Introduction

Neonatal sepsis is a serious infectious disease of the neonatal period, which refers to a systemic inflammatory response caused by pathogens that invade the blood of the newborn and grow, multiply, and produce toxins. Depending on the age of the infant, neonatal sepsis is divided into early-onset sepsis (EOS) and late-onset sepsis (LOS). LOS is more common in preterm infants.1 In the last 30 years, despite significant improvements in treatment technology, neonatal sepsis remains a serious and life-threatening disease in infants worldwide. Data from a study by Liu et al. showed that neonatal sepsis accounted for 15.2% of deaths in the neonatal period worldwide,1 and is the leading cause of infant mortality.2 In particular, very low birth weight infants (VLBWI) have a higher morbidity and mortality rate compared to normal birth weight infants.3 In addition, for VLBWI, death is not only the only poor outcome, but high disability rates should also be considered. Many studies have shown that short-term complications in neonates, such as brain injury,4 chronic lung disease,5 and retinopathy of prematurity (ROP),6 are closely associated with sepsis. Therefore, we propose a poor prognosis that includes not only VLBWI with death but also VLBWI with survival combined with severe complications.

The prognosis of VLBWI with LOS are influenced by various factors, including birth history, complications, genetic uniqueness, and timeliness of diagnosis and treatment. Early intervention and appropriate treatment strategies are essential to reduce sepsis mortality and disability rate among survivors, which requires early identification of the disease and accurate prediction of progression.7 The search for indicators to assess the prognosis of infants with established LOS has become a hot research topic. Many biological indicators such as procalcitonin, C-reactive protein (CRP), leukocyte count, and albumin (ALB) have been demonstrated to be indicators of poor prognosis in LOS. However, the sensitivity or specificity of these markers in determining the prognosis of LOS varies widely among studies. This limitation is primarily attributed to the complex pathophysiology of sepsis.8,9 Therefore, few indicators have relatively good sensitivity and specificity in the prognostic assessment of LOS. There is a preference for combining multiple indicators to assess the prognosis of LOS. In addition, for VLBWI with LOS, even though survival rates are gradually improving, the high rate of disability in survivors remains a concern.10,11,12 Clinicians need a comprehensive prediction that can combine individual predictors to predict the probability of poor prognosis in VLBWI with LOS.

In recent years, nomograms have been widely used as a simple statistical visualization tool to predict the occurrence and prognosis of various diseases.13,14,15 Compared to a single indicator, nomograms provide a more accurate estimate of the risk of poor prognosis in individual patients by combining multiple predictors.

We aimed to develop a clinical predictive model including clinical characteristics and biochemical indicators to determine the probability of poor prognosis more precisely in VLBWI with LOS.

Methods

Study population and grouping

Data were collected on all VLBWI with LOS admitted to the First Affiliated Hospital of Anhui Medical University, Jixi Road Campus and Gaoxin Campus between January 1, 2010 and November 1, 2022. Among these neonates, 309 eligible VLBWI with LOS were included in the study. All patients were randomized to the development cohort or the validation cohort in an 8:2 ratio. The nomogram based on the development cohort was created and tested for validity in the validation cohort.

According to the condition of the infants on discharge, we divided them into four grades: death, survival with severe complications, survival with mild complications, and survival without any of the following complication.16 Severe complications included: grade 3 or 4 intraventricular hemorrhage (IVH),17,18,19 periventricular leukomalacia (PVL),17,18,19 stage 3 or higher ROP,19 grade 3 necrotizing enterocolitis(NEC),20 or severe bronchopulmonary dysplasia (BPD).19,21,22,23,24 Many studies have noted that these disorders are associated with long-term acute and chronic outcomes and neurodevelopmental disorders, which usually require rehospitalization and ongoing treatment.17,18,19,20,21,22 Mild complications included: grade 1 or 2 IVH, stage 1 or 2 ROP, grade 2 NEC, and mild BPD.24 In general, newborns with these disorders require only post-discharge follow-up and do not leave long-term sequelae that would affect a child’s long-term prognosis. Therefore, in this study, severe complications and death were defined as poor prognoses, and mild complications and no complications were defined as good prognoses. Among them, IVH, PVL were collectively referred to as cerebral injury. For VLBWI, we routinely completed a cranial ultrasound within 3 days of birth, followed by weekly follow-up until 28 days after birth. If the child had no cerebral injury, the ultrasound was rechecked every 2–3 weeks; if the child had cerebral injury, the ultrasound was still completed weekly. All children completed a cranial magnetic resonance image (MRI) before discharge, and the diagnosis of IVH or PVL was confirmed by cranial ultrasound or cranial MRI.

