Antenatal Prediction Model for Neonatal Intensive Care Unit Admission in Late Preterm Infants

ABSTRACT

Objective

To develop and internally validate an antenatal prediction model for neonatal intensive care unit (NICU) admission among late preterm infants using routinely available maternal and obstetric parameters.

Methods

This retrospective observational cohort study included deliveries between 34⁰/₇ and 36⁶/₇ weeks of gestation at a tertiary referral center. Maternal, obstetric, and perinatal data were extracted from electronic medical records. The primary outcome was NICU admission. Variable selection was performed using the Least Absolute Shrinkage and Selection Operator (LASSO), followed by penalized multivariable logistic regression. Model performance was assessed using Harrell's concordance index (C‐index), bootstrap internal validation (1000 resamples), calibration analysis, and decision curve analysis. A nomogram was constructed to estimate individualized NICU admission risk.

Results

A total of 2007 late preterm pregnancies were included, and neonatal NICU admission occurred in 656 pregnancies (32.7%). Independent predictors of NICU admission were lower gestational age at delivery, fetal growth restriction, twin pregnancy, cesarean delivery, and antenatal corticosteroid exposure. The model demonstrated good discrimination with an apparent C‐index of 0.752 and an optimism‐corrected C‐index of 0.747. Calibration analysis showed excellent agreement between predicted and observed risks. Decision curve analysis indicated a positive net benefit across a wide range of clinically relevant threshold probabilities. The resulting nomogram enabled individualized antenatal risk estimation.

Conclusion

An antenatal model incorporating routinely available clinical variables may help estimate the risk of NICU admission among late preterm infants; however, external validation is necessary to confirm its generalizability.

1. Introduction

Late preterm infants, defined as those born between 34⁰/₇ and 36⁶/₇ weeks of gestation, are physiologically less mature than term infants and have limited ability to adapt to extrauterine life, despite often appearing similar in size and birth weight [1, 2]. For many years, these infants were considered “near‐term” and were commonly managed according to protocols designed for full‐term neonates; however, growing evidence has demonstrated that this assumption underestimates their clinical vulnerability. Extensive cohort studies have consistently shown that late preterm infants experience significantly higher rates of respiratory morbidity, metabolic disturbances, infections, and more extended hospital stays compared with term infants, with mortality increasing as gestational age decreases [3]. Due to this increased risk profile, preterm birth has become a leading cause of neonatal intensive care unit (NICU) admissions, representing a substantial proportion of NICU hospitalizations [4]. More recent data show that maternal and obstetric factors increasingly contribute to late preterm births and are often linked to adverse short‐term neonatal outcomes, including a high rate of NICU admission, especially among infants delivered via cesarean section [5]. National data from the United States show that NICU admission rates have increased in recent years, rising from 8.7% in 2016 to 9.8% in 2023 [6]. At a global level, preterm birth remains the leading cause of neonatal mortality, affecting nearly 10% of all live births worldwide, with no measurable reduction in rates over the past decade, underscoring the sustained burden on neonatal intensive care services [7].

Late preterm infants constitute a clinically distinct and heterogeneous population that remains vulnerable despite their relatively advanced gestational age. Ongoing physiological and structural immaturity predisposes these infants to multisystem morbidity and increased utilization of neonatal intensive care services [8]. Beyond gestational age, maternal and intrapartum factors substantially influence early neonatal outcomes. In singleton late preterm cohorts, maternal diabetes, intrapartum interventions, and fetal growth restriction have been independently associated with severe neonatal morbidity and NICU admission, underscoring that gestational age alone is insufficient for risk stratification [9]. In addition, population‐based data indicate that advanced maternal age is associated with increased risks of preterm birth and low birth weight, further contributing to adverse short‐term neonatal outcomes [10]. Moreover, multicenter studies have demonstrated that maternal and obstetric characteristics at the time of presentation with threatened late preterm birth can predict prolonged NICU stay, supporting the need for individualized risk assessment rather than reliance on gestational age thresholds alone [11].

