A Novel Classification and Regression Tree-Driven Decision Tree Combining Neutrophil-to-Lymphocyte Ratio and C-reactive Protein for Early Prognostication of Severe Acute Pancreatitis: A Prospective Vietnamese Cohort Study.

Acute pancreatitis (AP) is an inflammatory disorder of the pancreas, varying in severity from mild, self-limiting episodes to severe acute pancreatitis (SAP), accounting for 10%–30%, and increases mortality risks ( 1 – 3 ). Early prediction of SAP is critical to guide timely triage and reduce adverse outcomes. Traditional scoring systems such as Ranson Criteria, the Acute Physiology and Chronic Health Evaluation (APACHE), and the Computed Tomography Severity Index are reliable but complex, relying heavily on clinical and imaging parameters. This makes them less applicable in resource-constrained environments ( 4 – 6 ). The Bedside Index for Severity in Acute Pancreatitis (BISAP), developed in 2008, is a more straightforward and more accessible tool that uses 5 clinical markers—blood urea nitrogen, mental status, systemic inflammatory response syndrome (SIRS), age, and pleural effusion—to predict severity within 24 hours of admission ( 7 ). Although effective and introduced in guidelines for stratification, its reliance on imaging and detailed laboratory tests poses challenges in high patient volume, resource-limited environments, necessitating a more straightforward and sensitive method ( 1 , 8 – 14 ). There is growing interest in using routine complete blood count (CBC)-derived biomarkers, such as the neutrophil-to-lymphocyte ratio (NLR) and C-reactive protein (CRP), as simpler alternatives for assessing AP severity ( 5 , 6 , 15 , 16 ). NLR has gained attention because of its ability to reflect both inflammatory responses and immune regulation ( 17 ). Elevated NLR levels have been linked to greater inflammation and disease severity in various conditions, including AP ( 17 – 22 ). Similarly, CRP, a well-established acute-phase reactant, is another reliable marker for systemic inflammation and disease severity ( 5 , 23 , 24 ). Combining NLR and CRP has demonstrated synergistic benefits, offering improved accuracy in predicting AP severity ( 6 , 15 ). Although previous studies have shown promising results, most evidence comes from retrospective studies in China, with limited data from Southeast Asian populations, including Vietnam ( 15 , 25 ). Classification and regression tree (CART) analysis is a nonparametric statistical method that uses binary recursive partitioning to create decision trees. This intuitive approach is easy to interpret and practical for bedside stratification ( 26 , 27 ). Several studies have demonstrated the effectiveness of CART in developing accurate prognostic models for AP ( 7 , 28 ). We hypothesize that combining NLR and CRP in the CART model will outperform BISAP in predicting SAP, providing a more accessible tool for early identification in resource-limited settings. This study aims to evaluate the prognostic value of NLR and CRP in SAP using CART analysis and compare its performance with BISAP.

The derivation (n = 210) and validation (n = 130) cohorts were comparable in demographic and clinical characteristics, supporting the robustness of subsequent model analyses. The median age was 45.0 years in the derivation cohort and 46.0 years in the validation cohort ( P = 0.401), with male patients accounting for 67.7% and 70.0%, respectively ( P = 0.552). No significant differences were observed in the etiologies of AP, including biliary, alcohol-related, and hypertriglyceridemia-induced cases ( P = 0.519). Comorbidities such as hypertension, diabetes mellitus, and SIRS were similarly distributed across both cohorts. Supplementary Figure 1, http://links.lww.com/CTG/B375 , illustrates the distribution of NLR and CRP values, demonstrating markedly elevated and more variable levels in patients with SAP compared with those with nonsevere AP (Table 1 ). Demographics in derivation cohort and validation cohort (n = 340) BISAP, Bedside Index for Severity in Acute Pancreatitis; BUN, blood urea nitrogen; CRP, C-reactive protein; CTSI, Computed Tomography Severity Index; SIRS, systemic inflammatory response syndrome; WBC, white blood cell. Data were shown as a-median (interquartile range), but Hct variable was shown as mean ± SD. n (%). To evaluate the relationship between individual clinical parameters and disease severity, we performed a correlation analysis involving BISAP, NLR, and CRP. A moderate correlation was