Development and internal validation of an early postnatal nomogram for bronchopulmonary dysplasia in very preterm infants with positive respiratory tract Ureaplasma urealyticum results Authors: Kai Sun, Xiaowei Sun, Ling Li, Jiayu Ren, Rui Jing, Yang Li Published: 2026 ABSTRACT AbstractBackgroundEvidence suggests that Ureaplasma urealyticum (UU) colonization is associated with the onset and progression of bronchopulmonary dysplasia (BPD) in very preterm infants. However, clinically practical tools for early postnatal risk stratification in very preterm infants with positive respiratory tract UU results remain lacking. This study aimed to develop a 72-h early postnatal prediction model for BPD in very preterm infants with positive respiratory tract UU results using variables available by the end of the first 72 h after birth.MethodsThis retrospective study included very preterm infants (gestational age < 32 + 0 weeks) with a positive UU result from respiratory tract sampling within 48 h after admission at Qilu Hospital, Shandong University, between January 2020 and December 2024. Candidate predictors were restricted to perinatal and early postnatal variables available by the end of the first 72 h after birth; therefore, the model was intended for early postnatal risk stratification rather than delivery-room or immediate-at-birth prediction. LASSO regression with tenfold cross-validation was used for variable selection. Variables with non-zero coefficients at λ1se were entered into a full multivariable logistic regression model, and a parsimonious four-variable logistic regression model was refitted for nomogram construction. Internal validation of the final four-variable model was performed using bootstrap resampling (B = 500).ResultsOf 270 infants assessed for eligibility, 260 were included in the final analysis, including 90 infants with BPD and 170 without BPD. Gestational age, endotracheal intubation, neutrophil count, and lymphocyte count were retained in the final model and incorporated into a nomogram. The model showed high apparent discrimination in the derivation cohort, with an apparent C-index of 0.950 (95% CI, 0.925–0.975) and an apparent Brier score of 0.084. Bootstrap internal validation yielded an optimism-corrected C-index of 0.945, an optimism-corrected calibration intercept of 0.010, and an optimism-corrected calibration slope of 0.875. These findings indicate good internal performance but should be interpreted cautiously because the model was developed in a single-center cohort and lacks external validation.ConclusionsWe developed a 72-h early postnatal prediction model for BPD risk stratification in very preterm infants with positive respiratory tract UU results. The model should be interpreted as a 72-h early postnatal risk-stratification tool rather than a delivery-room prediction model. The model showed promising internal performance in this single-center cohort, but external validation is required before clinical implementation.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12887-026-07175-w.Keywords: Ureaplasma urealyticum, Preterm birth, Risk factors, Predictive model Introduction Ureaplasma urealyticum (UU) is an opportunistic pathogen that colonizes the mucosal surfaces of the female urogenital tract and is commonly detected in women of reproductive age [1, 2]. Vertical transmission to the infant may occur through ascending infection, hematogenous spread, or intrapartum exposure to an infected birth canal. During pregnancy, genital UU detection, positivity, or colonization has been associated with adverse obstetric outcomes, particularly preterm birth [3, 4]. In preterm infants, UU colonization or respiratory tract UU positivity has also been associated with the development of bronchopulmonary dysplasia (BPD), possibly through persistent airway inflammation, immune dysregulation, and impaired lung development [5, 6]. BPD remains one of the most important chronic complications in very preterm infants and is associated with prolonged hospitalization, recurrent respiratory morbidity, and adverse long-term outcomes. Although several prediction models for BPD have been reported, most were developed in general preterm populations rather than in infants with a positive UU result. As a result, their applicability to this UU-positive high-risk subgroup may be limited, because they may not adequately capture the early inflammatory and clinical features associated with UU exposure. Moreover, clinically practical tools for early postnatal risk stratification in UU-positive very preterm infants remain lacking. Nomograms are widely used to provide individualized risk estimation by integrating multiple clinical predictors into an intuitive and clinically applicable format [7]. Developing an early postnatal prediction model for very preterm infants with positive respiratory tract UU results may therefore help identify infants at increased risk of BPD during the first days after birth and provide a framework for closer monitoring and individualized risk assessment. Accordingly, the aim of this study was to develop a 72-h early postnatal prediction model for BPD in very preterm infants with positive respiratory tract UU results using perinatal and early postnatal variables available by the end of the first 72 h after birth. Methods Data collection and diagnosis Qilu Hospital, Shandong University is a tertiary referral center