Considering the clinical heterogeneity of late preterm infants and the limitations of gestational age‐based risk assessment, there is a clear need for reliable tools to estimate the likelihood of neonatal intensive care unit admission prior to delivery. Accurate antenatal risk stratification can facilitate individualized delivery planning and informed family counseling, particularly in settings with limited access to neonatal intensive care services. Accordingly, this study aimed to develop a predictive model to estimate the risk of NICU admission among late preterm infants using readily available maternal and obstetric parameters.

2. Material and Methods

2.1. Study Design and Data Collection

This study was designed as a retrospective observational cohort study conducted at Gaziantep City Hospital, a tertiary referral center. Women who delivered between 34⁰/₇ and 36⁶/₇ weeks of gestation were included.

Ethical approval was obtained from the Gaziantep City Hospital Local Ethics Committee on May 21, 2025 (approval number 216/2025). The study was conducted in accordance with the Declaration of Helsinki.

All data were collected retrospectively from the hospital's high‐security electronic medical record system, which provides standardized and comprehensive documentation of maternal, obstetric, and neonatal clinical information. Prior to analysis, all data were anonymized, and patient confidentiality was maintained in accordance with institutional data protection policies.

2.2. Study Population, Outcome, and Variables

Between November 2023 and March 2025, a total of 2064 pregnancies resulting in delivery between 34⁰/₇ and 36⁶/₇ weeks of gestation were assessed for eligibility. After exclusion of 35 pregnancies because of missing maternal, obstetric, or perinatal data, 6 because of major fetal congenital anomalies, and 16 because of intrauterine fetal demise, a total of 2007 pregnancies were included in the final analysis (Figure 1). Eligible pregnancies had available information regarding NICU admission.

FIGURE 1.

Flowchart of patient selection and study population. Flowchart illustrating the patient selection process and the final study population included in the analysis. Pregnancies were assessed for eligibility; exclusions were applied according to the predefined criteria, and the final cohort was categorized according to NICU admission status.

Cases with missing key maternal, obstetric, or perinatal data were excluded. Pregnancies complicated by major fetal congenital anomalies or intrauterine fetal demise were also excluded. Multiple pregnancies were included to reflect real‐world clinical practice, as they constitute a substantial proportion of late preterm births.

The primary outcome of the study was NICU admission following delivery. The decision for NICU admission was made by a neonatologist based on postnatal clinical assessment of the newborn, in accordance with current national guidelines and institutional clinical protocols. The unit of analysis was the pregnancy rather than the individual neonate. Accordingly, twin pregnancies were included as a single pregnancy‐level observation. For twin pregnancies, NICU admission was considered present if at least one twin required NICU admission. Therefore, no adjustment for within‐pregnancy clustering was required.

Maternal, obstetric, and perinatal variables were extracted from electronic medical records. These included maternal age, body mass index (BMI), gestational age at delivery, gravida, parity, and abortion history, preoperative hemoglobin level, pregnancy‐related complications (preeclampsia, gestational diabetes mellitus, gestational hypertension, hypothyroidism, asthma, and intrahepatic cholestasis of pregnancy), multiple pregnancy, preterm premature rupture of membranes (PPROM), fetal growth restriction (FGR), amniotic fluid index (AFI), mode of delivery, labor induction, and antenatal corticosteroid use.

Fetal growth restriction (FGR) was defined according to the Delphi consensus criteria using the Hadlock fetal growth chart [12]. Amniotic fluid volume was assessed using the amniotic fluid index (AFI), with AFI < 5 cm defined as oligohydramnios and AFI ≥ 24 cm defined as polyhydramnios; values between these thresholds were considered normal. Pregnancies complicated by FGR, oligohydramnios, or polyhydramnios were managed according to established national guidelines and institutional clinical protocols.

