Predicting neurodevelopmental outcomes in neonatal hyperbilirubinemia: a multidimensional nomogram integrating biomarkers and neurobehavioral scores Authors: Mei Xue, Jiansong Yin, Jun Lv, Jiaping Yu, Jing Wang Published: 2026 ABSTRACT ObjectiveThe long-term neurodevelopmental trajectories of neonates with hyperbilirubinemia lacking overt acute symptoms remain largely unknown. This study aimed to evaluate these trajectories and develop a predictive model integrating multidimensional biomarkers with early neurobehavioral screenings to forecast neurodevelopmental impairment at one year of age.MethodsIn this prospective observational study of 234 term neonates, clinical characteristics were recorded at admission, and laboratory results were obtained within the first 24 h. Early Neonatal Behavioral Neurological Assessment (NBNA) was performed, and neurodevelopment was assessed at 12 months of age using the Gesell Developmental Schedules. Key predictors were selected by integrating LASSO regression results with prior knowledge and clinical significance. A multivariable logistic regression model was subsequently constructed to develop a prognostic nomogram, internally validated using bootstrap resampling.ResultsAt the 12-month follow-up, 12.8% (30/234) of the infants exhibited neurodevelopmental impairment. Multivariable analysis identified elevated C-reactive protein (CRP, OR = 1.097, 95% CI: 1.026, 1.172), a higher bilirubin-to-albumin ratio (B/A, OR = 2.587, 95% CI: 1.497, 4.471), and increased neuron-specific enolase (NSE, OR = 1.285, 95% CI: 1.092, 1.513) as independent risk factors for adverse outcomes. Higher NBNA scores (OR = 0.749, 95% CI: 0.608, 0.922) served as an independent protective factor. The model yielded an apparent area under the curve (AUC) of 0.813 and an optimism-corrected AUC of 0.797 (95% CI: 0.696–0.902), demonstrating stable discriminative power. Calibration plots showed consistency between predicted and observed probabilities, with a Brier score of 0.09. Decision curve analysis indicated clinical net benefit across a range of threshold probabilities.ConclusionNeonates with hyperbilirubinemia may experience subclinical neurodevelopmental impairment. Our model facilitates risk stratification, supporting targeted long-term follow-up and developmental monitoring for high-risk infants. Introduction In the first week of life, neonatal hyperbilirubinemia ranks among the most common clinical conditions, with roughly 60% of term and 80% of preterm infants affected (1–3). While most cases follow a self-limiting physiological course, significantly elevated levels of free unconjugated bilirubin (Bf) pose a notable risk of irreversible central nervous system damage (1, 4, 5). This lipid-soluble fraction can penetrate the immature blood-brain barrier (BBB) and causes neuronal injury (4–6). Acute Bilirubin Encephalopathy (ABE) represents the acute manifestation of severe bilirubin toxicity. Without timely intervention, ABE may progress to a chronic condition marked by lifelong disabilities such as choreoathetoid cerebral palsy or auditory neuropathy (5). To prevent such permanent injury, research has long prioritized the early identification of ABE. Total bilirubin (TBIL) levels, the bilirubin-albumin (B/A) ratio, and neuroimaging metrics are regarded as useful for early ABE diagnosis (1, 4). Recently, attention has turned to the long-term prognosis of these infants. But existing cohorts tend to have small sample sizes, with a primary focus on populations presenting overt ABE symptoms or requiring exchange transfusion (6–11). These studies emphasize severe outcomes such as mortality and cerebral palsy, while neglecting more subtle neurodevelopmental impairments in the broader group of patients with hyperbilirubinemia. Considerable uncertainty persists regarding the long-term neurodevelopmental trajectories of neonates in the “subclinical risk zone” who lack overt acute symptoms. The concept of Bilirubin-Induced Neurologic Dysfunction (BIND) and the broader Kernicterus Spectrum Disorder (KSD) underscores that bilirubin neurotoxicity