Inclusion criteria and exclusion criteria

The inclusion criteria were:(1) Birth weight <1500 g; (2) Diagnosis of LOS; and (3) Hematology was performed within the first 24 h of the onset of clinical signs associated with LOS.

The exclusion criteria were: (1) Other diseases, such as congenital malformation of vital organs or genetic/metabolic diseases; (2) Diagnosis of BPD, ROP, NEC, or cerebral injury before the onset of sepsis; (3) Lack of critical clinical data, including no cranial imaging results or missing data on hematology >5%.

Definition

Based on the time of onset, neonatal sepsis is classified as early-onset sepsis (EOS, age of onset ≦ three days) and late-onset sepsis (LOS, age of onset > three days). Based on culture results, sepsis is classified as (1) culture-positive proven sepsis: positive results for one or more bacterial or fungal cultures obtained in the blood or cerebrospinal fluid of a neonate with clinical signs of infection, treated with antibiotics for five days or more or until death. Pathogens that may represent contamination (coagulase-negative staphylococci) are considered as proven sepsis only if there are laboratory signs of infection. (2) Culture-negative clinically diagnosed sepsis: with clinical infection symptoms without other reasons. For example, respiratory symptoms could be due to underlying BPD and not related to sepsis. And accompanied by any of the following, (i) ≥ two positive blood nonspecific tests, (ii) septic meningitis changes on cerebrospinal fluid examination and (iii) detection of pathogenic bacterial DNA in blood. Sepsis with clinical signs includes temperature instability, abdominal distention, vomiting, diarrhea, hepatomegaly, apnea, dyspnea, tachypnoea, retractions, flaring, grunting, cyanosis, irritability, lethargy, tremors, seizures, hyporeflexia, hypotonia, irregular respirations, full fontanel, high-pitched cry. Blood nonspecific tests include WBC count, neutrophil counts, immature to total neutrophils (I/T) ratio, PLT count, CRP, and procalcitonin.3,25

Severe BPD definition: a premature infant (<32 weeks gestational age) with BPD has persistent parenchymal lung disease, radiographic confirmation of parenchymal lung disease, and at 36 weeks PMA requires 1 of the following FiO2 ranges/oxygen levels/O2 concentrations for ≥3 consecutive days to maintain arterial oxygen saturation in the 90–95% range: Invasive intermittent positive pressure ventilation (IPPV) FiO2 concentrations>21%; nasal continuous positive airway pressure(N-CPAP), noninvasive positive pressure ventilation (NIPPV), or nasal cannula (NC) ≥ 3 L/min: FiO2 concentrations ≥30%.24 NEC definition: The diagnosis and staging of NEC were based on the Bell stage criteria.

Data collection

All the clinical data of VLBWI with LOS were collected as follows: (1) Demographics, including gestational age (GA), birth weight (BW), sex, age at onset of disease, incidence of bacterial meningitis, mechanical ventilation rate; (2) Laboratory parameters at presentation, including pH, ionized calcium (iCa), white blood cell count (WBC), neutrophil percentage (NEUT), lymphocyte percentage (LYMPH), monocyte percentage (MONO), red blood cell count (RBC), hemoglobin (HGB), hematocrit (HCT), mean red blood cell volume (MCV), mean red blood cell hemoglobin concentration (MCHC), red blood cell distribution width-SD value (RDW-SD), platelet count (PLT), plateletcrit (PCT), platelet distribution width (PDW), mean platelet volume (MPV), albumin (ALB), total bilirubin (TBIL), alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), gamma-glutamyl transferase (GGT), blood urea nitrogen (BUN), creatinine (CRE), uric acid (UA), estimated glomerular filtration rate (eGFR), serum potassium (K), serum sodium (Na); (3) Maternal factors include pregnancy-related complications, Maternal age, premature rupture of membranes (PROM), and cesarean section rate.