Antenatal corticosteroid exposure was classified based on the interval between administration and delivery. A partial course was defined as delivery occurring within 0–12 h after the first dose, precluding completion of the full regimen. A complete course was defined as delivery occurring after a sufficient interval following the first dose, allowing completion of the antenatal corticosteroid regimen.

All variables were selected because they are clinically relevant, routinely available in obstetric practice, and suitable for use in a predictive clinical framework.

2.3. Statistical Analysis and Modeling

All statistical analyses and figure generation were performed using R software (version 4.5.0; R Foundation for Statistical Computing, Vienna, Austria; https://www.r‐project.org). The distribution of continuous variables was assessed using the Shapiro–Wilk test. Non‐normally distributed variables were summarized as median and interquartile range (IQR) and compared between groups using the Mann–Whitney U test. Categorical variables were presented as counts and percentages and analyzed using the chi‐square test. A two‐sided p‐value < 0.05 was considered statistically significant.

The development of the prediction model was conducted in two stages. In the first stage, variable selection was performed using the Least Absolute Shrinkage and Selection Operator (LASSO), a penalized regression technique that shrinks uninformative coefficients toward zero and retains only variables with meaningful predictive contribution [13]. A total of 11 candidate predictors were entered into the LASSO model. The optimal penalty parameter (lambda) was selected using 10‐fold cross‐validation, and the lambda.min value (λ = 0.001095) was used as the final tuning parameter. At this threshold, all 11 candidate predictors had nonzero coefficients and were retained for the final Penalized Maximum Likelihood Estimation (PMLE) model. In the second stage, variables selected by LASSO were used to construct the final multivariable model using PMLE. The optimal penalty value for the PMLE model was determined using the pentrace() function from the rms package, and the final model was fitted using a penalty value of 5. PMLE reduces overfitting by penalizing model complexity and stabilizing regression coefficients, and is recommended for clinical prediction modeling when external validation datasets are not available [14]. The effects of predictors in the final model were reported as adjusted odds ratios (OR) with 95% confidence intervals (CI), derived from the penalized regression coefficients.

Model performance was evaluated in terms of discrimination and calibration. Discrimination was quantified using Harrell's concordance index (C‐index). Both the apparent C‐index, calculated directly from the original dataset, and the optimism‐corrected C‐index, obtained through 1000 bootstrap resamples, were reported. Bootstrap internal validation is recommended to reduce performance optimism in clinical prediction models [14].

Calibration was assessed using a bias‐corrected calibration curve generated with the calibrate() function from the rms package (B = 1000). Calibration accuracy was further quantified using the mean absolute error (MAE), mean squared error (MSE), and the 0.9 quantile of absolute error.

A nomogram was constructed based on the final PMLE model to allow individualized estimation of NICU admission risk. To explore the potential clinical usefulness of the model, Decision Curve Analysis (DCA) was conducted using the rmda package. Net benefit was compared with “treat‐all” and “treat‐none” strategies across a range of clinically relevant threshold probabilities. Because DCA was conducted on the same dataset used for model development, results were interpreted cautiously.

3. Results

3.1. Maternal and Perinatal Characteristics

Maternal and perinatal characteristics of the study population stratified by NICU admission status are presented in Table 1. A total of 2007 late preterm pregnancies were included in the analysis. NICU admission occurred in 656 (32.7%) pregnancies. Maternal age, gravida, parity, abortion history, and preoperative hemoglobin levels were comparable between the NICU and non‐NICU groups (all p > 0.05). Maternal body mass index was significantly higher among women whose neonates were admitted to the NICU (median 28.40 vs. 27.68 kg/m², p = 0.002).

TABLE 1.

Maternal and perinatal characteristics by NICU admission.