observed between NLR and CRP ( r = 0.359, P < 0.001). Subgroup analysis showed weaker correlations in the severe group ( r = 0.173) compared with nonsevere cases ( r = 0.233). BISAP correlated positively with both NLR ( r = 0.512) and CRP ( r = 0.530), with all correlations reaching statistical significance ( P < 0.001) (Figure 2 ). Correlation matrix illustrating relationships between neutrophil-to-lymphocyte ratio (NLR), C-reactive protein (CRP), and Bedside Index of Severity in Acute Pancreatitis (BISAP) scores using Spearman rank correlation analysis. Density plots represent distributions stratified by severity (nonsevere vs severe acute pancreatitis). Correlation coefficients (Corr) are indicated, with significance denoted as follows: *** P < 0.001. A decision tree model was constructed using CART analysis to classify patients with AP according to their risk of developing SAP, based on admission levels of NLR and CRP, which are depicted in Figure 3 , with a simplified version for practical application in Figure 4 . Classification and regression tree (CART)-derived decision tree for predicting severe acute pancreatitis (SAP) based on optimal cutoff values for neutrophil-to-lymphocyte ratio (NLR) and C-reactive protein (CRP) in the derivation cohort. Numbers and percentages indicate the proportion of patients developing SAP within each subgroup. NLR and CRP thresholds are provided at each decision node. Simplified decision tree derived from CART analysis for stratifying the risk of severe acute pancreatitis based on admission values of neutrophil-to-lymphocyte ratio (NLR) and C-reactive protein (CRP). Percentages represent the proportion of SAP cases within each subgroup. CART, classification and regression tree; SAP, severe acute pancreatitis. In the derivation cohort (n = 210), patients were first split based on NLR. Among those with NLR <11.4 (n = 130), 6.1% (8 patients) developed SAP, indicating low risk. By contrast, patients with NLR ≥11.4 (n = 80) were further stratified by CRP values. Those with CRP <173.3 (n = 32) had a 25.0% SAP rate, while those with CRP ≥173.3 (n = 48) showed a markedly higher SAP incidence of 81.3%. At the first decision node, NLR ≥11.4 showed a sensitivity of 81.3% and specificity of 92.5% in the derivation cohort, with a misclassification rate of 13.8%. The second decision node, based on CRP ≥173.3, demonstrated a sensitivity of 88.0% and specificity of 91.5%, with a misclassification rate of 10.5%. Regression analysis confirmed these findings (Table 2 ). Individually, NLR ≥11.4 was significantly associated with SAP, with adjusted ORs of 16.7 ( P < 0.001) in the derivation cohort and 8.15 ( P < 0.001) in the validation cohort. Similarly, CRP ≥173.3 was associated with SAP with adjusted ORs of 8.32 and 11.8, respectively (both P < 0.001). When combining both biomarkers, patients with NLR ≥11.4 and CRP ≥173.3 had the highest risk of SAP, with adjusted ORs of 55.1 and 87.7 in the derivation and validation cohorts, respectively (both P < 0.001). The subgroup with NLR ≥11.4 but CRP <173.3 showed increased SAP risk in the derivation cohort (OR = 4.67, P = 0.015), but this was not statistically significant in the validation cohort (OR = 2.01, P = 0.3). Evaluate models in predicting SAP in derivation and validation cohort AUC, area under the curve; CI-95%, confidence level-95%; CRP, C-reactive protein; −LR, negative likelihood ratio; NLR, neutrophil-to-lymphocyte ratio; −PTP, negative posttest predictive; NPV, negative predictive value; +LR, positive likelihood ratio; +PTP, positive posttest predictive; PPV, positive predictive value; SAP, severe acute pancreatitis. a P 1 and P 2 were, respectively, P value for NLR + CRP vs NLR and NLR + CRP vs CRP (using the Delong test). The proportions of patients classified into each risk group across both derivation and validation cohorts are summarized in Supplementary Digital Content (see Supplementary Figure 2, http://links.lww.com/CTG/B376 ). The diagnostic accuracy of NLR, CRP, and the combined NLR + CRP model in predicting SAP is summarized in Table 3 . Among all models, the combination of NLR ≥11.4 and CRP ≥173.3 yielded the best overall performance in both the derivation and validation cohorts. Combined use of CRP with NLR ratio in differentiating between SAP and non-SAP in patients 95% CI, confidence level-95%; CRP, C-reactive protein; NLR, neutrophil-to-lymphocyte ratio; OR, odd ratio; SAP, severe acute pancreatitis. In the derivation cohort, the NLR + CRP model demonstrated an