Routine Laboratory Data for Predicting 30-Day Emergency Department Revisits: The AXIS-2 Risk Score

Routine Laboratory Data for Predicting 30-Day Emergency Department Revisits: The AXIS-2 Risk Score | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Routine Laboratory Data for Predicting 30-Day Emergency Department Revisits: The AXIS-2 Risk Score Ferhat Demirci, Aylin Demirci This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7980356/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 13 Feb, 2026 Read the published version in BMC Emergency Medicine → Version 1 posted 14 You are reading this latest preprint version Abstract Background: Thirty-day emergency department (ED) revisit is a major quality indicator reflecting morbidity and healthcare burden. Laboratory data obtained during outpatient encounters may capture underlying biological stress axes. This study aimed to develop and externally validate an interpretable logistic regression model—the AXIS-2 (Anemia–Inflammation Two-Axis Composite Index)—for predicting 30-day ED revisits using routine laboratory parameters. Methods: This retrospective study included 92,386 adult outpatients who visited a tertiary academic hospital between January 2015 and August 2025. Laboratory variables were biologically grouped into two axes: Axis-1 (hematologic indicators of anemia and erythropoiesis: hemoglobin, lymphocyte count, mean corpuscular volume, neutrophil count, platelet count, red cell distribution width, and white blood cell count) and Axis-2 (inflammatory and catabolic markers: C-reactive protein, ferritin, albumin, neutrophil-to-lymphocyte ratio, and platelet-to-lymphocyte ratio). A multivariable logistic regression model was trained using stratified sampling and isotonic calibration. Model performance was assessed in both internal and external test datasets using discrimination (AUC-ROC, AUC-PR), classification (F1 score, MCC, sensitivity, specificity), calibration, and decision curve analysis. Results: The 30-day revisit rate was 30%. Both hematologic (Axis-1) and inflammatory (Axis-2) axes were independent predictors of revisit risk (Axis-1 OR 2.06; Axis-2 OR 2.31; p < 0.001). The model showed excellent discrimination (AUC-ROC 0.921 in the test set; 0.935 in the external set) and balanced classification accuracy (≈ 85% for both sensitivity and specificity). Negative predictive value reached 95%, and calibration metrics indicated strong concordance between predicted and observed probabilities. Decision curve analysis demonstrated clear net benefit within the 10–40% probability range. High-risk patients also exhibited higher 30-day readmission (≈ 45%), transfusion (≈ 15%), and 180-day mortality (≈ 3%) rates compared with the low-risk group. Conclusions: AXIS-2 is a transparent, laboratory-based model that accurately predicts 30-day emergency revisits and offers interpretable outputs suitable for integration into clinical workflows. Its strong calibration and decision-analytic benefit support its use as a cost-efficient tool for post-discharge monitoring and healthcare resource optimization. Figures Figure 1 Figure 2 Figure 3 1. INTRODUCTION A substantial proportion of patients presenting to outpatient clinics return to the emergency department within a short period of time. This phenomenon not only increases healthcare utilization but also reflects an elevated risk of morbidity and mortality. Therefore, early prediction of emergency department revisit risk using laboratory data obtained at the time of the outpatient encounter carries critical clinical importance. Similarly, among patients discharged from the emergency department, 30-day revisit is common and represents a key quality indicator. While the conventional 72-hour metric fails to capture many events, the 30-day revisit rate has been reported to be approximately 20% in previous studies [ 1 ]. In the context of revisits, anemia and inflammation emerge as two principal biological axes. Anemia contributes to adverse prognosis through alterations in erythrocyte indices and a reduction in oxygen-carrying capacity. In particular, parameters such as hemoglobin (HGB), red cell distribution width (RDW), and mean corpuscular volume (MCV) have been shown to be associated with short-term clinical outcomes [ 2 , 3 ]. Inflammation, on the other hand, can be readily monitored through parameters such as elevated C-reactive protein (CRP), decreased albumin, and an increased neutrophil-to-lymphocyte ratio (NLR). These markers have been found to be valuable predictors of short-term mortality and revisit risk in numerous studies [ 4 , 5 ]. In recent years, machine learning models have demonstrated promising results in predicting emergency