Variable	No NICU (n = 1351)	NICU (n = 656)	p‐value
Maternal age (years)	26 [22–31]	26 [22–32]	0.060
BMI (kg/m2)	27.68 [25.39–30.85]	28.40 [25.71–31.22]	0.002
Gestational age at delivery (weeks)	36.29 [35.79–36.57]	35.57 [34.71–36.29]	< 0.001
Gravida	3 [2–4]	3 [2–4]	0.854
Parity	1 [1–3]	1 [0–3]	0.632
Abortion history	0 [0–0]	0 [0–0]	0.134
Preoperative hemoglobin (g/dL)	11.7 [10.6–12.6]	11.8 [10.7–12.8]	0.059
Preeclampsia (PE)	26 (1.9%)	38 (5.8%)	< 0.001
Gestational diabetes (GDM)	43 (3.2%)	36 (5.5%)	0.018
Gestational hypertension (GHT)	60 (4.4%)	47 (7.2%)	0.015
Hypothyroidism	18 (1.3%)	19 (2.9%)	0.023
Asthma	10 (0.7%)	2 (0.3%)	0.357
Cholestasis	12 (0.9%)	10 (1.5%)	0.291
Multiple pregnancy (twin)	118 (8.7%)	134 (20.4%)	< 0.001
Amniyotic Fluid Index (AFI)
Normal	1066 (78.9%)	506 (77.1%)	0.632
Oligohydramnios	203 (15.0%)	109 (16.6%)
Polyhydramnios	82 (6.1%)	41 (6.2%)
PPROM	288 (21.3%)	146 (22.3%)	0.674
Fetal Growth Restriction (FGR)	139 (10.3%)	110 (16.8%)	< 0.001
Mode of delivery (C/S)	767 (56.8%)	524 (79.9%)	< 0.001
Labor induction	203 (15.0%)	51 (7.8%)	< 0.001
Antenatal corticosteroid use			< 0.001
Complete course	1113 (82.4%)	398 (60.7%)
Partial course	238 (17.6%)	258 (39.3%)

Note: Data are presented as median [interquartile range] for continuous variables and number (percentage) for categorical variables. NICU refers to the neonatal intensive care unit. PPROM indicates preterm premature rupture of membranes. BMI denotes body mass index, GDM gestational diabetes mellitus, GHT gestational hypertension, and C/S cesarean section. Antenatal corticosteroid use is categorized as complete or partial course. Continuous variables were compared using the Mann–Whitney U test, while categorical variables were analyzed using the chi‐square test. Bold p‐values indicate statistical significance (p < 0.05).

Gestational age at delivery was significantly lower in the NICU group compared with the non‐NICU group (35.57 vs. 36.29 weeks, p < 0.001). The prevalence of pregnancy‐related comorbidities, including preeclampsia (5.8% vs. 1.9%, p < 0.001), gestational diabetes mellitus (5.5% vs. 3.2%, p = 0.018), gestational hypertension (7.2% vs. 4.4%, p = 0.015), and hypothyroidism (2.9% vs. 1.3%, p = 0.023), was significantly higher among NICU‐admitted neonates. Rates of asthma and cholestasis did not differ significantly between the groups.

Multiple pregnancy was more common in the NICU group (20.4% vs. 8.7%, p < 0.001). Amniotic fluid index categories and the incidence of PPROM were similar between groups, whereas fetal growth restriction was significantly more frequent among neonates admitted to the NICU (16.8% vs. 10.3%, p < 0.001). Cesarean delivery was performed more frequently in the NICU group (79.9% vs. 56.8%, p < 0.001), while labor induction was less common (7.8% vs. 15.0%, p < 0.001). Regarding antenatal corticosteroid exposure, partial courses were more frequent among NICU‐admitted neonates compared with those not requiring NICU care (39.3% vs. 17.6%, p < 0.001).

3.2. Multivariable Analysis and Nomogram Development

Independent predictors of NICU admission were evaluated using penalized multivariable logistic regression, as shown in Table 2. After adjustment for relevant covariates, lower gestational age at delivery was a strong independent predictor of NICU admission (OR 0.40, 95% CI 0.34–0.46, p < 0.001).

TABLE 2.

Penalized multivariable logistic regression for NICU admission.