Statistical analysis

The SPSS 25.0 package (SPSS Inc, Chicago, IL), GraphPad Prism 9(GraphPad Software Inc, San Diego, CA), and R version (Version 4.0.3; http://www.R-project.org) were used for data processing. We randomly divided all patients into a development cohort (80%) and a validation cohort (20%). The development cohort was used to construct a nomogram and the validation cohort was used to perform internal validation. Categorical variables were described as frequency and percentage values, and differences between groups were determined using chi-square or Fisher’s exact test. The Shapiro-Wilk test is applied to continuous variables to verify that they conform to a normal distribution. Continuous variables were described as mean and standard deviation values or median and interquartile range values depending on whether the variables conformed to a normal distribution. Collinearity among all covariates was assessed using the Spearman correlation and Beasley collinearity tests. Logistic regression was used to identify independent predictors of poor prognosis in VLBWI with LOS. Predictors of poor prognosis were first identified using univariate logistic regression analysis, with P < 0.05 considered statistically significant, and then variables were screened using forward stepwise regression. The identified independent prognostic factors were again analyzed using logistic regression models, and the results were expressed as odds ratios (OR) and 95% confidence intervals (CIs). The final nomogram was constructed based on the independent predictors to predict the probability of poor prognosis in VLBWI with LOS. The nomogram is based on scaling each regression coefficient in a multiple logistic regression to 0–100 points. The scores for each variable can be summed to derive a total score, which corresponds to the predicted probability.

After model construction, we assessed the accuracy of the model using C-index and receiver operating characteristic (ROC) curve analysis. Model performance was measured by accuracy, sensitivity, specificity, positive and negative predictive values (PPV and NPV). Calibration plots were plotted using the rms package. A nomogram is considered to have excellent predictive value when the curve is close to the 45° angle.

Results

Baseline characteristics

Between 2010 and 2022, we admitted 383 VLBWI with LOS. Of these cases, 9 were excluded due to congenital disease, 49 were excluded due to comorbid BPD, NEC, and brain injury before the onset of sepsis, and 16 were excluded due to missing important information. The 309 patients were classified as 247 and 62 in the development and validation cohorts, respectively. The flow diagram was shown in Fig. 1. The median GA of the neonate in the developmental and validation cohorts was 29.2 weeks and 29 weeks, respectively. The mean BW was 1130.67 g and 1110.50 g, respectively. As expected, between development and validation cohorts, no significant differences were found in clinical characteristics, including sex, Maternal gestational factors. Similarly, there were no statistically significant differences in laboratory indicators between the two groups.

Fig. 1
figure 1

The flow diagram of this study.

Full size image

Nomogram construction

In the developmental cohort, we divided VLBWI with LOS into a good prognosis group (n = 100) and a poor prognosis group (n = 147) according to the prognosis of the neonate. In the poor prognosis group, mortality, the prevalence of severe BPD, stage 3 or higher ROP, and cerebral injury, grade 3 NEC was 19.8%,25.2%,19.7%,15.0%, and 16.3%, respectively. Tables 1 and 2 summarize the characteristics of participants in the different prognostic groups. As shown in Tables 1 and 2, there were no statistical differences between the good prognosis and poor prognosis groups in terms of sex, 5-min Apgar score, bacterial meningitis incidence and mechanical ventilation ratio, pH, WBC, NEUT, LYMPH, MONO, RBC, HGB, HCT, MCV, MCHC, PLT, MPV, TBIL, ALT, GGT, Na, Cl, Mg, hsCRP, procalcitonin, and maternal gestational factors. The BW, GA, iCa, PCT, PDW, ALB, ALP, eGFR, HCO3, TCa, and Phosphate were lower in the poor prognosis group than in the good prognosis group. The RDW-SD, AST, BUN, CRE, UA, and K were higher in the poor prognosis group than in the good prognosis group.

Table 1 Comparison of clinical characteristics between good prognosis and poor prognosis groups in the development cohort.
Full size table
Table 2 Comparison of laboratory indicators between good prognosis and poor prognosis groups in the development cohort.
Full size table

