Introduction

Moderate and late preterm (MLP) infants, born between 32 and 36 weeks of gestation (WG), account for approximately 85% of the 13.4 million preterm births worldwide each year.1,2 This population largely explains the rising rates of preterm birth observed in recent decades.1,3 Compared with term infants, MLP infants have higher risks of death, neonatal morbidity and poor long-term developmental outcomes.4,5,6,7,8,9 Optimizing perinatal management is therefore essential to reduce short- and long-term complications.

In high-income countries, perinatal regionalization has been extensively evaluated for very preterm and very low birth weight infants, demonstrating improved outcomes for births in tertiary centers.10,11,12,13 However, evidence guiding regionalization strategies for MLP infants remains limited, despite the existence of guidelines in many countries. The MLP population is heterogeneous, with prognosis influenced by gestational age, but also fetal growth restriction (FGR), multiple gestation, and maternal comorbidities such as hypertensive disorders and diabetes.14,15,16,17,18 This heterogeneity raises questions about the criteria used to guide level-of-care recommendations for this large and vulnerable population.

In France, as in most high-income countries,19,20 neonatal care is organized into a three-tiered system, including (i) level I maternity units, which have no neonatal unit; (ii) level II maternity units which have a neonatal unit without (IIa) or with (IIb) the capacity to provide non-invasive ventilation and short-term mechanical ventilation; and (iii) level III maternity units, which provide the highest level of care.21 National guidelines are primarily based on gestational age and estimated fetal weight.22 Birth in a level IIb or III maternity unit is recommended for moderate preterm births with an estimated fetal weight of 1500 g or more, while deliveries between 34+0 and 35+6 WG can be carried out in level IIa, IIb or III maternity units. Starting at 36+0 WG, care can be provided in any level of maternity unit.22

Although neonatal transfer is an integral component of regionalized systems, minimizing its occurrence is a key aim of care provision, given associated clinical risks23 and potential for disrupting early mother-infant bonding.24,25 Clinical risks include physiological instability, particularly cardiovascular, respiratory, and temperature instability, with adverse events reported in up to 25% of transfers26 as documented in a recent systematic review.27

Given the gap in knowledge about the regionalization of care for MLP infants, our study aimed to describe place of birth according to gestational age and other potential risk factors for neonatal morbidity and to assess how the birth setting and these risk factors influence the likelihood of neonatal transfer within the first 48 h of life. High early transfer rates are indicators of suboptimal care for infants and mothers and indicate misalignment between the infant’s healthcare needs and the available level of care at the birth hospital.

Materials and methods

Data sources and ethics approval

We performed a retrospective cohort study in MLP infants born in mainland France from January 1st, 2014, to September 30st, 2021. Data were extracted from the French National Health Data System (Système National des Données de Santé, SNDS). The database aggregates data from the hospital discharge abstract database (Programme de Médicalisation des Systèmes d’Informations), which collects diagnoses encoded using the International Classification of Diseases, 10th revision (ICD-10), and procedures performed during hospital stays using the common classification system for medical procedures (Classification commune des actes médicaux, CCAM), data from health insurance claims, and medical death certificates. In France, a specific neonatal death certificate for deaths before 28 days of life includes information about perinatal factors (such as gestational age and birthweight), in accordance with the WHO model.

In order to create our cohort of MLP infants, from birth to discharge home or death, including information on maternal characteristics and the delivery, we linked hospital discharge summaries for deliveries, births and hospital admissions within 1 year after birth. Data to confirm deaths were obtained from health insurance claims data and death certificates. Linkage was performed using unique pseudo-anonymized identification numbers, followed by the application of deterministic and probabilistic linkage methods (linkage rate of delivery and birth hospitalizations of 96.3%).28,29

This study was conducted in accordance with Article 6(1)(c) of the General Data Protection Regulation (GDPR), which allows the processing of pseudonymized data from the SNDS for reasons of public interest in the area of public health through controlled and secure access procedures. No informed consent or ethics committee approval was required, in compliance with French legislation (Loi n°78-17 du 6 janvier 1978 relative à l’informatique, aux fichiers et aux libertés, loi n° 2019-774 of 24 juillet 2019 and Décret n° 2021-848 of 29 juin 2021). This study was carried out as part of the QUALI-N study to validate indicators of mortality and morbidity in the French National Health Data System, which was approved by the Institut National de la Santé et de la Recherche Médicale (INSERM).