with a Level III neonatal intensive care unit. This study included very preterm infants (gestational age < 32 + 0 weeks) with a positive Ureaplasma urealyticum result who were assessed for eligibility at Qilu Hospital, Shandong University, between January 2020 and December 2024. The inclusion criteria were as follows: (a) birth in the obstetrics department of our hospital with immediate admission to the neonatal department after birth; and (b) a positive Ureaplasma urealyticum result detected from respiratory tract sampling within 48 h after admission. The exclusion criteria were as follows: (a) outborn infants transferred from other institutions; and (b) cases with incomplete, unreliable, or unclear medical records. A total of 270 infants were assessed for eligibility, of whom 10 were excluded, including 8 with missing key variables and 2 outborn infants transferred from other institutions. The final analysis included 260 infants (Fig. 1). BPD status was assessed at 36 + 0/7 weeks’ postmenstrual age. All infants included in the final analysis had evaluable outcome data at that time point, and no included infant died or was lost before outcome ascertainment. Flow diagram of patient selection and study enrollment. A total of 270 very preterm infants with a positive Ureaplasma urealyticum result were assessed for eligibility at Qilu Hospital, Shandong University, between January 2020 and December 2024. Ten infants were excluded, including 8 with missing key variables and 2 outborn infants transferred from other institutions. The final analysis included 260 infants This investigation adhered to the principles outlined in the Declaration of Helsinki and received the endorsement of the Ethics Committee at Qilu Hospital, Shandong University (Approval No. KYLL-202506–035). Given its retrospective nature, the requirement for informed consent was waived. Ureaplasma urealyticum testing and definition of positivity Respiratory tract specimens for UU testing were collected within 48 h after admission according to the routine neonatal intensive care unit protocol. For infants who were intubated at the time of sampling, endotracheal aspirates were obtained by aspirating tracheal secretions through the endotracheal tube using sterile suction equipment. For infants who were not intubated at the time of sampling, nasopharyngeal aspirates were obtained by aspirating nasopharyngeal secretions using sterile suction equipment. The specimens were submitted to the hospital clinical microbiology laboratory for UU testing as part of routine clinical care. UU testing was performed using a quantitative nucleic acid amplification assay. Because this was a retrospective study based on routine clinical laboratory reports, detailed analytical assay characteristics, including the commercial kit name, amplification target, platform, and formal analytical limit of detection, could not be fully retrieved. A positive UU result was defined according to the qualitative “positive” or “detected” result reported by the hospital clinical microbiology laboratory, rather than by applying a post hoc external copy-number threshold. Quantitative UU DNA copy number was extracted from the original laboratory report and expressed as copies/mL. After rechecking the original laboratory records, all included infants with laboratory-reported positive UU results had reportable UU DNA copy numbers of ≥ 500 copies/mL. This value was used to verify the lower reportable range of the included positive results and should not be interpreted as a universal diagnostic cutoff for confirmed clinical UU infection or as a universal positivity threshold across assays. Previous neonatal studies have used nucleic acid amplification approaches to detect and quantify Ureaplasma DNA in respiratory tract specimens from preterm infants, but positivity criteria are assay-specific [8, 9]. In this study, a positive UU result indicated detection of UU DNA in respiratory tract specimens and was interpreted as respiratory tract UU positivity or possible colonization rather than confirmed clinical UU infection. Risk factors Comprehensive data were collected, including maternal characteristics and early clinical variables of very preterm infants. Maternal factors included maternal age, mode of delivery, antenatal corticosteroid administration, gestational hypertension, gestational diabetes mellitus, chorioamnionitis, premature rupture of membranes (PROM), duration of PROM, use of assisted reproductive technology, meconium-stained amniotic fluid, and multiple gestation. To address the concern regarding information leakage and to clarify the intended timing of model application, the predictor ascertainment window was predefined as the period from birth to 72 h after birth. The intended time point for model application was the end of the first 72 h after birth, once the relevant perinatal and early postnatal information had become available. Therefore, the model was not designed as a delivery-room or immediate-at-birth prediction tool. Accordingly, only variables available by the end of the first 72 h after birth were considered as candidate predictors for model development. Maternal variables were obtained from the obstetric record at delivery. Neonatal variables included gestational age and birth weight recorded at birth, postnatal administration of pulmonary surfactant and endotracheal intubation performed immediately