
- 19.Givens M, Teal EN, Patel V, Manuck TA. Preterm birth among pregnant women living in areas with high social vulnerability. Am J Obstet Gynecol MFM. 2021. Sep;3(5). doi: 10.1016/j.ajogmf.2021.100414 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Bourque SL, Levek C, Melara DL, Grover TR, Hwang SS. Prevalence and Predictors of Back-Transport Closer to Maternal Residence After Acute Neonatal Care in a Regional NICU. Matern Child Health J. 2019. Feb 1;23(2):212–9. doi: 10.1007/s10995-018-2635-6 [DOI] [PubMed] [Google Scholar]
- 21.Aly H, Hoffman H, El-Dib M, Said L, Mohamed M. Factor affecting length of stay in late preterm infants: an US national database study. J Matern Fetal Neonatal Med. 2015. Mar 24;28(5):598–604. doi: 10.3109/14767058.2014.927428. [DOI] [PubMed] [Google Scholar]
- 22.Korvenranta E, Linna M, Häkkinen U, Peltola M, Andersson S, Gissler M, et al. Differences in the length of initial hospital stay in very preterm infants. Acta Paediatr. 2007;96(10):1416–20. doi: 10.1111/j.1651-2227.2007.00471.x [DOI] [PubMed] [Google Scholar]
- 23.Son M, Gallagher K, Lo JY, Lindgren E, Burris HH, Dysart K, et al. Coronavirus Disease 2019 (COVID-19) Pandemic and Pregnancy Outcomes in a U.S. Population. Obstet Gynecol. 2021. Oct;138(4):542–51. doi: 10.1097/AOG.0000000000004547. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Son M, Culhane JF, Louis JM, Handley SC, Burris HH, Greenspan J, et al. Severe maternal morbidity rates in a US-based electronic health record database, 2018–2022. J Perinatol. 2023. Oct;43(10):1316–8. doi: 10.1038/s41372-023-01765-7. [DOI] [PubMed] [Google Scholar]
- 25.Morrow CB, McGrath-Morrow SA, Collaco JM. Predictors of length of stay for initial hospitalization in infants with bronchopulmonary dysplasia. J Perinatol. 2018. Sep;38(9):1258–65. doi: 10.1038/s41372-018-0142-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Fu M, Song W, Yu G, Yu Y, Yang Q. Risk factors for length of NICU stay of newborns: A systematic review. Front Pediatr. 2023. Mar 13;11:1121406. doi: 10.3389/fped.2023.1121406 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Ford JB, Roberts CL, Algert CS, Bowen JR, Bajuk B, Henderson-Smart DJ. Using hospital discharge data for determining neonatal morbidity and mortality: a validation study. BMC Health Serv Res. 2007. Nov 20;7:188. doi: 10.1186/1472-6963-7-188. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Ryckman KK, Holdefer PJ, Sileo E, Carlson C, Weathers N, Jasper EA, et al. The Validity of Hospital Diagnostic and Procedure Codes Reflecting Morbidity in Preterm Neonates Born <32 Weeks Gestation. J Perinatol Off J Calif Perinat Assoc. 2023. Nov;43(11):1374–8. doi: 10.1038/s41372-023-01685-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Handley SC, Salazar EG, Greenberg LT, Foglia EE, Lorch SA, Edwards EM. Variation and temporal trends in delivery room management of moderate and late preterm infants. Pediatrics. 2022. Aug 1;150(2):e201055994. doi: 10.1542/peds.2021-055994. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Edwards EM, Horbar JD. Variation in use by NICU types in the United States. Pediatrics. 2018. Nov;142(5):e20180457. doi: 10.1542/peds.2018-0457. [DOI] [PubMed] [Google Scholar]
- 31.Bakewell-Sachs S, Medoff-Cooper B, Escobar GJ, Silber JH, Lorch SA. Infant Functional Status: The Timing of Physiologic Maturation of Premature Infants. Pediatrics. 2009. May 1;123(5):e878–86. doi: 10.1542/peds.2008-2568 [DOI] [PubMed] [Google Scholar]
- 32.Hatch LD, Clark RH, Carlo WA, Stark AR, Ely EW, Patrick SW. Changes in Use of Respiratory Support for Preterm Infants in the US, 2008–2018. JAMA Pediatr. 2021. Jul 6;175(10):1017–24. doi: 10.1001/JAMAPEDIATRICS.2021.1921 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Hibbs AM, Chen Z, Minich NM, Martin RJ, Raffay TM, MacFarlane PM, et al. Association between Intermittent Hypoxemia and NICU Length of Stay in Preterm Infants. Neonatology. 2024;121(3):327–35. doi: 10.1159/000535264. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Merritt TA, Pillers D, Prows SL. Early NICU discharge of very low birth weight infants: a critical review and analysis. Semin Neonatol. 2003. Apr 1;Community Neonatology: The continuing care of NICU graduates after hospital discharge; 8(2):95–115. doi: 10.1016/S1084-2756(02)00219-1 [DOI] [PubMed] [Google Scholar]
- 35.Handley SC, Passarella M, Lorch SA, Lee HC. Survey of preterm neuro-centric care practices in California neonatal intensive care units. J Perinatol. 2019. Feb 1;39(2):256–62. doi: 10.1038/s41372-018-0283-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Coughlin K, Posencheg M, Orfe L, Zachritz W, Meadow J, Yang K, et al. Reducing Variation in the Management of Apnea of Prematurity in the Intensive Care Nursery. Pediatrics. 2020. Feb 1;145(2):e20190861. doi: 10.1542/peds.2019-0861 [DOI] [PubMed] [Google Scholar]
- 37.Lee HC, Bennett MV, Crockett M, Crowe R, Gwiazdowski SG, Keller H, et al. Comparison of Collaborative Versus Single-Site Quality Improvement to Reduce NICU Length of Stay. Pediatrics. 2018. Jul;142(1):e20171395. doi: 10.1542/peds.2017-1395. [DOI] [PubMed] [Google Scholar]
- 38.Salazar EG, Passarella M, Formanowski B, Rogowski J, Edwards EM, Halpern SD, et al. The association of NICU capacity strain with neonatal mortality and morbidity. J Perinatol. 2025;45(12):1801–8. doi: 10.1038/s41372-025-02449-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Yip SASE, Lim QZX, Kong G, Amin Z, Ng YPM. Quality of Life of Parents of Premature Infants: A Systematic Review and Meta-Analysis. JAMA Netw Open. 2026. Jan 2;9(1):e2553712. doi: 10.1001/jamanetworkopen.2025.53712. [DOI] [PMC free article] [PubMed] [Google Scholar]