AUC of 0.913, with high sensitivity (90.9%) and specificity (84.5%). In the validation cohort, the model maintained strong performance, with an AUC of 0.866, sensitivity of 84.4%, and specificity of 77.6%. The model also produced the highest overall accuracy in both cohorts: 86.2% and 80.8%, respectively. Positive and negative predictive values were consistently superior in the combined model. Notably, the posttest probability of SAP after a positive test result increased to approximately 65% in both cohorts, compared with lower values seen with either NLR or CRP alone. Negative posttest probabilities (PTP) remained below 10%, supporting its rule-out capacity. The DeLong test confirmed that the combined model significantly outperformed CRP alone ( P = 0.04) and NLR alone ( P = 0.001) in the validation cohort. Supplementary Figure 3, http://links.lww.com/CTG/B377 , displays the ROC curves for all 3 models in the validation cohort. The calibration performance of the combined NLR + CRP model was assessed in the validation cohort. The Brier score was 0.108, reflecting good agreement between predicted probabilities and observed outcomes. The calibration curve demonstrated that predicted risks were well-aligned with actual SAP incidence across risk strata (see Supplementary Figure 4, http://links.lww.com/CTG/B378 ). The NLR + CRP model was compared with BISAP and the combined BISAP + NLR + CRP model in both cohorts. In the validation cohort, the AUC for NLR + CRP was 0.866, compared with 0.901 for BISAP. Although BISAP achieved the highest AUC, the differences between BISAP and NLR + CRP were not statistically significant ( P = 0.286) (Figure 5 and see Supplementary Table 1, http://links.lww.com/CTG/B379 ). Receiver operating characteristic curves comparing the diagnostic performance of the NLR + CRP model and the BISAP score in the validation cohort for predicting severe acute pancreatitis. NLR, neutrophil-to-lymphocyte ratio; CRP, C-reactive protein; BISAP, Bedside Index for Severity in Acute Pancreatitis. The DCA indicated that NLR + CRP provided clinical utility similar to that of BISAP in predicting SAP. Although BISAP demonstrated slightly higher net benefits across most threshold probabilities, NLR + CRP performed particularly well within the 75%–85% range (Figure 6 ). Decision curve analysis comparing the net clinical benefit of the NLR + CRP model vs the BISAP score for predicting severe acute pancreatitis in derivation ( a ) and validation ( b ) cohorts. NLR, neutrophil-to-lymphocyte ratio; CRP, C-reactive protein; BISAP, Bedside Index for Severity in Acute Pancreatitis.

This study is the first prospective evaluation in Vietnam of the CART model using NLR and CRP to predict SAP. The results show that combining NLR and CRP measurements within 24 hours of admission significantly improves predictive accuracy, providing a practical tool for early clinical decision-making. The CART model demonstrated strong performance, with an AUC of 0.866 in the validation cohort, outperforming NLR (AUC 0.783, P = 0.001) and CRP (AUC 0.819, P = 0.04), and showing noninferiority to BISAP (AUC 0.9, P = 0.286). Diagnostic metrics included LR + values of 5.864 and 3.768, and LR − values of 0.108 and 0.201 in the derivation and validation cohorts, respectively. Notably, 81.3% of high-risk patients identified by the model developed SAP, confirming the model's effectiveness in risk stratification (Figure 7 ). Decision curve for NLR + CRP and BISAP in predicting severe acute pancreatitis. Although the BISAP score remains a widely accepted tool for SAP prediction, the NLR + CRP model's simplicity—relying solely on accessible blood tests—makes it more feasible in resource-limited settings, where advanced imaging or BISAP components may not be available. Decision curve analysis further confirmed the clinical utility of the NLR + CRP model, positioning it as a viable alternative to BISAP in such settings. NLR reflects the inflammatory and immune responses triggered by pancreatic injury, mediated by cytokines like tumor necrosis factor-α and interleukins (interleukin [IL]-1β, IL-6, IL-8, and IL-18) ( 10 , 22 ). These cytokines promote neutrophil infiltration and lymphocyte reduction, contributing to tissue destruction, SIRS, and organ failure ( 10 ). Previous studies have consistently shown that higher baseline NLR values correlate with greater disease severity and organ failure in SAP ( 37 – 39 ). Elevated NLR has been linked to prolonged hospital stays, increased morbidity, and