Participant Characteristics and Hematological Parameters Analysis

Table 1 presents the demographic and serological characteristics of the study participants in both NHDNH and ABO-HDN groups. A total of 213 cases were included in NHDNH 118 cases were included in ABO-HDN, and 83 cases were excluded due to suspicious hemolysis triad test results. The NHDNH group had 130 male and 83 female infants, while the ABO-HDN group had 57 male and 61 female infants (P = 0.025). The ABO blood type distribution was significantly different between the two groups, with NHDNH having 89 cases of type A, 91 cases of type B, 31 cases of type O, and 2 cases of type AB, and ABO-HDN having 74 cases of type A, 44 cases of type B, 0 cases of type O, and 0 cases of type AB (P < 0.001). The average age of infants was significantly lower in the ABO-HDN group (2.89 ± 2.23 days) than in the NHDNH group (4.83 ± 3.17 days) (P = 0.0281). In terms of hematological parameters, the ABO-HDN group had a significantly higher white blood cell count than the NHDNH group, with a higher neutrophil count and lower lymphocyte count in ABO-HDN. The ABO-HDN group also had a significantly lower RBC count than the NHDNH group, along with higher mean corpuscular volume, mean corpuscular hemoglobin, RBC distribution width-coefficient of variation, and RBC distribution width-standard deviation (P < 0.001).

Table 1 Demographic and serological characteristics of non-hemolytic disease of the newborn hyperbilirubinemia (NHDNH) and ABO hemolytic disease of the newborn (ABO-HDN) patientsRisk Factor Prediction of Hemolytic Disease

A logistic regression model was utilized to predict the risk of hemolytic disease in neonates, specifically comparing patients with NHDNH and ABO-HDN. The model incorporated eight serological difference indicators, including WBC count, RBC count, neutrophil count, lymphocyte count, MCV, MCH, RDW-CV, and RDW-SD. Gender, ABO blood type, and age were also included as covariates in the model. Among them, the WBC count, RBC count, neutrophil count, and lymphocyte count should be assigned and then analyzed. (Supplemental Table S2). A binary logistic regression univariate analysis was performed on each indicator, with a test level of P < 0.1, and WBC count and ABO blood type were excluded from the covariates. Then other results were retained for multivariate analysis to continue to adjust and eliminate covariates with P > 0.1 in the model. The last result was only RBC count, MCV, RDW-CV, and RDW-SD (Table 2). The logistic regression analysis revealed significant associations between these indicators and the risk of hemolytic disease. The regression coefficients were calculated to quantify the impact of each indicator on the risk. Incorporating these variables into the logistic regression model allowed us to effectively predict the risk of hemolytic disease in neonates. This predictive capability is crucial for early identification and implementation of appropriate intervention strategies for affected individuals. The model's performance in predicting risk factors provides valuable insights for clinical decision-making and the management of the neonatal hemolytic disease.

Table 2 Selected risk factors for multivariate binary logistic regression analysis

In our comprehensive analysis investigating the potential risk factors and diagnosis of hemolytic disease, we observed significant associations with various indicators. The results are summarized in Table 3, presenting the beta coefficients, standard errors, p-values, and OR with their respective CI. Initially, several factors including gender, age, RBC count, NEUT count, LYMPH count, MCV, MCH, RDW-CV, and RDW-SD were identified as significant predictors of hemolytic disease. However, upon adjusting for confounding factors in the multivariate analysis, only RBC count, MCV, RDW-CV, and RDW-SD remained significant predictors. Importantly, we observed that an increase of one unit in RBC count was associated with a 51.6% higher risk of hemolytic disease (OR 1.516; 95% CI 1.044–2.201). Similarly, each unit increase in MCV corresponded to a 30.6% (OR 1.306; 95% CI 1.130–1.508) higher risk. Contrariwise, each unit increase in RDW-CV exhibited a significant positive association with a substantial 261.8% increase in the risk of hemolytic disease (OR 3.618; 95% CI 1.917–6.828). In contrast, each unit increase in RDW-SD was associated with a noteworthy 28.1% decrease in the risk of hemolytic disease (OR 0.719; 95% CI 0.586–0.884). These findings underscore the importance of considering RBC count, MCV, RDW-CV, and RDW-SD as potential indicators for predicting the risk of hemolytic disease. Higher levels of RBC count and MCV are associated with increased susceptibility to hemolytic disease, while elevated RDW-CV levels and lower RDW-SD levels indicate a higher risk. ABO blood type and WBC count did not emerge as significant predictors of hemolytic disease in the multivariate analyses. In conclusion, our findings emphasize the importance of RBC count, MCV, RDW-CV, and RDW-SD, as significant risk factors for hemolytic disease [19]. These factors should be taken into consideration in the risk stratification and management of patients with this condition.