Based on the above univariate regression analysis results, we used forward stepwise regression to screen the above variables (P < 0.05). Finally, BW, GA, ALB, HCO3, BUN and iCa were identified as independent predictors for the occurrence of poor prognosis in VLBWI with LOS (Table 3). Then, we developed an individualized nomogram prediction model (Fig. 2). The nomogram was applied as follows: according to the nomogram, we could obtain the score corresponding to each predictor, the sum of the scores was recorded as the total score, and the predicted risk corresponding to the total score was the probability of poor prognosis in VLBWI with LOS. For example, a VLBWI with LOS had a BW of 850 g (69 points), GA of 25.6 weeks (74.8 points), ALB of 25.2 g/L (51.64 points), HCO3 of 18.5 mmol/L (51.3 points), BUN of 12.2 mmol/L (60.02 points), and iCa of 3.32 mg/dl (60.16 points). The cumulative score of each predictive index was 69 + 74.8 + 51.64 + 51.3 + 60.02 + 60.16 = 366.92, corresponding to the predicted risk of poor prognosis of 0.983(98.3%) (Fig. 2). Based on the above-predicted probabilities, the child was at high risk of developing a poor prognosis.

Table 3 Logistic analysis results of predictors for poor prognosis for VLBWI with late-onset sepsis in a developmental cohort.
Full size table
Fig. 2: The nomogram predicts the probability of poor prognosis in very low birth weight infants with late-onset Sepsis.
figure 2

The red dots represent the indicators of a patient in our study population and the corresponding probability of poor prognosis. *P < 0.05, **P < 0.01 and ***P < 0.001 in multivariate logistic analysis. ALB albumin, iCa ionized calcium, BW birth weight, GA gestational age, BUN blood urea nitrogen.

Full size image

Nomogram validation

Model validation was based on discrimination and calibration. The discrimination ability of the nomogram for predicting the occurrence of poor prognosis in VLBWI with LOS was assessed by the area under the curve (AUC) of receiver operating characteristic (ROC) curve analysis. We compared the AUC of the prediction model with the six independent predictors described above. As shown in Figs. 3 and 4a, b, in the developmental cohort, the AUC of ROC showed that nomogram(AUC = 0.845, 95% CI, 0.795–0.894, P < 0.05) had a rather good predictive value for poor prognosis of neonatal LOS, superior than the BW (AUC = 0.749, 95% CI, 0.687–0.810), ALB (AUC = 0.588, 95% CI,0.517-0.659), HCO3 (AUC = 0.643, 95% CI,0.574–0.713), BUN (AUC = 0.661, 95% CI, 0.594–0.729) and iCa (AUC = 0.740, 95% CI, 0.676–0.803). Compared with the observed results, the nomogram predicted a poor prognosis in the development cohort with 79.8% accuracy, 83.7% sensitivity, 74.0% specificity, 75.5% negative predictive value, and 82.5% positive predictive value (Table 4). Similarly, the AUC for the validation group was 0.869(0.764–0.973), which was also superior to the six individual variables (Fig. 4b). The prediction of poor prognosis in the validation cohort had an accuracy of 84.4%, a sensitivity of 94.0%, specificity of 77.8%, a negative predictive value of 88.0%, and a positive predictive value of 82.1% (Table 4).

Fig. 3: ROC curves of the six independent predictors (BW, GA, ALB, HCO3, iCa, BUN) in the development cohort.
figure 3

ROC receiver operating characteristic, BW birth weight, GA gestational age, ALB albumin, iCa ionized calcium, BUN blood urea nitrogen.

Full size image
Fig. 4: ROC curves in development and validation cohorts.
figure 4

ROC curves for nomogram in the development cohort (a) and validation cohort (c); ROC curves for development cohort (b) and validation cohort (d). BW, birth weight; GA, gestational age; ALB, albumin; iCa, ionized calcium; BUN, blood urea nitrogen.

Full size image
Table 4 The performance of the nomogram for poor prognosis in very low birth weight infants with late-onset sepsis.
Full size table

Calibration refers to the agreement between the observed outcome and the prediction outcome, and we used the calibration plot and Hosmer-Lemeshow to assess the calibration of the prediction model. In both the development cohort and the validation cohort, the agreement between predictions and observations in the calibration plot of the nomogram was satisfactory (Fig. 5), as both the bias-corrected curve and apparent curve deviated only slightly from the reference line. The Hosmer-Lemeshow test showed no significant deviation between observed and predicted events in the development cohort (P = 0.755) and the validation cohort (P = 0.756).

Fig. 5: Calibration plots in the development cohort and the validation cohort.
figure 5

a Development cohort; b validation cohort.