operates as a continuum rather than an “all-or-nothing” phenomenon (12–14). BIND encompasses a range of subtle sequelae—developmental delay, cognitive impairment, and executive dysfunction—that can arise from jaundice less severe than that causing typical kernicterus (2, 12, 14). Notably, neuronal damage can occur even at TSB levels traditionally deemed safe (<342 μmol/L), producing dysfunction that may remain clinically silent in infancy and become apparent in later childhood (15, 16). While preliminary findings linked hyperbilirubinemia to decreased long-term functional assessment scores (17–19), there is still a lack of models in this field for predicting long-term developmental outcomes. Adopting a prospective cohort design, this research tracked neonates with hyperbilirubinemia through one year to systematically evaluate their neurodevelopmental trajectories. By integrating biochemical, inflammatory, brain injury biomarkers and early neonatal neurobehavioral assessments, we aimed to identify early predictors of neurodevelopmental impairment at 12 months in neonates with hyperbilirubinemia, and to develop a predictive model for the early identification of high-risk populations. This will facilitate the formulation of targeted follow-up plans and intervention strategies to improve long-term quality of life. Methods Study participants and data collection This prospective observational study enrolled neonates with hyperbilirubinemia admitted to the Department of Neonatology, the Second People's Hospital of Changzhou, the Third Affiliated Hospital of Nanjing Medical University, between January 2021 and December 2023. Inclusion criteria: (1) diagnosis of neonatal hyperbilirubinemia according to the standards of the Neonatology Group, Pediatric Society of the Chinese Medical Association; (2) gestational age ≥ 37 weeks; (3) birth weight ≥ 2000 g; and (4) signed informed consent from the legal guardian and agreement to complete follow-up at 1 year. Exclusion criteria: (1) congenital neurological malformations, inherited metabolic disorders, intrauterine infection, or chromosomal abnormalities; (2) hypoxic-ischemic encephalopathy (HIE), intracranial hemorrhage, or central nervous system infection; and (3) Severe infection, sepsis; (4) severely incomplete clinical data or loss to follow-up. All laboratory indicators refer to the first peripheral venous blood tests after admission (within 24 h). Neurological function assessment (NBNA) The 20-item Neonatal Behavioral Neurological Assessment (NBNA) was used to evaluate acute neurological function (20, 21). The assessment was performed as early as possible after ruling out the need for urgent intervention (usually within 72 h after admission). The NBNA scale comprises five domains: behavioral capacity (6 items), passive muscle tone (4 items), active muscle tone (4 items), primitive reflexes (3 items), and general response (3 items), with a maximum score of 40. All assessments were performed by neonatologists who received standardized training and passed inter-rater reliability testing. Evaluations were conducted while the infant was awake and quiet between two feeds. NBNA scores were used to reflect neonatal neurobehavioral status. Long-term neurodevelopmental outcome (GESELL assessment) All enrolled infants returned for neurodevelopmental follow-up at corrected (or chronological) age 12 months ± 2 weeks. The Gesell Developmental Schedules (GESELL) were administered by certified pediatricians who were blinded to group assignment and laboratory results. GESELL assesses five functional domains (22): adaptive behavior, gross motor, fine motor, language, and personal-social behavior. Developmental Quotients (DQ) for each domain were calculated as: DQ = (developmental age/corrected age) × 100. Neurodevelopmental impairment was defined as a DQ < 75 in at least one domain. Variable selection strategy Correlation analysis was first performed for all baseline data, clinical manifestations, laboratory indicators, and short-term functional scores; variables with an absolute correlation coefficient |r| > 0.7 were