Variable	β	SE	OR	95% CI	p‐value
Gestational age	−0.925	0.073	0.40	0.34–0.46	< 0.001
BMI	0.013	0.013	1.01	0.99–1.04	0.294
Preeclampsia	0.504	0.278	1.66	0.96–2.85	0.070
GDM	0.317	0.240	1.37	0.86–2.20	0.186
GHT	0.222	0.233	1.25	0.79–1.70	0.340
Hypothyroidism	0.519	0.327	1.68	0.89–3.19	0.112
PPROM	0.197	0.131	1.22	0.94–1.57	0.134
FGR	0.418	0.151	1.52	1.13–1.97	0.006
Twin pregnancy	0.474	0.151	1.61	1.19–2.04	0.002
Cesarean Delivery	0.924	0.126	2.52	1.96–3.23	< 0.001
ACS	0.295	0.122	1.34	1.06–1.70	0.016

Note: ACS: antenatal corticosteroid course; odds ratio represents partial course compared with complete course (reference category). Statistically significant p‐values (< 0.05) are shown in bold.

Abbreviations: β: regression coefficient; ACS: antenatal corticosteroid course (Complete/Partial); BMI: body mass index; CI: confidence interval; FGR: fetal growth restriction; GA: gestational age; GDM: gestational diabetes mellitus; GHT: gestational hypertension; NICU: neonatal intensive care unit; OR: odds ratio; PMLE: penalized maximum likelihood estimation; PPROM: preterm premature rupture of membranes; SE: standard error.

Fetal growth restriction (OR 1.52, 95% CI 1.13–1.97, p = 0.006), twin pregnancy (OR 1.61, 95% CI 1.19–2.04, p = 0.002), and cesarean delivery (OR 2.52, 95% CI 1.96–3.23, p < 0.001) were independently associated with increased odds of NICU admission. In addition, antenatal corticosteroid exposure was significantly associated with NICU admission (OR 1.34, 95% CI 1.06–1.70, p = 0.016). Maternal BMI, preeclampsia, gestational diabetes mellitus, gestational hypertension, hypothyroidism, and PPROM did not retain statistical significance in the multivariable model.

Based on the final multivariable model, a nomogram was constructed to estimate individualized risk of NICU admission (Figure 2). Each predictor was assigned a weighted point value proportional to its regression coefficient, and the total score corresponded to a predicted probability of NICU admission. The final PMLE model had an intercept of 29.947, and the corresponding regression coefficients were used to construct the nomogram.

FIGURE 2.

Nomogram predicting the probability of NICU admission in late preterm infants. The nomogram integrates routinely available antenatal maternal and obstetric variables to estimate the individualized risk of NICU admission. Total points derived from each variable correspond to the predicted probability shown on the bottom scale.

3.3. Model Performance and Internal Validation

Internal validation was performed using 1000 bootstrap resamples to evaluate model discrimination, calibration, and clinical utility. The apparent Harrell's C‐index was 0.752, indicating good discriminative ability. After correction for optimism, the bias‐corrected C‐index remained high at 0.747.

Calibration performance was assessed both graphically and quantitatively. The calibration plot (Figure 3) demonstrated excellent agreement between predicted and observed probabilities of NICU admission across the entire range of risk estimates, with the bias‐corrected curve closely overlapping the ideal 45‐degree reference line. Quantitative calibration metrics further supported these findings, with a mean absolute error of 0.018, a mean squared error of 0.00042, and a 0.9 quantile of absolute error of 0.028. The bootstrap‐corrected calibration slope was 0.998 and the calibration intercept was 0.0002, indicating excellent model calibration.

FIGURE 3.

Calibration plot of the nomogram for predicting NICU admission. The calibration plot illustrates the agreement between predicted and observed probabilities of neonatal intensive care unit (NICU) admission. The dashed line represents the ideal reference line indicating perfect calibration. The dotted line shows the apparent calibration of the model, while the solid line represents the bias‐corrected calibration obtained through internal validation. The light gray lines indicate the confidence limits (C.L.). The distribution of predicted probabilities is shown by the rug plot at the top of the figure.