Full size image

Discussion

Neonatal LOS is one of the major causes of neonatal death. Although clinicians have many years of clinical experience in treating neonatal sepsis, challenges remain, including the lack of early assessment of the prognosis of neonates with LOS. Many studies demonstrated that sepsis was associated with the development of complications such as BPD,5 ROP,6,26 brain injury,4,27 and NEC,20 which were also strongly associated with poor long-term prognosis.17,18,19,20,21 Therefore, death is not the only poor outcome in neonates with LOS, but these severe complications should also be taken into account. However, there are no validated scales to assess the prognosis and condition of VLBWI with LOS, causing delays in precise treatment and additional distress to families. It is essential to find some sensitive and straightforward clinical variables to predict the risk of poor prognosis in septic neonates. Early and aggressive intervention is also necessary.

Due to the complex pathogenesis of sepsis, multiple indicators continue to emerge in the field of sepsis prognosis. Chew et al.28 pointed out that elevated serum heparin-binding protein was a risk factor for early death in septic patients. The view that elevated BUN/CRE (BCR) increases mortality in patients with sepsis was reported by Han et al.29 . In recent years, some studies have also proposed that CRP/ALB (CAR),30 BUN,31 and lactate32 can be independent predictors for the diagnosis of neonatal sepsis and the severity of the disease. However, some of these variables are difficult to obtain in clinical applications, and some have varying degrees of deficiencies in accuracy, sensitivity, or specificity. It is also important to emphasize the bias in clinicians’ subjective judgments of sepsis prognostic variables. All these factors can lead to inaccurate clinicians’ judgment of sepsis prognosis. The search for more convenient and accurate sepsis prognostic variables remains a current research hotspot.

Our study showed that HCO3, ALB, iCa, GA, BW, and BUN were independent risk factors for poor prognosis in VLBWI with LOS. These results were used to construct a nomogram to assess the risk of poor prognosis in VLBWI with LOS. The validity of our nomogram model was determined by the AUC, calibration plot, and Hosmer-Lemeshow test. We compared the nomogram with the six indicators mentioned above. The results showed that the predictive model was superior to individual indicators in determining the poor prognosis of VLBWI with LOS.

In the developmental cohort of this study, HCO3 was significantly lower in the poor prognosis group than in the good prognosis group. HCO3 is one of the most critical indicators of acid-base balance in neonates. Low et al.33 noted that as the severity of metabolic acidosis increases, the higher the probability of respiratory, cardiovascular, and renal-related complications in neonates. Fee et al.34 also indicated that severe acidosis was a predictor of subsequent neurological deficits. Jonsson et al.35 showed that the presence of metabolic acidosis in the neonatal period affected the long-term prognosis of the newborn. All the above studies point out that acidosis is associated with neonatal prognosis, which is consistent with our findings. However, the threshold of acidosis that causes poor prognosis in neonates varies in different studies. Moreover, there are many confounding factors of acidosis, and a single variable cannot accurately determine the prognosis of LOS.

ALB is also an independent predictor of poor prognosis in VLBWI with LOS. ALB is the most abundant protein in plasma and is synthesized only in the liver. ALB has several important physiological roles, including the maintenance of blood colloid osmotic pressure between blood vessels and tissues, binding, and participation in the transport of many small molecules, detoxification and reprocessing of metabolites, and antioxidant and radical scavenging functions.36,37 Traditionally, ALB has been viewed as an indicator of malnutrition. However, Soeters et al.38 also demonstrated a poor correlation between nutritional intake and serum albumin levels. Many studies have shown a strong link between ALB and inflammation.39,40,41,42 It has also been suggested that sepsis patients were often comorbid with hypoalbuminemia, which can lead to further exacerbation of the disease and increase the mortality rate.43,44,45 Yang et al 46 also noted that lower ALB levels in neonatal sepsis might be associated with a worse prognosis, which was consistent with our study results. There are several possible mechanisms for developing hypoalbuminemia in patients with sepsis. First, one of the pathological mechanisms contributing to morbidity and mortality in patients with sepsis is cytokine release.47 Cytokines such as TNF-α and interleukin 1 may reduce ALB levels by regulating ALB gene expression and catabolism.48,49 Second, sepsis may damage the liver through altered hemodynamics or attack on hepatocytes, which further reduces the ability of the liver to synthesize ALB.50 Third, ischemia, hypoxia, and oxidative damage often occur after severe infection, and ALB is a primary extracellular plasma protein target of oxidative stress.51 In addition to this, intravascular redistribution of ALB in critically ill patients may lead to increased capillary leakage after sepsis, which would significantly increase the risk of death.52 All of the above mechanisms could explain the results obtained in our study, with lower ALB at the onset in neonates with a poor prognosis compared to neonates with a good prognosis. Therefore, patients with decreased ALB levels should receive special attention to improve their clinical outcomes.