higher mortality in SAP ( 2 , 10 , 21 ). Han et al ( 39 ) reported that NLR measured within 48 hours of admission achieved the highest AUC for predicting SAP, with a cutoff value of 6.66 and showed a positive correlation with Ranson scores and hospital stay duration. Similarly, Zhang et al ( 21 ) found that elevated NLR was associated with persistent organ failure and higher mortality, particularly in Asian populations. Kong et al ( 22 ) conducted a meta-analysis to evaluate the diagnostic value of NLR in predicting AP severity. The study found NLR to have moderate diagnostic performance with an AUC of 0.82, sensitivity of 79%, and specificity of 71%. In addition, CRP, a liver-produced marker of inflammation, peaks 72–96 hours after symptom onset and has established prognostic value in AP, especially in Asian populations where severe complications are more prevalent ( 15 , 23 ). A systematic review and meta-analysis assessed the diagnostic value of CRP in predicting AP severity. CRP demonstrated good diagnostic performance with an AUC of 0.85, sensitivity of 0.76, and specificity of 0.79, supporting its role as a reliable biomarker in clinical settings ( 24 ). Our study sought to address limitations in prior research, such as small sample sizes and varying intervals between symptom onset and hospital admission ( 21 , 37 , 40 ). Our combined model significantly improved the AUC compared with CRP and NLR alone, consistently outperforming the individual markers in the validation cohort. This highlights the complementary roles of neutrophil elevation and lymphocyte suppression in the pathophysiology of SAP ( 21 ). Similarly, Lu et al ( 15 ) similarly reported that integrating BISAP, CRP, and NLR improved early severity prediction. However, their study had a retrospective design, small sample size, and lack of prospective validation limit its applicability. On the other hand, traditional scoring systems such as BISAP, APACHE-II, and Ranson often yield PTP below 50%, even with positive results ( 9 ). By comparison, our model achieved positive PTP of approximately 66% and negative PTP of 9.6%, offering significant clinical value for screening and early identification of SAP in primary healthcare settings. Although less sophisticated than AI-based models, the CART approach integrates accessible clinical variables, making it practical for resource-constrained settings. The strength of this study lies in the novel assessment of NLR and CRP in a Vietnamese population, addressing a local research gap. CART analysis provided actionable cutoff values, enhancing predictive accuracy. Calibration, bootstrap, and k-fold cross-validation increased the reliability of the model, while DCA confirmed its clinical utility. The CART model is simple and easy to implement, requiring only 2 common blood tests, unlike more complex systems such as BISAP that rely on specific clinical data or imaging. This makes the model particularly suitable for low-resource settings where access to imaging is limited ( 37 , 38 , 40 ). However, the study also has limitations. First, residual confounders such as undiagnosed inflammatory conditions may have an influenced the results. Second, the single-center design limits the generalizability of our findings, particularly to community or primary care settings. Third, the absence of long-term follow-up data prevents evaluation of the model's predictive value for chronic outcomes. Fourth, NLR values could be affected by other inflammatory conditions or treatments, such as antibiotics, which were not considered in this study ( 38 ). Fifth, the higher proportion of SAP cases in our cohort may not reflect broader populations, and regional differences in AP etiology necessitate locally specific cutoff values. For example, hypertriglyceridemia-induced pancreatitis is more common in Vietnam than in Europe, where alcohol and gallstones predominate ( 6 , 25 ). In addition, racial disparities also exist, with pancreatic necrosis being more common in Asian patients than White patients ( 41 ). Finally, we did not use the APACHE score due to its complexity in resource-limited settings. In conclusion, the CART-based NLR + CRP model is a practical, reliable tool for early severity stratification in AP. Its strong predictive performance, simplicity, and comparable accuracy to BISAP make it particularly valuable for identifying high-risk patients, especially in resource-limited settings. Further validation in diverse cohorts will solidify its clinical utility.