excluded. Collinearity diagnostics were performed for the preselected variables by calculating variance inflation factor (VIF). Severe multicollinearity was defined as VIF > 5 and highly correlated variables were removed based on clinical judgment. Remaining variables were entered into Least Absolute Shrinkage and Selection Operator (LASSO) regression for feature dimensionality reduction, retaining predictors with nonzero coefficients. To ensure stability, a 5 × 5-fold cross-validation was used to determine the optimal lambda. Variables selected by LASSO underwent a second collinearity diagnosis to confirm absence of multicollinearity. Finally, variables entering the multivariable model were comprehensively determined by integrating prior knowledge in neonatal hyperbilirubinemia and clinical accessibility. Model construction and evaluation The finalized variables were incorporated into a binary Logistic regression model to analyze the association between each factor and neurodevelopmental impairment at 1 year of age. Odds ratios (OR) and their 95% confidence intervals (CI) were calculated. Model calibration was assessed using the calibration curve. Model discrimination was evaluated by the area under the receiver operating characteristic (ROC) curve (AUC). Internal validation of the model was performed using the Bootstrap resampling method (1,000 resampling) to evaluate model stability. Statistical analysis All statistical analyses and figures were produced using R software (version 4.5.0, R Foundation for Statistical Computing, Vienna, Austria). Continuous variables were expressed as mean ± standard deviation or median (interquartile range) and compared using Student's t-test or Mann–Whitney U-test as appropriate. Categorical variables were presented as frequencies and percentages (%) and compared using the Chi-square test or Fisher's exact test. Variables with a missing rate > 20% were excluded from the analysis. Mean or median imputation was used if the missing rate was <5%; multiple imputation (mice package in R) was applied if the missing rate ranged from 5% to 20%. Results Study population and baseline characteristics A total of 234 neonates with hyperbilirubinemia were included. At the 12-month follow-up, 204 infants (87.2%) showed no neurodevelopmental impairment (all DQ ≥ 75), while 30 infants (12.8%) had neurodevelopmental impairment (at least one domain DQ < 75). There were no significant differences in gestational age, birth weight, mode of delivery, feeding method, age at jaundice onset (postnatal day), pre-admission disease duration, or ABO blood type between tow groups (P > 0.05). Compared with the normal-development group (ND group), the neurodevelopmental impairment group (NDI group) had a higher proportion of males (73.3% vs. 51.5%, P = 0.032) and a longer hospital stay (median 4.5 days vs. 4.0 days, P = 0.003). Laboratory testing showed that C-reactive protein (CRP), white blood cell count (WBC), total bilirubin (TBIL), conjugated bilirubin (CBIL), bilirubin/albumin ratio (B/A), lactate dehydrogenase (LDH), and neuron-specific enolase (NSE) were significantly higher in the NDI group than in the ND group (P < 0.05). Other hematologic, thyroid, hepatic, renal and electrolyte indices did not differ significantly between groups (P > 0.05) (Table 1). Table 1Baseline characteristics and laboratory parameters of the study population.CharacteristicND group (n = 204)NDI group (n = 30)P-valueDemographic & clinical Gestational age, weeks39.0 (38.0, 39.0)39.0 (38.0, 40.0)0.747 Birth weight, g3,350.0 (3,050.0, 3,600.0)3,330.0 (3,050.0, 3,650.0)0.749Sex, n (%)0.032 Male105 (51.5)22 (73.3) Female99 (48.5)8 (26.7)Mode of delivery, n (%)0.307 Vaginal delivery130 (63.7)22 (73.3) Cesarean section74 (36.3)8 (26.7)Feeding method, n (%)0.884 Breastfeeding48 (23.5)6 (20.0) Formula10 (4.9)2 (6.7) Mixed146 (71.6)22 (73.3) Onset age, days1.50 (1.00, 2.00)1.50 (1.00, 2.00)0.931 Length of hospital stay, days4.00 (3.00, 5.00)4.50 (3.00, 6.00)0.003Thyroid Function TSH, mIU/L7.49 (5.15, 10.81)7.43 (4.57, 10.52)0.630 FT3, pmol/L5.06 (4.23, 5.75)5.01 (4.51, 