after birth, and UU DNA copy number determined from respiratory tract samples collected within 48 h after admission. Hematological parameters obtained immediately after birth were also analyzed, including procalcitonin, white blood cell (WBC) count, neutrophil (NEU) count, lymphocyte (LYM) count, monocyte (MONO) count, hemoglobin (HB), platelet (PLT) count, and C-reactive protein (CRP). Early neonatal complications considered within this 72-h predictor ascertainment window included early-onset sepsis, purulent meningitis, neonatal respiratory distress syndrome (NRDS), intraventricular hemorrhage (IVH), and patent ductus arteriosus (PDA). Early-onset sepsis and purulent meningitis were recorded if diagnosed within the first 3 days after birth, whereas IVH and PDA were recorded based on ultrasonography performed on postnatal day 3. These variables were included because they were available by the end of the first 72 h after birth. BPD was defined according to the 2018 NICHD proposed criteria and operationalized in this study as a binary outcome, namely the presence or absence of any requirement for respiratory support at 36 + 0/7 weeks’ postmenstrual age [10]. Respiratory support included supplemental oxygen, high-flow nasal cannula, continuous positive airway pressure, noninvasive positive-pressure ventilation, or invasive mechanical ventilation, as applicable according to the 2018 NICHD respiratory support categories summarized in Supplementary Table S2. A formal room-air challenge test was not routinely performed; therefore, BPD status was determined based on respiratory support documented in the medical records at 36 + 0/7 weeks’ postmenstrual age. IVH was defined and graded according to the modified Papile classification [11]. In the final model, gestational age was analyzed in weeks, neutrophil count and lymphocyte count were analyzed in × 109/L, and endotracheal intubation was treated as a binary variable (yes/no) based on whether intubation was performed immediately after birth. Statistical analysis All statistical analyses were performed using R software (version 4.1.2; https://www.r-project.org/). Continuous variables were assessed for normality using the Shapiro–Wilk test and were expressed as mean ± standard deviation (SD) when normally distributed or as median (interquartile range [IQR]) when non-normally distributed. Categorical variables were presented as frequencies and percentages. Group comparisons were performed using the Student’s t test or Mann–Whitney U test for continuous variables, as appropriate, and the chi-square test or Fisher’s exact test for categorical variables. To address the concern regarding information leakage, the predictor ascertainment window was predefined as birth through 72 h after birth. The resulting model was therefore considered a 72-h early postnatal risk-stratification model, intended for use after the relevant information available by the end of the first 72 h had been ascertained, rather than for delivery-room or immediate-at-birth prediction. Only variables available within this time window were considered as candidate predictors for model development. Cases with missing key variables were excluded before model development, and the final analysis was therefore based on a complete-case dataset. The eight excluded infants had missing data for key early postnatal predictors, mainly including hematological parameters and UU DNA copy number. Because the proportion of excluded cases was small (8/270, 3.0%), complete-case analysis was considered acceptable for model development. However, the missing-data mechanism could not be formally verified, and missingness completely at random could not be assumed. No additional outlier trimming or winsorization was performed before model development. Least absolute shrinkage and selection operator (LASSO) regression was performed using the “glmnet” package to identify candidate predictors associated with BPD. Continuous candidate predictors were entered into the LASSO model in their original form and standardized internally by the glmnet procedure. The penalty parameter (λ) was selected by tenfold cross-validation, and the λ1se criterion was chosen for model development to obtain a more parsimonious and robust model. All eight variables with non-zero coefficients at λ1se were entered simultaneously into a full multivariable logistic regression model. No automated stepwise selection procedure was performed. The complete results of this eight-variable model, including regression coefficients, standard errors, Wald χ2 statistics, odds ratios, 95% confidence intervals, and P values, are provided in Supplementary Table S1. A parsimonious final four-variable logistic regression model was then refitted for nomogram construction using gestational age, endotracheal intubation, neutrophil count, and lymphocyte count. Table 2 presents the final model used for nomogram construction. Table 2Final multivariable logistic regression model for predicting bronchopulmonary dysplasiaVariableBSEWaldχ2P valueOR (95% CI)Gestational age−0.6930.13825.297< 0.0010.500 (0.382, 0.655)Endotracheal intubation1.9920.7167.7370.0057.328 (1.801, 29.822)Neutrophil0.1050.03012.624< 0.0011.111 (1.048, 1.177)Lymphocyte−0.3990.12011.0300.0010.671 (0.531, 0.849)ORs are expressed per 1-week increase in gestational age and per 1 × 10⁹/L increase in neutrophil and lymphocyte counts. Endotracheal intubation