Clinical utility of the prediction model was evaluated using decision curve analysis (Figure 4). Across a wide range of threshold probabilities, approximately between 0.05 and 0.65, the model demonstrated a positive net benefit compared with both the “treat‐all” and “treat‐none” strategies. The highest relative net benefit was observed within the threshold probability range of approximately 0.10–0.30.

FIGURE 4.

Decision curve analysis of the nomogram for predicting NICU admission. Decision curve analysis demonstrates the clinical usefulness of the NICU prediction model across a range of threshold probabilities. The red line represents the net benefit of the proposed NICU model, while the gray line indicates the strategy of treating all infants and the black horizontal line represents treating none. The model shows a higher net benefit than both default strategies across a wide range of clinically relevant threshold probabilities. The lower x‐axis displays the corresponding cost–benefit ratios.

4. Discussion

Late preterm infants remain a clinically heterogeneous group in whom gestational age alone does not fully capture the risk of early neonatal instability and subsequent need for intensive care [15, 16]. Population‐based and health‐systems data have shown that NICU admission among late preterm births is common and is influenced by both neonatal physiology and maternal and obstetric risk factors and local care pathways [17, 18]. This supports the view that gestational age alone is an imperfect proxy for risk in late preterm birth and that maternal conditions, fetal status, and perinatal management pathways together contribute to NICU utilization [17]. In our tertiary referral cohort of 2007 late preterm pregnancies, nearly one‐third of neonates (32.7%) required NICU admission, which highlights the burden of late preterm birth on neonatal services [19]. To address the need for antenatal risk stratification, we developed an internally validated prediction model using routinely available maternal and obstetric parameters to estimate the risk of NICU admission among late preterm infants [20]. The model showed good discrimination and excellent calibration, suggesting that within our population the predicted probabilities aligned closely with observed NICU admission rates [20, 21]. Decision curve analysis also supported potential clinical usefulness across a broad range of threshold probabilities [22].

Lower gestational age at delivery emerged as the strongest independent predictor of NICU admission, consistent with prior work demonstrating that even within the late preterm period, small differences in maturity can affect respiratory transition, feeding, thermoregulation, and glucose stability, which may prompt admission for observation or treatment [15, 23, 24, 25]. These findings reinforce that late preterm birth should not be treated as a single risk category, because risk changes meaningfully week by week [23]. Consistent with this, prior work has described a gestational age–dependent gradient in short‐term morbidity within the late preterm window, with higher respiratory and infectious morbidity among infants born at earlier weeks [23, 25].

Fetal growth restriction was independently associated with NICU admission. Growth restriction likely reflects reduced physiologic reserve and vulnerability to hypoglycemia, hypothermia, and respiratory compromise, which increases the likelihood of admission for monitoring and treatment [26]. Fetal growth restriction may reflect placental dysfunction and reduced physiologic reserve, which can increase vulnerability to hypoglycemia, temperature instability, and difficulties with postnatal transition. Gardosi and colleagues reported that twin‐specific growth charts can better identify small‐for‐gestational‐age–associated perinatal risk in twins, whereas applying singleton standards may misclassify risk [27]. This is relevant for model transportability, because differences in local growth references and FGR definitions may affect calibration across settings. In twin pregnancies, the interpretation of growth status is more complex because classification of small for gestational age and growth restriction can vary depending on the growth standard used. Recent work on customized growth assessment in twins highlights how SGA/FGR identification and associated perinatal risk can vary by the growth standard used [27]. This supports careful framing of FGR as a clinically meaningful risk marker, while acknowledging that measurement and classification approaches may affect transportability across settings.

Twin pregnancy itself also remained an independent predictor. Multiple gestations are common in tertiary referral centers and are more often complicated by earlier delivery, growth issues, and the need for closer neonatal surveillance; therefore, higher NICU utilization is expected [25, 27]. This aligns with literature showing an elevated risk of adverse short‐term outcomes among late preterm infants [28, 29, 30] and with emerging evidence that risk profiling should be tailored to growth and risk patterns specific to multiple pregnancies [27, 31].