6.15)0.439 FT4, pmol/L25.15 (22.77, 27.92)25.02 (23.62, 27.40)0.883Hematology CRP, mg/L0.65 (0.50, 2.63)0.94 (0.63, 5.35)0.032 WBC, ×10⁹/L9.83 (7.91, 12.06)11.82 (9.76, 13.31)0.028 NEUT%42.60 (34.35, 54.20)44.25 (34.30, 49.10)0.977 LYM%38.85 (29.65, 48.50)36.55 (28.40, 44.30)0.484 BASO%0.40 (0.30, 0.60)0.40 (0.30, 0.70)0.876 EO%3.60 (2.70, 5.25)4.35 (2.90, 5.70)0.337 HCT, %48.00 (43.35, 52.05)47.25 (43.90, 53.10)0.809 RBC, ×10¹²/L4.71 (4.26, 5.09)4.58 (4.29, 5.21)0.859 HGB, g/L166.00 (153.00, 180.50)167.00 (151.00, 183.00)0.643 MCV, fL102.20 (99.80, 104.55)101.30 (98.80, 104.40)0.353 MCH, pg35.10 (34.30, 36.10)34.95 (34.20, 35.50)0.424 MCHC, g/L344.00 (338.00, 350.00)345.50 (338.00, 351.00)0.734 MPV, fL10.10 (9.60, 10.60)10.20 (9.60, 10.70)0.377 PLT, ×10⁹/L295.50 (247.00, 342.00)298.00 (263.00, 365.00)0.839Biochemistry TP, g/L52.10 (49.80, 55.00)53.20 (50.40, 55.10)0.643 ALB, g/L38.25 (36.70, 40.10)38.90 (36.50, 40.60)0.376 GLB, g/L13.70 (12.10, 15.25)13.70 (10.80, 15.90)0.716 A/G ratio2.80 (2.50, 3.20)2.85 (2.50, 3.60)0.469 TBIL, μmol/L242.40 (214.40, 280.90)290.15 (236.00, 339.50)<0.001 CBIL, μmol/L0.30 (0.10, 0.90)4.45 (0.30, 11.35)<0.001 B/A ratio3.76 (3.33, 4.26)4.54 (3.60, 5.19)0.002 ALT, U/L11.95 (9.00, 15.00)11.90 (9.60, 14.00)0.931 AST, U/L40.00 (31.70, 49.55)39.70 (32.00, 45.00)0.746 ALT/AST ratio0.30 (0.24, 0.39)0.30 (0.23, 0.35)0.507 ALP, U/L164.50 (137.50, 203.00)168.00 (142.00, 209.00)0.816 GGT, U/L128.00 (88.50, 185.50)124.00 (92.00, 189.00)0.972 LDH, U/L451.50 (382.00, 524.00)509.50 (418.00, 570.00)0.041 CHE, U/L5,596.00 (4,780.00, 6,722.50)5,791.50 (4,757.00, 6,814.00)0.684 TBA, μmol/L12.10 (8.80, 18.30)10.65 (8.30, 16.30)0.243 ADA, U/L13.60 (11.65, 15.30)14.45 (11.80, 15.50)0.296 BUN, mmol/L2.00 (1.20, 2.95)2.60 (1.60, 3.70)0.071 UA, μmol/L159.95 (130.10, 196.80)175.20 (147.20, 224.70)0.178 K⁺, mmol/L4.96 (4.60, 5.29)4.95 (4.68, 5.25)0.911 Na⁺, mmol/L139.30 (137.10, 140.90)139.50 (137.30, 140.70)0.734 Cl⁻, mmol/L108.05 (106.10, 110.00)107.75 (105.90, 109.80)0.703 CK, U/L196.50 (121.50, 294.50)233.50 (149.00, 291.00)0.315 CK-MB, U/L36.95 (30.95, 45.50)39.65 (32.00, 45.00)0.777 GLU, mmol/L4.23 (3.70, 4.77)4.06 (3.69, 4.84)0.762 UCr, μmol/L38.35 (33.30, 45.15)40.00 (32.60, 47.50)0.734 NSE, ng/mL34.92 (27.65, 43.76)41.69 (36.03, 76.97)<0.001TSH, thyroid-stimulating hormone; FT3, free triiodothyronine; FT4, free thyroxine; CRP, C-reactive protein; WBC, white blood cell count; NEUT%, neutrophil percentage; LYM%, lymphocyte percentage; BASO%, basophil percentage; EO%, eosinophil percentage; HCT, hematocrit; RBC, red blood cell count; HGB, hemoglobin; MCV, mean corpuscular volume; MCH, mean corpuscular hemoglobin; MCHC, mean corpuscular hemoglobin concentration; MPV, mean platelet volume; PLT, platelet count; TP, total protein; ALB, albumin; GLB, globulin; A/G, albumin/globulin ratio; TBIL, total bilirubin; CBIL, conjugated bilirubin; B/A, bilirubin/albumin ratio; ALT, alanine aminotransferase; AST, aspartate aminotransferase; ALP, alkaline phosphatase; GGT, gamma-glutamyl transferase; LDH, lactate dehydrogenase; CHE, cholinesterase; TBA, total bile acids; ADA, adenosine deaminase; BUN, blood urea nitrogen; UA, uric acid; CK, creatine kinase; CK-MB, creatine kinase myocardial band; GLU, glucose; UCr, urinary creatinine; NSE, neuron-specific enolase; NBNA, neonatal behavioral neurological assessment; Bold values indicate statistical significance (P < 0.05). Baseline characteristics and laboratory parameters of the study population. TSH, thyroid-stimulating hormone; FT3, free triiodothyronine; FT4, free thyroxine; CRP, C-reactive protein; WBC, white blood cell count; NEUT%, neutrophil percentage; LYM%, lymphocyte percentage; BASO%, basophil percentage; EO%, eosinophil percentage; HCT, hematocrit; RBC, red blood cell count; HGB, hemoglobin; MCV, mean corpuscular volume; MCH, mean corpuscular hemoglobin; MCHC, mean corpuscular hemoglobin concentration; MPV, mean platelet volume; PLT, platelet count; TP, total protein; ALB, albumin; GLB, globulin; A/G, albumin/globulin ratio; TBIL, total bilirubin; CBIL, conjugated bilirubin; B/A, bilirubin/albumin