was analyzed as yes versus no according to whether intubation was performed immediately after birth. Hematological parameters were obtained immediately after birth. All eight predictors selected by LASSO at λ1se were initially entered simultaneously into a full multivariable logistic regression model, and the complete results of that model are provided in Supplementary Table S1. The four predictors shown in this table were refitted in the final model used to construct the nomogram Final multivariable logistic regression model for predicting bronchopulmonary dysplasia ORs are expressed per 1-week increase in gestational age and per 1 × 10⁹/L increase in neutrophil and lymphocyte counts. Endotracheal intubation was analyzed as yes versus no according to whether intubation was performed immediately after birth. Hematological parameters were obtained immediately after birth. All eight predictors selected by LASSO at λ1se were initially entered simultaneously into a full multivariable logistic regression model, and the complete results of that model are provided in Supplementary Table S1. The four predictors shown in this table were refitted in the final model used to construct the nomogram Because of the modest sample size, the full cohort of 260 infants was used as the derivation dataset for model development, and no separate training/validation split was performed. Model performance was evaluated in the derivation cohort by discrimination and calibration. Discrimination was assessed using the apparent concordance index (C-index) and receiver operating characteristic (ROC) analysis. Calibration was evaluated using the apparent Brier score, apparent calibration intercept and slope, and bootstrap calibration. Internal validation was performed using bootstrap resampling (B = 500) to estimate optimism in discrimination and calibration. Bootstrap internal validation was performed on the final four-variable logistic regression model used to construct the nomogram, rather than repeating the LASSO-based variable-selection process within each bootstrap sample. The calibration plot displays the apparent, bias-corrected, and ideal curves, and optimism-corrected C-index, optimism-corrected calibration intercept and slope were also calculated. Decision curve analysis (DCA) and clinical impact curve (CIC) were used to assess the potential clinical utility of the model. A two-sided P value < 0.05 was considered statistically significant. Results Baseline and early postnatal characteristics A total of 270 very preterm infants with a positive UU result were assessed for eligibility. Ten infants were excluded, including 8 with missing key predictor data and 2 outborn infants transferred from other institutions, leaving 260 infants for the complete-case analysis. Among the 260 infants included in the final analysis, 90 developed BPD and 170 did not develop BPD. The missing variables among the 8 excluded infants mainly involved hematological parameters and UU DNA copy number. No infant in the final analysis cohort died or was lost before BPD assessment at 36 weeks’ postmenstrual age. This absence of deaths or losses should be interpreted in the context of the selected single-center cohort. The patient selection process is shown in Fig. 1, and Table 1 summarizes the baseline characteristics, early postnatal interventions, and early complications of the study population. After verification of the raw data, all included infants had a gestational age of less than 32 + 0 weeks. In Table 1, gestational age is reported as decimal weeks to avoid apparent inconsistencies caused by rounding. Table 1Baseline and early postnatal characteristics of the included very preterm infants with positive Ureaplasma urealyticum results (n = 260)VariableBronchopulmonary dysplasia group (n = 90)Non-bronchopulmonary dysplasia group (n = 170)Test statisticP valueSex (male), n (%)60 (66.67)89 (52.35)4.9280.035Gestational age (weeks)28.00 (27.00, 29.00)31.50 (30.00, 31.70)−11.312< 0.001Birth weight (g)1110.28 ± 249.841847.76 ± 524.5815.337< 0.001Postnatal administration of pulmonary surfactant, n (%)85 (94.44)54 (31.76)92.922< 0.001Implementation of endotracheal intubation, n (%)87 (96.67)59 (34.71)91.757< 0.001Copy number of Ureaplasma urealyticum DNA (copies/mL)48,300.00 (13,350.00, 407,250.00)4025.00 (531.50, 16,675.00)8.756< 0.001White blood cells (× 109/L)17.50 (10.84, 32.25)10.38 (7.43, 15.17)5.926< 0.001Neutrophil (× 109/L)13.23 (7.44, 23.90)6.07 (3.98, 9.79)6.919< 0.001Lymphocyte (× 109/L)3.09 (2.15, 3.81)3.22 (2.49, 4.44)−1.9640.050Monocyte (× 109/L)1.34 (0.74, 2.06)0.62 (0.32, 1.02)6.434< 0.001Hemoglobin (g/L)152.67 ± 20.34163.79 ± 20.384.189< 0.001Blood platelet count (× 109/L)244.04 ± 79.63258.51 ± 69.781.5130.131C-reactive protein (mg/L)1.88 (0.49, 10.92)0.36 (0.00, 2.08)5.137< 0.001Procalcitonin (ng/ml)5.05 (1.42, 20.14)1.71 (0.45, 10.82)3.561< 0.001Intraventricular hemorrhage, n (%)29 (32.22)27 (15.88)9.2970.004Purulent meningitis, n (%)22 (24.44)12 (7.06)15.648< 0.001Early-onset sepsis, n (%)27 (30.00)12 (7.06)24.291< 0.001Neonatal respiratory distress syndrome, n (%)89 (98.89)90 (52.94)57.923< 0.001Patent ductus arteriosus, n (%)38 (42.22)20 (11.76)31.497< 0.001Mothers’ information Age (years)32.50 (30.75, 35.00)32.00 (30.00, 35.00)0.2740.784 Cesarean section, n (%)46 (51.11)90 (52.94)0.0790.795 Antenatal corticosteroids, n (%)83 (92.22)143 (84.12)3.4000.082 Gestational