Population

Our study population included all infants born between 32+0 WG and 36+6 WG in all French maternity units during the study period (N = 370,333). Women in overseas territories were excluded because of differences in the organization of care (n = 26,315 infants). We excluded terminations of pregnancy (n = 3641), infants with congenital malformations (n = 22,267), stillbirths (n = 5331), and out-of-hospital births (n = 1726). In addition, infants with implausible birthweight values defined as birthweight less than or greater than four standard deviations from the mean for each GA (n = 1517). The final sample included 309,062 infants. Between 2014 and 2021, the number of Level I maternity units decreased from 229 to 175 and the number of Level IIa units from 140 to 137, while the number of Level IIb increased from 82 to 85. The number of Level III maternity units remained stable at 60.

Outcome

Our primary outcome was early neonatal transfer, defined as the transport of a newborn to a neonatal unit at a different hospital within 48 h after birth. To identify transfers, we linked hospital stays consecutive to the birth-related hospitalization for each newborn using unique identification numbers and through a combination of deterministic and probabilistic methods when direct linkage was not successful. We also used the discharge disposition variable indicating neonatal transfer. This approach allowed cleaning and validation of discharge disposition data and, consequently, of transfer status. In cases where the discharge disposition was “home”, but a transfer was identified, the infant was classified as transferred. In cases where the transfer disposition was “transfer”, but the subsequent stay could not be identified, the infant was classified as a transfer, but information was not available on the transfer hospital. We defined upward transfers as transfers within 48 h of birth to a hospital of a higher level of care or for Level III units to an equivalent level of care. Early deaths in this population are infrequent but constitute a competing risk for transfer. We therefore also included death within the first 48 h in the main outcome.

Covariables

Covariables included maternal, neonatal and organizational characteristics hypothesized to affect the probability of transfer based on previous research. Maternal and pregnancy characteristics were maternal age, multiple pregnancy, and pregnancy complications defined using ICD-10 codes, including preterm labor (O60.0–O60.1), preterm premature rupture of membranes (pPROM; O42.0–O42.2, O42.9, O75.5–O75.6, P01.1), hypertensive disorders of pregnancy (HDP; O10–O11, O13–O16, O14.0–O14.2, O14.9, O15.0–O15.9, I10–I15, P00.0), diabetes (chronic and gestational; O24.0–O24.9, E10–E11, E13–E14, P70.0–P70.2), placental abruption (O45), placenta previa (O44.0–O44.1), and antenatal FGR (O36.5). Infant characteristics were gestational age at delivery (in completed WG), sex, and small for gestational age (SGA) (<3rd percentile) using the French intrauterine growth curves.30

Statistical analysis

We first described population characteristics according to gestational age and the level of maternity care at birth. We then compared maternal and neonatal characteristics associated with early neonatal transfer, and we calculated the prevalence of early neonatal transfer across gestational ages and clinical risk groups.

To better characterize the clinical profile of infants at risk of early transfer, we categorized key pregnancy-related risk factors into four mutually exclusive groups. The first group, multiple pregnancy, included all multiple pregnancies. The second group, SGA/FGR, included singleton pregnancies with either a diagnostic code for FGR and/ or SGA status at birth. The third group, maternal pathology, comprised singleton pregnancies complicated by at least one of the following conditions: HDP (chronic or gestational), diabetes (chronic or gestational), and placenta previa. The fourth group, none/other, included singleton pregnancy with none of the above risk factors.

To facilitate visualization of heterogeneity across clinical–birth setting subgroups, we used a priori thresholds (<10%, 10%–25% and >25%) to highlight subgroups with low, intermediate, and high prevalence of neonatal transfer at each gestational age. These thresholds were selected for descriptive purposes to facilitate comparisons across gestational ages, clinical profiles, and birth settings. They do not reflect established clinical standards for acceptable transfer rates, as no such consensus exists in the literature.

We conducted two sensitivity analyzes of the main results by (1) excluding early deaths from the outcome and (2) restricting the exposure to upward transfers.

Finally, we described (1) the distribution of at-risk MLP infants born in Level I and/or IIa maternity units by gestational age (i.e., the number of at-risk MLP infants in each gestational age group divided by the total number of at-risk MLP infants born in Level I or IIa maternity units); (2) the relative size of each group (i.e., number of MLP in the group divided by total number of MLPs); and (3) each group’s contribution to transfer prevalence (i.e., number of transfers in the group divided by the total number of transfers). This study followed the Strengthening the Reporting of Observational Studies in Epidemiology reporting guidelines (Online supplementary Table S1). All analyzes were conducted using SAS Guide software.

Results

A total of 309,062 MLP infants born between 32+0 and 36+6 weeks of gestation from 2014 to 2021 were included in the study. Of these, 42,880 (13.9%) were moderate preterm (32+0–33+6 WG), and 266,182 (86.1%) were late preterm (34+0–36+6 WG). These infants represented 5.7% of the 5,406,746 total births during the study period (0.8% for moderate and 4.9% for late).