ratio; ALT, alanine aminotransferase; AST, aspartate aminotransferase; ALP, alkaline phosphatase; GGT, gamma-glutamyl transferase; LDH, lactate dehydrogenase; CHE, cholinesterase; TBA, total bile acids; ADA, adenosine deaminase; BUN, blood urea nitrogen; UA, uric acid; CK, creatine kinase; CK-MB, creatine kinase myocardial band; GLU, glucose; UCr, urinary creatinine; NSE, neuron-specific enolase; NBNA, neonatal behavioral neurological assessment; Bold values indicate statistical significance (P < 0.05). Neonatal neurobehavioral assessment There was no significant difference in the postnatal age at which the NBNA was performed between groups (P = 0.422). The NBNA total score was significantly lower in the NDI group than in the normal-development group (median 35.5 vs. 38.0, P < 0.001). On subscale analyses, scores for behavior, passive muscle tone, active muscle tone, and primitive reflexes were all significantly lower in the NDI group (P ≤ 0.005) (Table 2). Table 2Neonatal behavioral neurological assessment (NBNA) scores by neurodevelopmental outcome.CharacteristicND group (n = 204)NDI group (n = 30)P-valueNBNA assessment age, day7.00 (5.00, 9.00)7.50 (6.00, 9.00)0.422Behavior10.00 (10.00, 11.00)9.00 (9.00, 11.00)0.005Passive tone8.00 (8.00, 8.00)8.00 (7.00, 8.00)0.005Active tone7.00 (7.00, 8.00)7.00 (7.00, 7.00)0.002Primary reflexes6.00 (6.00, 6.00)6.00 (5.00, 6.00)0.004Total NBNA score38.00 (36.00, 39.00)35.50 (35.00, 37.00)<0.001Bold values indicate statistical significance (P < 0.05). Neonatal behavioral neurological assessment (NBNA) scores by neurodevelopmental outcome. Bold values indicate statistical significance (P < 0.05). Gesell developmental outcomes Gesell Developmental Scale assessment demonstrated that DQ scores in all domains were significantly lower in the NDI group compared with the ND group (P < 0.001), with the largest difference observed in the language domain (Table 3). Table 3Gesell developmental scale (GESELL) domain quotients by neurodevelopmental outcome.CharacteristicND group (n = 204)NDI group (n = 30)P-valueAdaptive behavior DQ90.50 (88.10, 95.00)78.00 (75.00, 80.00)<0.001Fine motor DQ90.30 (88.00, 95.00)73.00 (70.00, 79.00)<0.001Language DQ89.00 (85.00, 91.20)68.00 (58.00, 70.00)<0.001Personal-social DQ91.30 (88.00, 95.00)78.00 (72.00, 80.00)<0.001Gross motor DQ92.00 (89.00, 95.30)80.00 (78.00, 85.00)<0.001Bold values indicate statistical significance (P < 0.05). Gesell developmental scale (GESELL) domain quotients by neurodevelopmental outcome. Bold values indicate statistical significance (P < 0.05). Variable selection Correlation analysis identified 21 variable pairs with strong intercorrelation (|r| > 0.5, P < 0.05) (Figure 1). Using |r| > 0.7 as a threshold and guided by clinical expertise, 42 variables were preliminarily selected for collinearity diagnostics. The VIF for all variables are all less than 5. Via 5 × 5-fold cross-validation, the LASSO regression model (lambda.min) yielded six predictors with nonzero coefficients: CRP, basophil percentage (BASO%), CBIL, B/A ratio, NSE, and NBNA total score (Supplementary Table S1, Supplementary Figures S1, S2). CBIL primarily reflects hepatobiliary conjugation function, whereas the core neurotoxic moiety in neonatal hyperbilirubinemia is unconjugated (indirect) bilirubin. Because B/A ratio, which comprehensively reflects the bilirubin burden, is already included, CBIL offered no incremental predictive value, nor has prior literature supported its prognostic utility in this setting. BASO% lacks a biologically plausible link to bilirubin metabolism or neurotoxicity, has no established prognostic value in neonatal hyperbilirubinemia, and demonstrates poor measurement stability in neonates. Therefore, to preserve model parsimony and optimize stability given the limited sample size and number of NDI events, we ultimately selected CRP, B/A ratio, NSE, and NBNA total score for inclusion in the multivariable logistic regression model. Correlation heatmap of clinical variables. Only 21 variables with high correlation (∣r∣>0.5) were presented. Model development