Cesarean delivery showed a strong association with NICU admission. This may reflect both causal and noncausal pathways: cesarean delivery can be linked to transient respiratory morbidity and early adaptation problems, which may prompt admission [32]. At the same time, cesarean delivery in the late preterm period is often selected in higher‐acuity clinical contexts, so the association can also reflect confounding by indication [32, 33]. Studies examining prenatal predictors of NICU admission and respiratory distress similarly show how delivery context and pathway selection can shape NICU admission rates [32]. Accordingly, cesarean delivery may be interpreted primarily as a marker of clinical pathway selection and underlying indication rather than as a purely modifiable independent cause [17]. Because delivery mode and antenatal corticosteroid exposure are influenced by clinical decision‐making and underlying acuity, it is important to define the intended point of use for the model and to interpret these predictors primarily as markers of care pathways [21]. This association likely reflects, at least in part, confounding by indication, whereby cesarean delivery is more frequently performed in pregnancies already considered to be at increased risk of adverse neonatal outcomes rather than representing a direct causal effect of the delivery mode itself.

Antenatal corticosteroid exposure was independently associated with NICU admission in our multivariable model. This finding should be interpreted cautiously because in observational cohorts, steroids are often given when clinicians anticipate higher risk or limited time before delivery, so steroid exposure can act as a marker of imminent delivery and clinical concern rather than a direct cause of NICU admission [17, 20]. Studies evaluating neonatal outcomes in late preterm pregnancies similarly emphasize that maternal and obstetric risk clustering, gestational age, mode of delivery, and time‐critical deliveries influence neonatal outcomes and NICU admission, highlighting the importance of interpreting antenatal corticosteroid exposure within the broader clinical context rather than in isolation [34, 35, 36, 37]. The association may therefore capture residual acuity and care‐pathway intensity rather than a detrimental effect of corticosteroids themselves. Therefore, the observed association should not be interpreted as a causal effect of antenatal corticosteroid administration but rather as a reflection of confounding by indication inherent to observational studies.

Our findings align with and extend prior efforts to create antenatal tools for identifying late preterm infants at risk of NICU admission [38]. Notably, a maternal–obstetric risk score developed specifically for late preterm NICU admission identified similar high‐yield antenatal factors and emphasized practical bedside use. Similar to that risk score, our model suggests that a small number of routinely available antenatal variables can provide useful risk stratification. At the same time, differences in baseline NICU admission rates and admission thresholds between centers make external validation and, where necessary, recalibration essential before broader use [38]. More recent nomogram approaches in selected cohorts such as older maternal age groups also show that prenatal prediction is feasible, although differences in inclusion criteria and case mix limit direct comparisons [39]. Together, these studies support that routinely available antenatal variables can provide clinically useful risk stratification, while external validation and recalibration remain essential [20, 21].

From a clinical perspective, the model offers a pragmatic approach to antenatal risk stratification for late preterm pregnancies using routinely available variables [36]. Such an approach may be particularly valuable in referral systems and in settings with constrained NICU capacity, where identifying higher‐risk pregnancies can support referral planning, allocation of perinatal resources, and early involvement of neonatal teams [18, 19]. By facilitating individualized risk estimation before delivery, the model may also contribute to informed family counseling and multidisciplinary perinatal decision‐making [15]. In clinical practice, the nomogram is intended to be applied during antenatal assessment after the diagnosis of late preterm delivery and before birth, when key maternal and obstetric variables are available. For example, in a woman presenting with threatened late preterm birth, the estimated risk may assist clinicians in antenatal counseling, planning the place and mode of delivery, considering referral to a tertiary care center with NICU facilities when appropriate, and anticipating neonatal resource requirements.