hypertension, n (%)5 (5.56)14 (8.24)0.6240.617 Gestational diabetes mellitus, n (%)12 (13.33)27 (15.88)0.3000.716 Chorioamnionitis, n (%)14 (15.56)13 (7.65)3.9550.056 Premature rupture of membranes, n (%)58 (64.44)107 (62.94)0.0570.892 Time of premature rupture of membranes (hours)48.00 (0.00, 120.00)19.50 (0.00, 120.00)0.6210.535 Multiple gestation, n (%)19 (21.11)22 (12.94)2.9570.107 Assisted reproductive technology, n (%)26 (28.89)41 (24.12)0.7000.457 Meconium-stained amniotic fluid, n (%)7 (7.78)11 (6.47)0.1560.798Intraventricular hemorrhage and patent ductus arteriosus were diagnosed by ultrasonography on postnatal day 3. Early-onset sepsis and purulent meningitis were diagnosed within the first 3 days after birth. Hematological parameters were obtained immediately after birthAll included infants had a gestational age of < 32 + 0 weeks according to the eligibility criteria. Gestational age is reported as decimal weeks to two decimal places to avoid apparent inconsistencies caused by roundingUU DNA copy number was extracted from the original clinical laboratory reports and expressed as copies/mL. After review of the original records, all laboratory-positive UU results included in the analysis had reportable copy numbers of ≥ 500 copies/mL Baseline and early postnatal characteristics of the included very preterm infants with positive Ureaplasma urealyticum results (n = 260) Intraventricular hemorrhage and patent ductus arteriosus were diagnosed by ultrasonography on postnatal day 3. Early-onset sepsis and purulent meningitis were diagnosed within the first 3 days after birth. Hematological parameters were obtained immediately after birth All included infants had a gestational age of < 32 + 0 weeks according to the eligibility criteria. Gestational age is reported as decimal weeks to two decimal places to avoid apparent inconsistencies caused by rounding UU DNA copy number was extracted from the original clinical laboratory reports and expressed as copies/mL. After review of the original records, all laboratory-positive UU results included in the analysis had reportable copy numbers of ≥ 500 copies/mL Selection of perinatal characteristics LASSO regression was performed to identify candidate predictors of BPD from variables available by the end of the first 72 h after birth, rather than from delivery-room variables alone (Fig. 2A–C). As the tuning parameter λ decreased, the coefficients of less informative variables gradually shrank toward zero, whereas the more important variables retained non-zero coefficients. Cross-validation identified λmin = 0.015 and λ1se = 0.039. We selected λ1se for model development because it provided a more parsimonious model while retaining clinically relevant predictors. Based on the LASSO results at λ1se, the original 30 candidate variables were reduced to eight predictors with non-zero coefficients, including sex, birth weight, gestational age, postnatal administration of pulmonary surfactant, endotracheal intubation, NEU, LYM, and CRP. All eight predictors were then entered simultaneously into a full multivariable logistic regression model, and no automated stepwise selection was performed. The complete results of this eight-variable model are presented in Supplementary Table S1. To construct a parsimonious and clinically interpretable nomogram, gestational age, endotracheal intubation, NEU, and LYM were retained and refitted in the final four-variable logistic regression model. The final four-variable model was selected based on clinical interpretability, model parsimony, and avoidance of redundant predictor information rather than statistical significance alone. Because gestational age and birth weight both reflect developmental maturity and are closely related, gestational age was retained as the more stable and clinically interpretable marker of prematurity for the nomogram. Sex, birth weight, postnatal surfactant administration, and C-reactive protein were not included in the final nomogram to preserve model parsimony and reduce potential redundancy or overfitting in this single-center cohort. The retained predictors represented key domains of developmental maturity, early respiratory support, and early hematological inflammatory or immune status. Table 2 presents the final four-variable model used for nomogram construction. LASSO-based predictor selection for the 72-h early postnatal bronchopulmonary dysplasia risk-stratification model. A Coefficient profiles of the 30 candidate predictors across different values of the tuning parameter λ. The upper axis indicates the number of predictors with non-zero coefficients at each λ value. B Labeled coefficient profiles highlighting the eight predictors with non-zero coefficients at λ1se: sex, birth weight, gestational age, postnatal surfactant administration, endotracheal intubation, neutrophil count, lymphocyte count, and C-reactive protein. Among these predictors, gestational age, endotracheal intubation, neutrophil count, and lymphocyte count were retained in the final nomogram. The left and right vertical dotted lines indicate λ1se and λmin, respectively, because the x-axis is − Log(λ). C Ten-fold cross-validation plot for selection of the penalty parameter λ. The left and right vertical dotted lines indicate λ1se and λmin, respectively, because the x-axis is − Log(λ). In this study, λ1se was selected for model development to obtain a more parsimonious