department visits [ 6 , 7 ]. However, the interpretability of these models remains limited, making their widespread adoption as decision-support tools by clinicians challenging. Therefore, there is a need for scoring systems that not only provide strong predictive performance but are also computationally feasible and transparent. In this context, multivariable models such as logistic regression are widely used, as they offer interpretability and transparency. For studies of this nature, the TRIPOD [ 8 ] and PROBAST [ 9 ] frameworks are recommended for reporting and risk of bias assessment. Moreover, calibration metrics—including calibration-in-the-large (CITL) and calibration slope—should always be reported and recalibrated when necessary. To illustrate clinical utility in a threshold-based manner, Decision Curve Analysis, which relies on the net benefit approach, is commonly advocated [ 10 , 11 ]. In this study, we aimed to develop a novel risk score, termed AXIS-2 (Anemia–Inflammation Two-Axis Composite Index). The objectives were to evaluate the predictive performance of the AXIS-2 score in forecasting 30-day emergency department revisits among outpatients, to test its validity in an independent dataset, and to present it as a transparent risk assessment tool that may contribute to clinical decision-making. Additionally, this study investigated comparative outcomes in patients identified by AXIS-2 as being at high risk of revisit, including hospitalization rates, transfusion requirements, and mortality during follow-up. 2. MATERIALS AND METHODS 2.1 Study Population / Subjects This study was conducted at SBÜ İzmir Tepecik Training and Research Hospital, a tertiary care academic center, and its affiliated service building. The primary dataset included patients who presented on an outpatient basis, excluding emergency department visits, between January 1, 2015, and August 31, 2025. Inclusion criteria: 1. Patients aged >18 years. 2. Patients presenting to non-emergency outpatient clinics of our hospital between January 1, 2015, and August 31, 2025 (only the first eligible visit was considered). 3. Patients for whom the following laboratory tests were requested completely and without missing values: hemoglobin (HGB), white blood cell count (WBC), absolute neutrophil count (NEU), absolute lymphocyte count (LYM), platelet count (PLT), red cell distribution width (RDW), mean corpuscular volume (MCV), C-reactive protein (CRP), vitamin B12, folate, ferritin, iron, albumin, creatinine, alanine aminotransferase (ALT), and aspartate aminotransferase (AST). Exclusion criteria: 1. Patients with outpatient visits other than the first eligible outpatient encounter during the study period (to avoid duplicate results).İlk uygun poliklinik başvurusu öncesi aynı gün acil başvurusu olan hastalar 2. Patients with an emergency department visit on the same day prior to the first eligible outpatient encounter. 3. Patients younger than 18 years of age. 4. Patients whose residential address was transferred outside of İzmir in the Central Population Management System of Turkey (MERNIS) registry within 30 days. 5. Patients with missing laboratory test results. 6. Patients whose laboratory test results did not contain numeric data. 7. Patients who did not provide a sample within 2 hours after the outpatient visit. 8. Patients with a lymphocyte count of zero, precluding calculation of NLR and PLR. 9. Patients with more than one protocol number recorded on the same day. 10. Pregnant patients. 11. Patients with active oncologic disease or hematological malignancy. 12. Forensic cases. 13. Patients diagnosed with acute coronary syndrome (ACS), pulmonary embolism (PE), aortic dissection/aneurysm rupture, subarachnoid hemorrhage/stroke-TIA, or acute massive gastrointestinal bleeding (as these conditions require immediate spontaneous emergency department admission). Whole blood samples were collected into K2-EDTA tubes by phlebotomy nurses and analyzed within 30 minutes using a Sysmex XN-1000 analyzer (Kobe, Japan). Complete blood count (CBC) reagents and calibrators were obtained from the respective manufacturer, certified, and officially registered products, while internal quality control materials were supplied by Sysmex Corporation (Kobe, Japan). Serum for biochemical assays, including C-reactive protein (CRP), ferritin, iron, albumin, creatinine, alanine aminotransferase (ALT), and aspartate aminotransferase (AST), was obtained from plain tubes after centrifugation and analyzed within 2 hours on a Beckman Coulter AU-5800 analyzer (California, USA). All reagents and calibrators for biochemical measurements were provided by the respective manufacturer, certified, and officially registered, with internal quality control materials supplied by Bio-Rad Corporation (California, USA). 