The distribution of births across the four levels of maternity care varied with gestational age. Moderate preterm births were more frequently concentrated in level III maternity units, whereas those from 34+0 to 36+6 weeks were more often born in level I and level IIa units. Specifically, only 12.7% of births at 32 WG occurred in level I and IIa units, compared with 44.9% at 36 WG (Fig. 1).

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Place of birth by gestational age.

Neonatal and maternal characteristics differed substantially depending on the level of maternity care. Level III maternity units had higher rates of multiple pregnancies (30.1%), preterm labor (74.5%), pPROM (31.1%), diabetes (17.5%), hypertensive disorders of pregnancy (18.3%), FGR (20.7%), and SGA < 3rd percentile (16.9%) compared with lower-level units (Table 1).

Table 1 Maternal and neonatal characteristics for moderate and late preterm birth by level of care
Full size table

Among infants delivered in lower-level maternity units (Levels I and IIa), a substantial proportion were classified as higher-risk, accounting for 51.9% of births at 32–33 WG and 41.8% at 34–36 WG (Online supplementary Table S2).

A total of 20,657 infants (6.7%) were either transferred to another hospital (20,198 infants; 6.5%) or died (459 infants; 0.2%) within the first 48 h of life. The prevalence of these outcomes was highest in infants born in level I units (19.0%) and decreased with higher levels of care: level IIa (9.5%), level IIb (3.5%), and level III (3.0%). Among transferred infants, 13,710 (80.0%) underwent upward transfers, while information on transfer direction was missing for 3476 infants (16.8%). The proportion of upward transfers was 99.1% for infants born in level I units, 85.1% in level IIa units, 84.5% in level IIb units, and 17.5% in level III units.

Gestational age was strongly associated with early neonatal transfer or death: 14.7% of infants born at 32 WG versus 3.4% at 36 WG. A higher prevalence was associated with several maternal and neonatal characteristics, with differences by level of care of the birth hospital, including multiple pregnancy (23.8% in level I), SGA <3rd (54.1% in level I), placenta previa (36.8% in level I), and HDP (24.7% in level I) (Table 2).

Table 2 Risk factors associated with early neonatal transfer or death within 48 h
Full size table

Table 3 shows the prevalence of early neonatal transfer or death by gestational age and clinical subgroups, stratified by level of maternity care. Prevalence was highest among births in level I maternity units, with rates exceeding 95% at 32–33 WG regardless of clinical subgroup. Among late preterm infants, a substantial proportion of infants born in level I units were still transferred, particularly those with SGA and/or FGR, with transfer rates of approximately 88.2% at 34 WG, 56.2% at 35 WG, and 33.3% at 36 WG. At 36 weeks’ gestation, transfers exceeded 10% for multiple pregnancies (11.7%) and maternal complications (14.4%). In contrast, among infants without identified risk factors (none/other group), the transfer rate was lower (6.6%).

Table 3 Prevalence of early neonatal transfer in moderate and late preterm infants by gestational age, level of care and risk factors.
Full size table

Elevated transfer rates were also observed among infants born in level IIa maternity units, although the prevalence was notably lower than in level I units. Across all clinical subgroups, transfer rates exceeded 25% at 33 WG and 10% at 34 WG. As in level I units, infants with SGA and/or FGR in level IIa units had the highest transfer rates compared to other subgroups, with prevalence exceeding 76.1% at 32 WG, 39.6% at 33 WG, 20.2% at 34 WG, and 11.1% at 35 WG. Transfer rates in level IIb maternity units were substantially lower across all gestational ages and clinical subgroups, rarely exceeding 10%, except at 32 WG. In level III units, early neonatal transfer was uncommon across all risk categories. The only notable exception was among multiple pregnancies between 32 and 34 WG, for which transfer rates ranged from 6% to 8% (Table 3).

Figure 2 shows the prevalence of transfer and early death rates by gestational age, illustrating clinical-birth setting subgroups in which transfer probabilities exceed 10% or 25%. At 32 WG, only infants born in level III had rates below 10%, regardless of clinical complication group, whereas for those born from 33 to 36 WG, this was true for infants in level III and level IIb units. This figure also illustrates the heterogeneity of risk by clinical birth setting subgroup, with infants with SGA/FGR in level I having very high transfer rates (>25%) up to 36 WG, and those with SGA/FGR or maternal pathology born in level IIa units having persistently elevated transfer rates (>10%) up to 35 WG. Sensitivity analyzes excluding early deaths and restricting to upward transfers yielded identical results (Online Supplementary Figs. S1 and S2).