model. The corresponding values were λmin = 0.015 and λ1se = 0.039. The complete results of the eight-variable multivariable logistic regression model are provided in Supplementary Table S1. LASSO, least absolute shrinkage and selection operator; BPD, bronchopulmonary dysplasia; λmin, value of λ that minimized the cross-validated binomial deviance; λ1se, largest value of λ within one standard error of the minimum cross-validated deviance Development and validation of the nomogram A 72-h early postnatal nomogram for BPD risk stratification was constructed using the 4 predictors retained in the final model. Although the retained predictors were generally available early after birth, the model should be interpreted within the predefined 72-h predictor ascertainment framework (Fig. 3). In the derivation cohort, the model showed high apparent discrimination, with an apparent C-index of 0.950 (95% CI, 0.925–0.975). The apparent Brier score was 0.084, with an apparent calibration intercept of 0.000 and an apparent calibration slope of 1.000. Bootstrap internal validation with 500 resamples yielded an optimism-corrected C-index of 0.945, an optimism-corrected calibration intercept of 0.010, and an optimism-corrected calibration slope of 0.875. These results suggest that the model maintained good discrimination after correction for optimism within the derivation cohort; however, because both the apparent and optimism-corrected performance estimates were derived from internal validation in a single-center cohort, they should not be interpreted as evidence of external transportability. External validation remains necessary. The internal performance and exploratory clinical utility assessment of the nomogram are presented in Fig. 4A–D, including the bootstrap calibration plot in Fig. 4B. DCA and CIC suggested potential net benefit across a range of threshold probabilities; however, these findings should be considered exploratory and supportive rather than definitive evidence of clinical usefulness because they were derived from the same single-center cohort without external validation. Therefore, no threshold probability or clinical decision strategy should be considered validated for routine use at this stage. Nomogram for individualized 72-h early postnatal risk stratification of bronchopulmonary dysplasia in very preterm infants with a positive Ureaplasma urealyticum result. The nomogram is intended for use after the relevant information available by the end of the first 72 h after birth has been ascertained, rather than in the delivery room or immediately at birth. The nomogram integrates gestational age, endotracheal intubation performed immediately after birth, neutrophil count, and lymphocyte count into a quantitative risk estimate. Gestational age is shown in weeks, and neutrophil and lymphocyte counts are shown as × 10⁹/L. This nomogram should be applied only to infants meeting the study inclusion criteria, namely very preterm infants with gestational age < 32 + 0 weeks and positive respiratory tract UU results. BPD, bronchopulmonary dysplasia Internal performance and exploratory clinical utility assessment of the 72-h early postnatal nomogram for bronchopulmonary dysplasia risk stratification in very preterm infants with a positive Ureaplasma urealyticum result. The model was intended for use after information available by the end of the first 72 h after birth had been ascertained, rather than for delivery-room or immediate-at-birth prediction. A Receiver operating characteristic (ROC) curve showing the apparent discrimination performance of the model in the derivation cohort. B Bootstrap calibration plot (B = 500) showing the apparent, bias-corrected, ideal, and confidence limit curves. C Decision curve analysis showing the nomogram model, treat-all, and treat-none strategies across threshold probabilities. D Clinical impact curve showing the number of infants classified as high risk and the number of high-risk infants with the outcome across threshold probabilities. The DCA and CIC findings should be interpreted as exploratory assessments of potential clinical utility rather than definitive evidence of clinical usefulness, because they were derived from the same single-center derivation cohort and the model has not yet been externally validated. BPD, bronchopulmonary dysplasia; ROC, receiver operating characteristic Discussion UU is a small, cell wall–deficient bacterium of the mycoplasma family and a common commensal in the human urogenital tract. Although often part of the normal flora, UU can become pathogenic under certain conditions, particularly in immunocompromised individuals such as pregnant women, neonates, and preterm infants [12]. In addition, evidence suggests that colonization with UU is associated with the onset and progression of BPD in preterm infants [13]. In this study, we developed a 72-h early postnatal prediction model for BPD risk stratification in very preterm infants with a positive Ureaplasma urealyticum result. The intended time point for model application is the end of the first 72 h after birth, once the relevant perinatal and early postnatal variables have been ascertained. Therefore, the model should not be interpreted as a delivery-room or immediate-at-birth prediction model. Gestational age, endotracheal intubation, neutrophil count, and lymphocyte count were retained in the final model. The potential