Table 3 Adjusted results of multivariate binary logistic regression analysisDevelopment of L Score

A logistic regression model was utilized to predict the likelihood (L score) of HDN based on four serological predictive factors, namely RBC count, MCV, RDW-CV, and RDW-SD. The L score can be utilized as a more straightforward and accessible tool for risk assessment and management of the HDN. \(P=\frac^}}}\)

The logistic regression equation was expressed as logit(P) = – 30.069 + 0.416X1 + 0.267X2 + 1.286X3-0.329X4. By transforming the logistic equation, the individual's predicted probability equation was obtained as P = 1/[1 + e^(-(-30.069 + 0.416X1 + 0.267X2 + 1.286X3-0.329X4))]. If the predicted probability P was greater than 0.5, the newborn was diagnosed with hemolytic disease. However, since the predicted probability cannot be directly observed, the logistic equation was transformed to simplify its calculation. Each coefficient in the model equation was divided by the smallest coefficient, resulting in the L score equation [13, 17]:

$$ } = - . + .}_} + }_} + .}_} - .}_} $$

Evaluation of the L Score and its Constituent Variables

Our study aimed to assess the predictive capabilities of the L score and its constituent variables for hemolytic disease in neonates. Table 4 and Fig. 1 present the comprehensive results of our investigation. We utilized ROC curves to evaluate the performance of different indicators. We calculated several performance parameters, including the AUC, optimal cutoff values, sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), positive predictive value (PPV), negative predictive value (NPV), and accuracy. Remarkably, the L score demonstrated the highest AUC (0.746) among all tested variables, underscoring its superior predictive performance. The optimal cutoff value for the L score (-3.0816) yielded sensitivity and specificity values of 80.5% and 60.6%, respectively. The calculated PLR (2.041) and NLR (0.322) further support the utility of the L score in identifying high-risk cases. In contrast, individual covariate variables displayed varying AUC values ranging from 0.579 to 0.661, with distinct sensitivity and specificity values. Notably, RDW-SD exhibited the second-highest AUC value (0.661), while RBC had the lowest AUC value (0.579). Our findings emphasize the practicality and accessibility of the L score as a risk assessment tool for managing hemolytic disease. It outperforms individual covariate variables in terms of predictive accuracy. The determined optimal cutoff value of the L score enables accurate identification of neonates at high risk, facilitating timely intervention and treatment.

Table 4 Comparison of AUC, optimal cut-off values, and other parameter estimates for each indicatorFig. 1

ROC curves predicting neonatal hemolytic disease for L-score and each of the raw covariates. Receiver operating characteristic (ROC) curves were generated to evaluate the predictive performance of the L-score and each raw covariate for neonatal hemolytic disease. The AUC values were calculated to quantify the predictive ability of each predictor. The Figure shows that the L-score had the highest AUC value, indicating that it was the most accurate predictor of neonatal hemolytic disease

Comparison of ROC Curves

We used Receiver Operating Characteristic (ROC) curves to evaluate and compare the performance of the L score and its constituent variables The results of the ROC curve analysis for different indicators are summarized in Table 5. We determined the statistical significance of each comparison by calculating the z-statistic and corresponding p-value. Among the indicators assessed, the RBC-L score demonstrated the highest z-statistic (5.60) and the lowest p-value (P < 0.0001), indicating its superior predictive capability for the desired outcome. Other indicators, such as the MCV-L score (z-statistics: 3.93; P = 0.0001), RDW-CV-L score (z-statistics: 2.91; P = 0.0036), and RDW-SD-L score (z-statistics: 3.2; P = 0.0014), also exhibited statistically significant differences in their ROC curves. However, comparisons between other scores did not yield statistically significant results. Our analysis revealed that the L score exhibited a statistically significant clinical diagnostic value compared to individual covariate variables [13, 20, 21].

Table 5 Comparison of ROC curves for different hemolytic disease predictorsL Score Analyzes 83 Cases Excluded Due to Suspicious Hemolysis Triad Test Results

Additionally, we utilized the collected data to diagnose the suspect groups in three trials using the L score. We assigned and substituted the RBC count, MCV, RDW-CV, and RDW-SD, as significant risk factors for hemolytic disease into the L score formula. Based on the optimal critical value (− 3.0816), a positive result was determined if the value exceeded the critical threshold, indicating the presence of hemolytic disease in the newborn. Conversely, a negative result was assigned if the value fell below the critical threshold. By substituting the results of the original 83 suspicious cases into the L score formula, we diagnosed 47 cases as hemolytic diseases of newborns. According to the positive predictive value of the L score formula, the true positive rate accounted for 53.1%, representing approximately 25 true positive results. The utilization of the L score significantly improves the detection rate of hemolytic disease in the newborn. In conclusion, our findings demonstrate that the L score, along with other relevant indicators, exhibits significant discriminatory power in distinguishing suspected cases. Utilizing the L score formula and determining the optimal critical value allowed us to make accurate diagnoses, leading to improved detection rates for hemolytic disease in newborns.