and performance The results of the multivariate logistic regression analysis are shown in Table 4. CRP (OR = 1.097; 95% CI, 1.026, 1.172; P = 0.007), B/A ratio (OR = 2.587; 95% CI, 1.497, 4.471; P = 0.001), and NSE (OR = 1.285; 95% CI, 1.092, 1.513; P = 0.003) were independent risk factors for NDI, whereas NBNA total score (OR = 0.749; 95% CI, 0.608, 0.922; P = 0.006) was an independent protective factor. Specifically, each 10 ng/mL increase in NSE was associated with a 54.4% increase in the odds of developmental impairment; for the other continuous predictors the reported ORs correspond to the change in odds per one-unit increase. Table 4Multivariable logistic regression analysis of factors associated with neurodevelopmental impairment.VariableOdds Ratio (95% CI)χ²P-valueCRP, mg/L1.097 (1.026, 1.172)7.330.007B/A ratio2.587 (1.497, 4.471)11.600.001NSE, ng/mL1.285 (1.092, 1.513)9.080.003NBNA score0.749 (0.608, 0.922)7.420.006CRP, C-reactive protein; B/A, bilirubin/albumin ratio; NSE, neuron-specific enolase; NBNA, neonatal behavioral neurological assessment. For CRP, B/A ratio, and NBNA, ORs represent the change in odds per one-unit increase; for NSE, the OR represents the change per 10 ng/mL. Multivariable logistic regression analysis of factors associated with neurodevelopmental impairment. CRP, C-reactive protein; B/A, bilirubin/albumin ratio; NSE, neuron-specific enolase; NBNA, neonatal behavioral neurological assessment. For CRP, B/A ratio, and NBNA, ORs represent the change in odds per one-unit increase; for NSE, the OR represents the change per 10 ng/mL. Internal validation was performed using 1,000 bootstrap resamples. Apparent AUC was 0.813; optimism-corrected AUC was 0.797 (95% CI, 0.696–0.902) (Table 5). The optimism-corrected calibration intercept was −0.171 (95% CI, −0.969–0.851), the calibration slope was 0.887 (95% CI, 0.445–1.419), and the Brier score was 0.090 (95% CI, 0.063–0.118). The calibration plot is presented in Figure 2. At the cutoff value determined by the Youden index, the model exhibited a sensitivity of 0.833, specificity of 0.804, accuracy of 0.808 (Supplementary Table S2). Table 5Internal validation and calibration performance of the prediction model.Performance measureOriginalOptimism-corrected95% CI (Corrected)C-index0.8,1340.7,974(0.6,961, 0.9,017) R 2 0.3,2210.2,688(0.0,715, 0.4,460)Calibration intercept0.0,000−0.1,710(–0.9,689, 0.8,508)Calibration slope1.0,0000.8,865(0.4,449, 1.4,190)Brier score0.0,8210.0,898(0.0,626, 0.1,183)Internal validation was performed using 1,000 bootstrap resamples. Internal validation and calibration performance of the prediction model. Internal validation was performed using 1,000 bootstrap resamples. Calibration curves of the multivariable logistic regression model. The bias-corrected line represents the bias-corrected calibration estimated using 1,000 bootstrap resamples, with the gray line indicating the 95% confidence intervals. Decision curve analysis (DCA) and nomogram DCA demonstrated that the prediction model yielded greater net benefit than the “treat-all” and “treat-none” strategies across a wide range of threshold probabilities (Figure 3). Decision curve analysis (DCA) for the prediction model. The blue line represents the nomogram-based prediction model. The horizontal green line (“Treat None”) assumes no patient receives intervention (net benefit of 0), and the red line (“Treat All”) assumes all patients receive intervention. A nomogram based on the logistic regression model was constructed to provide individualized risk estimates (Figure 4). For example, an infant with CRP 0.7 mg/L, B/A ratio 4.9, NSE 73 ng/mL, and NBNA total score 35 would be assigned individual point values from the top point scale: CRP ≈ 1 point, B/A ratio ≈ 64 points, NSE ≈ 35 points, NBNA ≈ 31 points. Summing these yields a total score of 131, which corresponds to a predicted risk of neurodevelopmental impairment at 1 year-old exceeding 50% on the nomogram risk scale. Nomogram for predicting the risk of neurodevelopmental impairment. For example, if a patient has a CRP of 0.7 mg/L (approximately 