added value of this nomogram lies in its ability to formalize and quantify 72-h early postnatal risk assessment in this UU-positive high-risk subgroup. Gestational age and endotracheal intubation are familiar clinical risk markers for BPD, but bedside assessment based on these factors alone remains subjective and may vary across clinicians. By integrating gestational age, endotracheal intubation, neutrophil count, and lymphocyte count into a single individualized risk estimate, the nomogram provides a reproducible framework for early risk stratification among very preterm infants with positive respiratory tract UU results. Compared with general BPD prediction models developed in broader preterm populations, the present model focuses specifically on infants with a positive UU result and incorporates early hematological indicators that may reflect host inflammatory response within the first 72 h after birth. However, its incremental value over existing BPD prediction models or routine clinical judgment has not yet been established. Therefore, at this stage, the model may help identify infants who warrant closer respiratory monitoring, more structured follow-up, and consideration for enrollment in future external validation or interventional studies, rather than directly guiding preventive interventions in routine practice. The resulting nomogram showed promising internal performance in this single-center cohort; however, external validation is required before clinical implementation. Although the model showed high apparent discrimination, this finding should be interpreted cautiously. Bootstrap internal validation yielded an optimism-corrected C-index of 0.945, suggesting that the optimism in discrimination was limited in the present cohort. Importantly, the high C-index values observed in this study should be interpreted as internal performance estimates rather than as proof that the model will perform similarly in external populations. Nevertheless, the model was developed and internally validated using the same single-center dataset, and its performance may still decline when applied to independent populations with different clinical practices, case mix, or UU testing strategies. Therefore, external validation in larger and more diverse cohorts is essential before clinical implementation. Future external validation studies should also directly evaluate the model’s incremental value over routine clinical judgment and existing BPD prediction tools before it is considered for clinical implementation. Similarly, the DCA and CIC findings should be interpreted as exploratory and hypothesis-generating. Because these analyses were performed in the same single-center derivation cohort, they provide only supportive evidence of potential clinical utility and should not be regarded as definitive evidence of clinical usefulness until the model is externally validated. These considerations underscore that the present study should be interpreted primarily as a model-development study rather than as evidence of immediate clinical implementation. In the final model, lower gestational age remained an important predictor, consistent with its established role in BPD risk. In UU-positive very preterm infants, lower gestational age may reflect greater immaturity of lung and vascular development and increased vulnerability to inflammatory and respiratory-support-related injury [13, 14]. Therefore, gestational age should be interpreted as a core developmental risk marker rather than a modifiable causal factor. Endotracheal intubation performed immediately after birth was also retained in the final model. This variable likely captures early respiratory severity and the need for invasive support during the initial postnatal period. Although invasive ventilation may contribute to lung injury, particularly in the context of UU-related inflammatory susceptibility, in this prediction model it should primarily be interpreted as an early clinical severity marker available within the predefined 72-h window [15, 16]. Neutrophil and lymphocyte counts were retained as early hematological predictors. These variables may capture the host inflammatory and immune response shortly after birth, complementing gestational age and early respiratory support in risk stratification [17, 18]. However, their role in this model should be interpreted as predictive rather than causal, and further studies are needed to determine whether these markers improve prediction across different clinical settings. Although UU DNA copy number was included as a candidate predictor, it was not retained in the final multivariable model. This suggests that, within this UU-positive cohort, microbial load alone may not add substantial predictive value beyond early clinical severity and host inflammatory response. Therefore, variation in BPD risk among infants who were already UU-positive may be better captured by early postnatal clinical and hematological features than by UU DNA burden itself. Because all infants included in this study had a positive UU result, the present model should be interpreted as a tool for risk stratification within a UU-positive subgroup, rather than as evidence that UU burden directly determines subsequent BPD. However, some limitations should be acknowledged. First, this was a single-center retrospective derivation study with a modest sample size. Although bootstrap internal