2.2 Study Design Prior to the initiation of the study, approval was obtained from the Non-Interventional Ethics Committee of SBÜ İzmir Tepecik Training and Research Hospital (Decision Date 11.09.2025 and No: 2025 / 08-19). To ensure data confidentiality, all patient identifiers were anonymized. A dataset including age, sex, complete blood count (CBC) parameters—white blood cell count (WBC), neutrophil count (NEU), lymphocyte count (LYM), platelet count (PLT), hemoglobin (HGB), red cell distribution width (RDW), and mean corpuscular volume (MCV)—as well as calculated ratios such as the neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR), and biochemical tests including C-reactive protein (CRP), vitamin B12, folate, ferritin, iron, albumin, creatinine, alanine aminotransferase (ALT), and aspartate aminotransferase (AST), was constructed from the records of 131,867 patients and transferred to Microsoft Excel 2021 (USA) for analysis. Among the numerous tests included in the complete blood count and basic biochemistry panel, we selected parameters with established prognostic relevance supported by the literature and high accessibility in routine clinical practice. These variables were grouped into two axes: Axis-1 (E1), representing indicators related to anemia and erythropoiesis (Hb, LYM, MCV, NEU, PLT, RDW, WBC), and Axis-2 (E2), reflecting inflammation and catabolism (CRP, ferritin, albumin, NLR, PLR). In this way, the model was designed to capture a biologically meaningful structure that integrates two fundamental processes in a comprehensive manner. Reference ranges for the laboratory tests included in the study are provided in Supplementary Material, Table SM1. After applying the exclusion criteria, the final dataset comprised 92,386 patients. This dataset was transferred to the Python environment (version 3.11, USA) for regression analysis. Non-numeric categorical variables were converted into numerical values as follows: male = 1, female = 2; 30-day emergency department revisit (Emergency30) = 0/1. The distribution of Emergency30 events was examined to assess class balance in the cleaned dataset. To prevent overfitting and preserve class proportions during dataset partitioning, a stratified sampling approach was employed. In addition, a class-weighting method was applied during model development to mitigate class imbalance. Following data preprocessing, the dataset was randomly divided into three parts using stratified sampling based on the binary outcome variable while maintaining the original class distribution: 60% training set, 20% internal test set (hereafter referred to as the test set), and 20% external test set. The number of patients allocated to each dataset (training, test, external test) is detailed in a flow diagram (Figure 1), in accordance with the TRIPOD guidelines. 2.3 Data Preprocessing and Training of Machine Learning Algorithms The cleaned dataset was processed in Python 3.11 (USA) using both Kaggle Notebook (USA) and Google Colab (USA) platforms. Logistic regression (LR) was employed for classification, with class_weight = "balanced" applied to mitigate class imbalance. The input set consisted of Axis-1 (hematology axis), Axis-2 (inflammation axis), sex, age, creatinine, ALT, and AST variables. The axes were defined by z-scoring the corresponding laboratory variables based on the training set, multiplying each variable by a sign reflecting its clinical risk direction (i.e., positive for risk-increasing markers and negative for protective ones), and calculating the composite axis score as the mean of these terms. Continuous variables were standardized using the “StandardScaler” function, and categorical variables were encoded using one-hot encoding. Stratified sampling was used during dataset partitioning to preserve class proportions across the sets. The model was trained on the development (training) dataset and internally validated using an independent internal test set derived from the same source population. To improve the calibration of predicted probabilities, isotonic calibration was applied, ensuring better agreement between predicted probabilities and observed outcomes. The optimal decision threshold was determined by maximizing the Youden J index on the internal test set. The finalized model, together with this fixed threshold, was subsequently evaluated on an external validation dataset to assess its generalizability and real-world applicability. In all analyses, the same predefined threshold derived from the internal test set was used for evaluation of both the internal and external validation datasets, without further recalibration. 