1 point), a B/A ratio of 4.9 (approximately 64 points), an NSE of 73 ng/mL (approximately 35 points), and an NBNA score of 35 (approximately 31 points), the total score would be 133 points. The total score corresponds to a predicted probability of approximately 0.5, indicating a 50% risk of the child having neurodevelopmental impairment at 1 year of age. Discussion Tracking the neurodevelopmental trajectories of neonates with hyperbilirubinemia, this prospective cohort study revealed that 12.8% (30/234) of infants exhibited potential neurodevelopmental impairment at 12-month (Gesell DQ < 75 in at least one domain). This suggests that neonates without overt acute brain injury may still suffer neuronal damage. Using feature selection via LASSO regression followed by multivariate logistic regression, we identified the B/A ratio, NSE, and CRP as independent risk factors for adverse outcomes, while the NBNA score served as an independent negative predictor. These findings suggest that the long-term neurodevelopmental outcomes of neonates with hyperbilirubinemia may be associated with bilirubin-related metabolic indicators, systemic inflammatory responses, and the baseline neurobehavioral functional status. For many years, clinicians have focused primarily on infants with extreme bilirubin levels or overt ABE symptoms. Under the updated KSD framework, however, BIND characterizes the subclinical or mild end of the spectrum, encompassing subtle neurological sequelae such as developmental delay and central nervous system dysfunction. Crucially, BIND may manifest within TSB levels traditionally considered safe (12, 14). Our data and previous evidence (2, 15) suggest that BIND may represent a chronic subclinical exposure that might have been overlooked in the past. The 12.8% incidence of neurodevelopmental impairment in our study population likely reflects infants within the BIND spectrum who were asymptomatic in the neonatal period but later developed minor neurological dysfunction (MND). The assessment of long-term functional scores explored the possible impact of this subclinical toxicity. We found that children with poor outcomes had lower DQ scores across all functional domains, with the language domain being the most significantly affected. This is consistent with Boskabadi et al. (18), who observed a strong positive correlation between hyperbilirubinemia severity and delays in the language and social domains at a six-month follow-up. These trends also align with data from Ke et al. (17), which showed that severe jaundice is associated with lower scores in cognitive (βadj = −0.44) and fine motor domains (βadj = −0.19$). Slower bilirubin clearance and delayed clinical intervention, as noted by Elmazzahy et al. (19), further raise the likelihood of unfavorable motor and language outcomes. These domain-specific deficits suggest the possibility that high-metabolic brain regions and auditory pathways have a selective vulnerability to bilirubin, even when the degree of injury is insufficient to cause classical cerebral palsy. Among the identified risk factors, the B/A ratio demonstrated the strongest independent association (OR = 2.587, P = 0.001). The B/A ratio serves as a more effective surrogate marker for Bf. An elevated B/A ratio indicates that more lipid-soluble Bf passes through the immature BBB and accumulate preferentially in metabolically active regions including the basal ganglia and auditory brainstem nuclei (2). This accumulation triggers a cascade of neuroinflammation, mitochondrial dysfunction, and oxidative stress, potentially inhibiting neurotransmitter synthesis and eventually contributing to neuronal membrane leakage and apoptosis (2, 6). Kang et al. (8) reported that infants in the adverse outcome group had a significantly higher B/A ratio (10.1 ± 1.7 mg/g) than those in the favorable group (7.5 ± 1.4 mg/g). Our data suggest that a higher B/A ratio may to some extent reflect the severity of subclinical damage, and such damage may subsequently manifest as multi-domain developmental delays. NSE provided potential evidence of neuronal