validation yielded an optimism-corrected C-index of 0.945, overfitting cannot be completely excluded given the number of candidate predictors relative to the number of BPD events, and the model’s discrimination, calibration, and clinical utility may decline when applied to external cohorts. Second, this retrospective study was based on positive respiratory tract UU testing; therefore, we could not reliably distinguish UU colonization from clinically confirmed UU infection. Accordingly, the present model should be interpreted as a risk stratification tool for very preterm infants with positive respiratory tract UU results, rather than for infants with confirmed UU infection. In addition, respiratory specimen type differed according to intubation status, with endotracheal aspirates obtained from intubated infants and nasopharyngeal aspirates obtained from non-intubated infants; this difference may have influenced UU detection sensitivity and measured DNA copy number, although UU DNA copy number was not retained in the final nomogram. Third, complete-case analysis was used after excluding 8 infants with missing key predictor data. Although the proportion of excluded infants was small, the missing-data mechanism could not be formally verified, and selection bias related to missingness cannot be fully excluded. Fourth, the predefined 72-h predictor ascertainment window reflects the early postnatal clinical course rather than a strict birth-time, delivery-room, or immediate-at-birth prediction model. Therefore, the model should be interpreted as a 72-h early postnatal risk-stratification tool based on information available by the end of the first 72 h after birth. Although the final nomogram variables may be available earlier in many clinical settings, the model-development framework was based on this predefined 72-h window. Fifth, all infants included in the final analysis had evaluable BPD outcomes at 36 weeks’ postmenstrual age, and no deaths or losses occurred before outcome ascertainment. While this ensured complete outcome assessment among analyzed infants, it may also reflect the selected nature of this single-center cohort and may limit the applicability of the model to extremely unstable infants or infants who die before 36 weeks’ postmenstrual age. Sixth, although the 2018 NICHD respiratory support categories are summarized in Supplementary Table S2 to clarify the operational definition of BPD, BPD was analyzed as a binary outcome in this study and severity grades were not incorporated into model development, which may limit the clinical interpretability of the outcomes. Finally, because the model was developed in a single-center cohort and only internally validated, future studies using larger and more diverse cohorts with external validation are warranted before clinical implementation. Conclusions We developed a 72-h early postnatal prediction model for BPD risk stratification in very preterm infants with a positive Ureaplasma urealyticum result. The model is intended for use after information available by the end of the first 72 h after birth has been ascertained. External validation and assessment of incremental value over routine clinical judgment or existing BPD prediction tools are required before clinical implementation. Supplementary Information Supplementary Material 1. Acknowledgements Not applicable. Abbreviations Ureaplasma urealyticum Bronchopulmonary dysplasia Premature rupture of membranes White blood cell Neutrophil Lymphocyte Monocyte Hemoglobin Platelet C-reactive protein Intraventricular hemorrhage Neonatal respiratory distress syndrome; PDA, patent ductus arteriosus Receiver operating characteristic Decision curve analysis Clinical impact curve Authors’ contributions Conceptualisation, Yang Li; methodology, Kai Sun and Yang Li; software, Kai Sun and Ling Li; validation, Yang Li; formal analysis, Kai Sun and Rui Jing; investigation, Rui Jing; resources, Jiayu Ren; data curation, Yang Li; writing—original draft preparation, Yang Li; writing—review and editing, Xiaowei Sun; visualisation, Jiayu Ren; supervision, Xiaowei Sun; project administration, Yang Li. All authors have read and agreed to the published version of the manuscript. Funding This study was supported by the Shandong Provincial Natural Science Foundation [grant number ZR2024QH105]. Data availability The clinical data used and/or analyzed during the current study involve patient privacy and are therefore not publicly available. These data can be obtained from the corresponding author upon reasonable request and with permission from the institutional ethics committee. Declarations Ethics approval and consent to participate This investigation adhered to the principles outlined in the Declaration of Helsinki and received the endorsement of the Ethics Committee at Qilu Hospital, Shandong University (Approval No. KYLL-202506–035). Given its retrospective nature, the requirement for informed consent was waived. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Kai Sun and Xiaowei Sun contributed equally to this work and are co–first authors. References Associated Data Data Availability Statement The clinical data used and/or analyzed during the current study involve patient privacy and are therefore not publicly available. These data can be obtained from the corresponding author upon reasonable request and with permission from the institutional ethics committee.