Several maternal comorbidities were more frequent among NICU‐admitted neonates in unadjusted comparisons, but they did not retain independent significance after penalized multivariable modeling. This pattern is consistent with the idea that maternal conditions may exert their neonatal impact indirectly through downstream pathways—earlier gestational age at delivery, impaired fetal growth, or clinical decisions around timing and mode of delivery—rather than acting as independent predictors once those downstream factors are included [17, 31]. Population‐based studies of NICU admission among preterm newborns also show that maternal factors can appear influential in crude comparisons but may attenuate when obstetric course and fetal condition are accounted for [31]. Maternal age may behave similarly: its association with neonatal outcomes often operates through downstream pregnancy complications and case‐mix, so it may not emerge as a stable independent predictor across cohorts [39, 40].

The strengths of this study include the focus on a well‐defined late preterm population, a large cohort with a substantial event rate, and use of internal validation with comprehensive assessment of discrimination, calibration, and clinical utility. The model relies on routinely available antenatal information and therefore has potential for implementation in referral counseling and perinatal resource planning. However, the study has several limitations. First, the single‐center retrospective design limits generalizability, particularly because NICU admission is influenced by local admission criteria, staffing models, and clinical thresholds. Second, predictors such as cesarean delivery and corticosteroid exposure are partially influenced by clinical decision‐making and underlying acuity. Third, external validation is necessary before broader adoption, and recalibration may be necessary in settings with different baseline NICU admission rates or different practice patterns.

Because NICU admission can be practice dependent, future studies should also evaluate prediction of more objective neonatal morbidity outcomes, such as respiratory support, hypoglycemia requiring intravenous glucose therapy, sepsis treatment, and prolonged length of stay. Postnatal models that incorporate early neonatal variables may further refine prognostication after admission and can be viewed as complementary to antenatal tools rather than alternatives [41]. In addition, subgroup‐focused models may also be useful in distinct pathways such as prelabor rupture of membranes. Prior work in PPROM populations supports the value of tailored risk tools when the underlying clinical course differs [41]. External validation across diverse healthcare settings and prospective evaluation of clinical impact on referral patterns, resource utilization, and neonatal outcomes will be important next steps.

In conclusion, late preterm infants represent a clinically vulnerable population with a substantial risk of NICU admission. In this study, an internally validated antenatal prediction model based on routinely available maternal and obstetric parameters demonstrated good discrimination, calibration, and decision‐analytic evidence of potential clinical utility for estimating NICU admission risk among late preterm infants. Lower gestational age, fetal growth restriction, twin pregnancy, cesarean delivery, and antenatal corticosteroid exposure were the primary independent predictors. With external validation and refinement for varying clinical workflows, this tool may support individualized delivery planning, perinatal resource allocation, and timely referral to centers with NICU capabilities.

Author Contributions

E.Y. and İ.T. conceived and designed the study. E.Y., F.D.Y.Y., N.M., and C.B. collected the data. S.Ö. was responsible for neonatal data acquisition and interpretation. İ.T. performed the statistical analysis and interpreted the results. S.S. and S.Sın. provided clinical supervision and methodological oversight. E.Y. and İ.T. drafted the manuscript, and all authors critically revised the manuscript for important intellectual content, approved the final version, and agree to be accountable for all aspects of the work.

Funding

The authors have nothing to report.

Disclosure

The authors have nothing to report.

Ethics Statement

This study was approved by the Gaziantep City Hospital Local Ethics Committee (Approval No. 216/2025; May 21, 2025). The study was conducted in accordance with the ethical principles of the Declaration of Helsinki.

Consent

The requirement for informed consent was waived by the Gaziantep City Hospital Local Ethics Committee because of the retrospective nature of the study.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study consist of de‐identified individual clinical data obtained retrospectively from hospital medical records. Public sharing of the dataset is restricted because the data were collected as part of routine clinical care and used for research under institutional ethics committee approval. De‐identified data may be made available by the corresponding author upon reasonable request, subject to applicable institutional and ethics committee requirements.