injury (OR = 1.285 per 10 ng/mL increment, P = 0.003), with every 10 ng/mL increase in NSE associated with a 54.4% increase in the risk of neurodevelopmental impairment. As a glycolytic enzyme released during neuronal membrane leakage, elevated NSE may reflect microstructural disruption prior to the onset of clinical symptoms. Zhang et al. (23) demonstrated that NSE is not only significantly elevated in neonates with brain injury (P < 0.001) but also negatively correlates with acute neurobehavioral scores (r = −0.785). NSE levels showpromising predictive value for adverse prognosis in neonatal brain injury (AUC = 0.827). Additional evidence indicates that NSE is significantly elevated in hyperbilirubinemic neonates with overt brain injury (24). These data suggest a potential biological link between early structural damage and long-term functional outcomes. Systemic inflammation may aggravate neurotoxicity by increasing BBB permeability and triggering microglial activation (5, 25). This synergistic effect may explain why neonates with concomitant infections are more susceptible to chronic subclinical exposure, leading to subsequent MND (2, 15). Another study did not find a significant correlation between CRP as well as other inflammatory factors and adverse neurodevelopmental outcomes. A recent systematic review (2) supports a meaningful connection between neonatal CRP and pro-inflammatory cytokine levels and favorable long-term motor and neurodevelopmental courses. The NBNA score in our model likely represents the integrated neurological response to hyperbilirubinemia (OR = 0.749, P = 0.006). Infants with severe hyperbilirubinemia have been found to carry notably lower NBNA scores relative to those with milder jaundice (4). Unlike reactive microscopic laboratory indicators, the NBNA may reflect the macroscopic effects of elevated bilirubin on behavioral ability, muscle tone, and primitive reflexes (20). In this study, higher early NBNA scores were associated with better long-term neurodevelopmental outcomes. Such scores may reflect neuromodulatory capacity and early functional resilience. Early behavioral screening may provide an important functional baseline for identifying children at risk within the BIND spectrum (4, 20). The primary strengths of this study lie in its prospective cohort design and the integration of multidimensional indicators. By combining biochemical (B/A ratio), structural (NSE), inflammatory (CRP), and functional (NBNA) markers, we constructed a multidimensional predictive framework. Furthermore, the use of LASSO regression allowed for the scientific reduction of over 40 clinical variables, resulting in a preliminary and internally validated model. However, several limitations should be acknowledged to avoid over-interpretation. First, this was a single-center study with a relatively small sample size in the delay group (n = 30), and events-per-variable ratio is 7.5:1. This may lead to an overestimation of the diagnostic performance of our predictors. Second, although internal validation was performed, the absence of external validation in an independent cohort limits the generalizability of our findings. Third, we could not entirely rule out the influence of unmeasured confounders, such as genetic susceptibility. Finally, the 12-month follow-up period may be insufficient to detect subtle cognitive or behavioral disorders, such as attention-deficit/hyperactivity disorder or autism spectrum disorder, which typically manifest during school age (2, 11). Conclusion Our study suggests that the B/A ratio, NSE, CRP, and NBNA scores may serve as early predictors of neurodevelopmental impairment at 12 months of age in term neonates with hyperbilirubinemia. The nomogram developed based on these variables offers a visual and individualized tool for the early identification of high-risk infants. We suggest implementing a multimodal screening strategy in clinical practice and enrolling high-risk children in early intervention programs and long-term neurodevelopmental follow-up to ultimately enhance their long-term quality of life.