iCa is involved in blood coagulation, muscle contraction, neurotransmitter synthesis, and release, and plays an essential role in cell adhesion and maintenance of cell membrane function. The study by Kelly et al.53 also showed that hypocalcemia was often found in critically ill patients and that iCa levels correlated with disease severity, as confirmed in many studies. Liu et al.54 also demonstrated that hypocalcemia was common in septic neonates and was significantly associated with organ dysfunction and sepsis-related mortality. In our study, neonates in the poor prognosis group had lower iCa levels than those in the good prognosis group. Furthermore, our findings also suggested that the lower the iCa level, the higher the risk of poor prognosis in VLBWI with LOS. Therefore, early monitoring of iCa levels may help predict the progression of VLBWI with LOS.

BUN is a waste product produced by the liver. It reaches the kidneys through the blood, and the kidneys filter it out of the blood. Several studies have shown that acute kidney injury is one of the common complications of sepsis.55,56,57 Sepsis causes a significant decrease in renal blood flow and renal function, which can increase BUN levels. Our study showed higher BUN levels in the poor prognosis group than in the good prognosis group. Li et al.31 also noted a correlation between high BUN levels and the presence and severity of neonatal sepsis, which is consistent with our findings. However, it is worth noting that besides renal function, BUN levels are also influenced by various confounding factors such as a high protein diet, glucocorticoid therapy, etc.57 Therefore, clinicians need to be alert when a septic child has an elevated BUN, but it cannot be used as a single indicator to assess prognosis.

Our study also revealed that GA and BW are independent predictors of poor prognosis in VLBWI with LOS. Studies have shown that GA and BW were negatively associated with the incidence of LOS, while the lower the GA or, the lower the BW, the higher the probability of distant neurological abnormalities.58,59

Clinical prediction models are used to investigate the relationship between future outcome events and laboratory indicators at the onset, and baseline health status in patients with specific conditions. The model can integrate the results of traditional analyses and predict the probability of certain outcome events through a scoring system. LOS remains one of the leading causes of neonatal death, and LOS is strongly associated with surviving neonates leaving distant neurological sequelae. Therefore, it is crucial to develop clinical models to predict the probability of poor prognosis in neonates with LOS. Huang et al.60 already constructed a nomogram to predict the probability of sepsis in neonates. This study revealed that the lower the birth weight, the higher the probability of sepsis in newborns. Our study population was VLBWI with LOS. We used laboratory indicators at presentation, maternal status, and the birth history of these VLBWI as study variables to construct a prognostic model. Different from other studies, the model constructed in this study is the first model to predict poor prognosis in VLBWI with LOS. The model could be applied clinically to predict the high-risk groups with poor prognosis and take targeted preventive measures, such as strict aseptic operation, standardized use of ventilators, minimizing the duration of mechanical ventilation, using non-invasive ventilation modes as much as possible, and strict control of transfusion guidelines, to reduce the occurrence of BPD, IVH, NEC and other diseases. In the meantime, clinicians can use the scoring results of the model to assess patient conditions accurately, predict prognostic outcomes, communicate more effectively with family members, help them understand more clearly the severity and possible regression of their conditions, and allow them to jointly develop treatment plans to improve cooperation and maximize the prevention of poor prognosis.

The limitations and directions for the improvement of our model are as follows. Major limitations: (1) This study was a single-center study and lacked data validation from other centers; (2) Even though we attempted to make data collection more rigorous with strict inclusion and exclusion criteria, it was a retrospective study, so there may be better predictors that were not found. Also, early in this study, we determined the neonatal intracranial condition by cranial ultrasound may have reduced the positive rate of PVL. Directions for improvement: (1) Our findings should be confirmed by large-scale prospective studies; (2) More and better predictor variables should be identified and combined in the model.

Conclusion

In this study, independent predictors of poor prognosis in VLBWI with LOS were identified and used as the basis for a predictive model. The nomogram is easy to use, highly accurate and has excellent calibration. To extend the use of the prognostic model, it needs to be validated using data from a greater number of institutions.