Guarantor of article: Thong Duy Vo, MD, PhD. Specific author contributions: T.M.H. and T.D.V.: conceptualized the study and drafted the manuscript. T.M.H., D.T.T., and Y.T.H.D.: collected the data. T.M.H. and A.T.: analyzed the data. T.M.H. and T.D.V.: provided critical revisions and supervised the study. All authors read and approved the final manuscript. Financial support: None to report. Potential competing interests: None to report. AI disclosure: This manuscript was prepared by the authors. Artificial intelligence tools were not used for content generation or analysis. Grammarly was used solely for correcting typographical and grammatical errors. Consent for publication: Informed consent was obtained from all participants and/or their legal guardians for study participation and publication of identifying information/images in an online open-access publication, in accordance with relevant guidelines and regulations. Availability of data and materials: The data sets generated by/or analyzed during this study are not publicly available due to data confidentiality policies. However, the data are available from the corresponding author Thong Duy Vo on reasonable request. To request access to the research data, contact the Ethics Committee of the University of Medicine and Pharmacy at Ho Chi Minh City at [email protected] . Study Highlights WHAT IS KNOWN ✓ Severe acute pancreatitis (SAP) is life-threatening and requires early risk stratification. ✓ Traditional tools like BISAP, while effective, have limitations in resource-limited settings. ✓ NLR and CRP are promising biomarkers for SAP prediction, reflecting inflammation and immune responses. WHAT IS NEW HERE ✓ A CART-based decision tree model combines NLR ≥11.4 and CRP ≥173.3 for SAP prediction. ✓ The model shows high accuracy (AUC 0.866), sensitivity (90.9%), and specificity (84.5%). ✓ Simpler and cost-effective, it is ideal for resource-limited clinical environments. ✓ Severe acute pancreatitis (SAP) is life-threatening and requires early risk stratification. ✓ Traditional tools like BISAP, while effective, have limitations in resource-limited settings. ✓ NLR and CRP are promising biomarkers for SAP prediction, reflecting inflammation and immune responses. ✓ Severe acute pancreatitis (SAP) is life-threatening and requires early risk stratification. ✓ Traditional tools like BISAP, while effective, have limitations in resource-limited settings. ✓ NLR and CRP are promising biomarkers for SAP prediction, reflecting inflammation and immune responses. ✓ A CART-based decision tree model combines NLR ≥11.4 and CRP ≥173.3 for SAP prediction. ✓ The model shows high accuracy (AUC 0.866), sensitivity (90.9%), and specificity (84.5%). ✓ Simpler and cost-effective, it is ideal for resource-limited clinical environments. ✓ A CART-based decision tree model combines NLR ≥11.4 and CRP ≥173.3 for SAP prediction. ✓ The model shows high accuracy (AUC 0.866), sensitivity (90.9%), and specificity (84.5%). ✓ Simpler and cost-effective, it is ideal for resource-limited clinical environments.

This prospective observational study was conducted at Cho Ray Hospital, Vietnam, from November 2022 to September 2023. Patients with AP were enrolled based on clinical symptoms, elevated serum amylase or lipase levels (>3 times the upper limit of normal), and/or imaging findings consistent with AP ( 13 ). The study was approved by the institutional review board, with written consent obtained from all participants, and followed Strengthening the Reporting of Observational Studies in Epidemiology guidelines ( 29 ). Patients aged 18 years or older with a first episode of AP were included. The exclusion criteria were as follows: (i) underlying conditions that may confound CRP and NLR levels, including chronic pancreatitis, recurrent acute pancreatitis, trauma-related pancreatitis, autoimmune diseases, malignancies, immunosuppressive therapy, pregnancy, and severe preexisting illnesses (e.g., New York Heart Association class IV heart failure, decompensated cirrhosis, or end-stage renal disease); (ii) delayed hospital admission, defined as ≥72 hours after symptom onset, to ensure biomarker measurements reflected early inflammatory changes pertinent to initial prognostic evaluation; and (iii) refusal to provide informed consent. Blood samples were collected within the first 24 hours of admission to measure CBC and CRP levels. Subsequently, the BISAP score and NLR were calculated. Imaging data obtained from spiral computed tomography (CT) scans or magnetic resonance imaging (MRI) performed between 24 and 72 hours after admission were also reviewed. CBC analysis was conducted using the ADVIA 2120i Hematology System (Siemens Healthineers, Erlangen, Germany) within 6 hours of sampling. Blood was drawn into EDTA tubes to prevent clotting. Neutrophil and lymphocyte counts were used to calculate the NLR, defined as the ratio of absolute neutrophil count to absolute lymphocyte count. The ADVIA 2120i Hematology System