Fig. 2: Prevalence of early neonatal transfers or death within 48 h among moderate and late preterm infants by gestational age, level of care, and risk factors.
Fig. 2: Prevalence of early neonatal transfers or death within 48 h among moderate and late preterm infants by gestational age, level of care, and risk factors.The alternative text for this image may have been generated using AI.
Full size image

* The category “none/other” included singleton pregnancies with no multiple pregnancy, no SGA or FGR, and no maternal pathology (hypertensive disorders of pregnancy, diabetes, and placenta previa). SGA small for gestational age, FGR fetal growth restriction.

Figure 3 (percentages provided in Online supplementary Table S3) provides the contribution of each clinical complication-birth setting group to the overall early transfer or death rate for each gestational-age week. For instance, infants at 32 WG born in level I maternity units contributed 2% to the overall transfer rate at 32 WG. If all births at 32 WG occurred in level IIa, IIb and III units, the rate would be reduced to less than 12%. If they occurred only in level IIb and level III units, the transfer rate would be 6%.

Fig. 3: The contribution of each clinical sub-group by level of care to early neonatal transfer and deaths rates for each week of gestational age.
Fig. 3: The contribution of each clinical sub-group by level of care to early neonatal transfer and deaths rates for each week of gestational age.The alternative text for this image may have been generated using AI.
Full size image

* The category “none/other” included singleton pregnancies with no multiple pregnancy, no SGA or FGR, and no maternal pathology (hypertensive disorders of pregnancy, diabetes, and placenta previa). SGA: Small for gestational age. FGR: Fetal growth restriction Groups with prevalence of transfer or death >10% are marked with one hatch; groups with prevalence >25% are marked with cross-hatching.

Discussion

This nationwide cohort study aimed to describe the place of birth according to gestational age and clinical risk factors, and to assess whether the initial level of care met the needs of MLP infants, using early neonatal transfer or death within 48 h as an indicator of suboptimal care. Overall, among MLP infants, 6.5% experienced an early neonatal transfer, while early neonatal deaths were rare (0.2%). As expected, lower gestational age was related to a higher likelihood of transfer, with transfer rates decreasing sharply from 14.7% at 32 weeks to 3.4% at 36 weeks. The highest prevalence of early neonatal transfer or death was observed in lower-level maternity units (19.0% in level I and 9.5% in level IIa). Our findings showed that the prevalence of early transfer was significantly elevated among infants classified as SGA and/or FGR born in level I and IIa units. The prevalence of transfers remained high in this group, even among late preterm infants, with transfer prevalence exceeding 25% in level I units at 36 WG and 10% in level IIa units at 35 WG.

Interpretation of findings

The prevalence of early neonatal transfer among MLP infants observed here (6.7%) is slightly higher than reports from other countries, including 4.6% in a recent U.S cohort study within the Vermont Oxford Network,31 4.0% in a retrospective cohort study from Canada,32 and 4.2% in a study from Japan.23 However, the transfer rate is lower than for very preterm infants for whom transfer rates approach 17% in France.21 The availability of more maternity units with the capacity to provide care to MLP infants contributes to low transfer rates, even among infants with known risk factors. We also observed that pregnant women with key risk factors were more likely to deliver in level IIb or level III units.

Our findings confirm that French national guidelines for the regionalization of preterm birth based on gestational age are generally followed and are appropriate for infants born at 32–33 WG, seeing as 12.7% of these births took place in level I or level IIa units and almost all of them were transferred. These transfers may partly reflect situations in which in utero transfer was not feasible (e.g., placental abruption or preterm labor with rapid progression), or complications were not detected (in the case of FGR, for instance), leading to subsequent neonatal transfers. Increasing travel distances to appropriate facilities may further complicate in utero transfer decisions: longer distances have been associated with a higher risk of neonatal transfer among very preterm infants in France.21 However, these findings also highlight the importance of improving risk stratification within gestational age groups to better align guidelines with clinical practice. For instance, the disproportionately high transfer rates from level I and IIa units for late preterm infants, even within a relatively small subset of clinically complex pregnancies, underscore the limitations of a gestational-age-based approach and highlight the need for a more individualized, risk-stratified perinatal management strategy to better capture clinical complexity and anticipate neonatal needs.

A significant proportion (41.8%) of births occurring in lower-level maternity units at later gestational ages (34–36 WG) involve higher-risk infants, despite the limited neonatal care capabilities of these units. This contributes to substantial early transfer rates, particularly among infants born at 34 WG and for infants who are SGA and have FGR. A challenge in determining the optimal place of birth for these infants is the limited performance of screening strategies for FGR. Research in France and other settings has shown that antenatal detection of growth restriction remains incomplete, with reported rates ranging from approximately 20% to 50%, depending on the definition used (e.g., <3rd or <10th percentile).33,34,35,36 Note that we used the 3rd percentile as the cutoff for severe SGA to capture pathological growth because diagnostic codes identifying FGR may be incomplete or inconsistently applied. However, infants with birthweight less than the 10th percentile may also be growth-restricted and require higher levels of care. Future research should investigate whether FGR was detected during pregnancy in these infants and, more broadly, how FGR should be defined for MLP regionalization guidelines.

We used early neonatal transfer as a proxy for the inadequacy of the birth setting, even though investigating an overall indicator of neonatal morbidity and mortality would provide more information because differences in care quality by level of care may affect the health outcomes of non-transferred infants. However, studying MLP outcomes by level of care is challenging due to limitations in the capacity to control for all potential confounders in the large administrative and medical databases needed to investigate these outcomes. Mothers with a history of adverse pregnancy outcomes or pregnancy complications are more likely to receive care in higher-level units, as seen in our study using indicators of risk available in the SNDS. Women with social risk factors are also more likely to deliver in public tertiary centers. Two recent U.S. studies suggested that high-volume level II units may have better outcomes than low-volume level III units for MLP infants,31,37 but the inability to take into consideration residual confounding may affect these outcomes. In the French SNDS, the available data does not have sufficient granularity and coding homogeneity to confidently control for preexisting risks. Nonetheless, taken together with our findings, these studies highlight the need for more research on health outcomes and MLP-specific quality indicators.38

Neonatal transfers are well established as a valid outcome measure in perinatal research and quality assessment.39 They are often needed when infants are born in facilities without the appropriate level of neonatal care, and thus reflect organizational and clinical shortcomings rather than the infant’s intrinsic risk alone.40 They may also reflect limited capacity within MLP units, as indicated by postnatal transfer rates in level III units (4.9–5.8% at 32–34 weeks). Transfers have been incorporated into composite quality indicators,15,40,41 particularly for low- and moderate-risk populations such as MLP infants, making reducing avoidable transfers a key objective from a health system perspective. High early neonatal transfer rates not only reflect inadequate matching of birth setting to infant needs but also constitute an important outcome in themselves, given their consequences for care quality and family wellbeing.23,24,25,26,27

Strengths and limitations

Our study presents previously unavailable data on place of birth and early neonatal transfer patterns in MLP infants across the entire French healthcare system using high-quality, linked administrative and clinical data. The large sample size of over 300,000 births enhances the precision and generalizability of our estimates and made it possible to stratify the analysis by gestational age, level of care, and clinical subgroup. This stratification was necessary for identifying the combinations of clinical complications and birth settings that were associated with the highest transfer rates. Furthermore, our categorization of maternal and fetal risk groups allowed for a nuanced analysis of how specific conditions may have interacted with organizational factors to influence neonatal outcomes.

Several limitations should be acknowledged. First, we lacked information on the specific reasons for the transfer, which may have been due to the newborn’s medical condition, a lack of available beds (including unit capacity constraints), or maternal rather than neonatal complications. Furthermore, some transfers may be unavoidable despite optimal planning, while others might reflect precautionary decisions. Nevertheless, early transfers of infants born in level I or IIa maternity units likely reflect immediate clinical needs after birth. Second, although the SNDS is a comprehensive national database, misclassification or underreporting of clinical variables may have occurred, which could have influenced the accuracy of our transfer rates. Third, the proportion of missing data (16.8%) regarding the level of the neonatal intensive care unit (NICU) of the receiving hospital, due unlinked hospital stays, limited the analysis of care trajectories. However, sensitivity analyzes restricted to upward transfers yielded the same results. Finally, while our analyzes relied on the French classification of maternity care levels, most high-income countries use comparable three- or four-tiered systems, suggesting that our findings are likely to be informative beyond the French context.

Conclusion

This large nationwide cohort of MLP infants shows that early neonatal transfer or death within 48 h remains relatively frequent, affecting 6.7% of births, with higher risks for some subgroups. Clinically complex subgroups such as SGA/FGR infants were more affected, with transfer rates exceeding 30% even at 36 weeks in level I units. These findings underscore the limitations of a regionalization strategy based solely on gestational age and suggest that a more individualized, risk-stratified approach to perinatal care is needed to ensure that the most at-risk infants are delivered in appropriately equipped settings, resulting in low transfer risks.