demonstrates an analytical sensitivity of 0.1 × 10 9 /L for leukocytes and a coefficient of variation of ≤3.0% for cell counts at 5 × 10 9 /L (Siemens Healthineers, 2012) ( 30 ). CRP levels were measured using the standard CRP assay on the ADVIA 1800 Chemistry System (Siemens Healthineers, Erlangen, Germany). Results were expressed in milligram per liters. The CRP assay has an analytical sensitivity (limit of detection) of 4 mg/L and coefficient of variation of 1.1% at 31 mg/L, as reported by the manufacturer (Siemens Healthineers, 2011) ( 31 ). Abdominal CT and MRI scans were performed using GE Healthcare (Chicago, IL) using 32-slice, 64-slice, and 128-slice detector row CT systems and GE MRI systems, with contrast-enhanced or noncontrast protocols. Contrast-enhanced CT used an iodine-based contrast agent, while MRI used gadolinium-based contrast, following standard injection protocols. BISAP was calculated within 24 hours of admission, based on 5 parameters: blood urea nitrogen >25 mg/dL, Glasgow Coma Scale 60 years, and pleural effusion on imaging (7). Scores >3 indicated higher SAP risk ( 6 , 7 ). Based on the Revised Atlanta Classification (2012), patients were classified as having mild, moderately severe, or severe AP ( 32 ). AP severity was classified as mild (no organ failure or complications), moderately severe (transient organ failure 48 hours, with or without complications). Organ failure was defined as respiratory (PaO 2 /FiO 2 170 μmol/L), or cardiac (systolic blood pressure 48 hours, while mild and moderately severe AP were grouped as non-SAP. All patients were treated per hospital guidelines, adjusted based on the severity ( 2 , 13 ). Sample size calculation was based on detecting a significant difference between the decision tree and BISAP models using the area under the curve (AUC). With an α of 0.05, β of 0.2 (80% power), and assuming AUCs of 0.82 for BISAP ( 7 ) and 0.7 for the decision tree model, the required sample size was estimated at 124 patients. A rank correlation of 0.6 between the 2 models was hypothesized. The final sample consisted of 60% derivation cohort (210 patients) and 40% validation cohort (130 patients), ensuring sufficient power for the DeLong test. Demographic analysis was performed using descriptive statistics: mean (SD) for normally distributed variables, median (Q1–Q3) for skewed variables, and frequency (%) for qualitative variables. T-tests, Wilcoxon rank-sum tests, and Fisher exact test were used for comparisons. The distribution of NLR and CRP was visualized with violin plots, stratified by severity of pancreatitis. Correlations between BISAP, CRP, and NLR were assessed using Spearman rank correlation and visualized through a heatmap. The models were developed as follows: NLR alone (i) and CRP alone (ii) were created using logistic regression, while the decision tree (iii) was built using the CART method ( 28 , 34 ). Cross-validation (10-fold) was used to determine the optimal hyperparameter (Complexity Parameter) for the decision tree ( 35 ). After finding the best complexity parameter value, final models were developed. The cutoffs for models (i) and (ii) were determined by the Youden index, and odds ratios (OR) for SAP prediction were calculated for each model using univariate and multivariate logistic regression. Confounders, including age, sex, etiology, diabetes mellitus, and white blood cell count, were adjusted based on clinical relevance or changes in effect estimates greater than 10% ( 3 ). Model performance was evaluated using AUC, accuracy, sensitivity, specificity, positive predictive value, negative predictive value, likelihood ratios, and pretest probability of SAP in both cohorts. AUC values were visualized with ROC curves, and the DeLong test was used to compare the decision tree model to BISAP ( 36 ). A model with an AUC above 0.7 was considered useful, and an AUC between 0.8 and 0.9 indicated excellent accuracy ( 28 ). The combined NLR + CRP model was calibrated using the Brier score to assess model accuracy, with smaller values indicating better fit. Finally, Decision Curve Analysis (DCA) quantified the clinical utility of each model. All analyses were conducted in R (v4.4.1) using the package for ROC analysis in R, rpart, and dcurves packages. Statistical significance was defined as a 2-sided P < 0.05. The study flowchart is shown in Figure 1 . Flowchart of the study. ROC, receiver operating characteristic. This script has been made publicly available on GitHub ( https://github.com/leeantran/DecisionTree_SAP ). This investigation forms part of our broader research project focusing on biomarkers and clinical parameters associated with acute pancreatitis. The study was conducted in accordance with the Declaration of Helsinki. Ethical approval was obtained from the Ethics Committee of the University of Medicine and Pharmacy at Ho Chi Minh City (IRB number: 829/ĐHYD-HĐĐĐ, sign: November 3, 2022). Written informed consent was obtained from all participants.