2.4 Performance Evaluation The logistic regression model was comprehensively evaluated using both the test set and the external test set. To ensure the reliability and clinical interpretability of the model, a multidimensional set of performance metrics was employed. Classification Performance Metrics · Area under the receiver operating characteristic curve (AUC-ROC) · Area under the precision–recall curve (AUC-PR) · Confusion matrix analysis at the Youden J threshold · Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, and F1 score · Matthews correlation coefficient (MCC) · Likelihood ratios (LR⁺/LR⁻) and diagnostic odds ratio (DOR) · Relative risk (RR) with 95% confidence intervals (CIs) Calibration and Clinical Utility Metrics · Calibration-in-the-large (CITL) and calibration slope (with 95% CIs) · Calibration plots comparing predicted and observed risks · Decision Curve Analysis (DCA) to evaluate net clinical benefit across a range of probability thresholds Validation Results · The optimal decision threshold was determined by maximizing the Youden J index on the internal test set and then directly applied to the external validation set without any further adjustment. · All primary performance metrics were reported with 95% confidence intervals obtained through bootstrapping. · Additional analyses included threshold sweeps (±20% around the Youden J point) and subgroup evaluations (sex, age tertiles, anemia status) to enhance the clinical interpretability of the model. This structured and multidimensional evaluation process demonstrated both the discriminative ability and the clinical applicability of the logistic regression model. 2.5 Statistical Analysis Comparisons across the three datasets (training, test, and external test) were performed using one-way ANOVA for continuous variables and the Pearson chi-square test for categorical variables. For variables with statistically significant differences, post-hoc pairwise comparisons were conducted to identify the source of variation. Continuous variables were standardized using z-scores, and odds ratios (ORs) were reported per one standard deviation increase. All statistical analyses were conducted in Python (v3.11) using the pandas , scipy , and statsmodels libraries, with two-sided p-values <0.05 considered statistically significant. 3. RESULTS The baseline characteristics of the study population, including demographic data, hemogram parameters, and relevant biochemical variables, are summarized in Table 1 , stratified by training, test, and external test sets. In comparisons across groups, no significant differences were observed in sex and age distributions (p > 0.05). Similarly, the majority of hematological parameters did not differ significantly between the datasets. However, statistically significant differences were identified in albumin (p = 0.017) and ferritin (p = 0.0058) levels, which were attributable to variations in the test set. In addition, the outcome variable Emergency30 differed significantly across the training, test, and external test sets (p < 0.001), with this difference also driven by the test set. In the training dataset, the mean age of patients was 52.93 ± 19.46 years, with males having a mean age of 54.06 ± 19.56 years and females 52.14 ± 19.36 years. The sex distribution consisted of 58.97% females and 41.03% males. The test and external test sets demonstrated similar age and sex distributions; however, the rate of emergency department revisits was higher in the test set compared with the other two sets. Potential linear associations among variables included in the modeling were assessed using Pearson correlation, while monotonic associations were evaluated using Spearman correlation. The results and visualizations of these analyses are provided in the Supplementary Material (Figure SM1). Following these baseline comparisons, a multivariable logistic regression model was developed to predict short-term emergency department revisits, which constituted the primary objective of the study. The model incorporated hematologic (E1) and inflammatory (E2) scores, defined along two biologically meaningful axes, together with age, sex, creatinine, ALT, and AST. In defining the axes, coefficients for variables known to be inversely associated with risk—such as hemoglobin (HGB), albumin, and lymphocytes (LYM)—were assigned negative signs, whereas parameters associated with increased risk (e.g., NEU, RDW, CRP) were assigned positive signs. In this way, a biologically grounded and interpretable logistic regression framework was constructed (Table 2 A). Vitamin B12 and folate levels showed strong correlations with MCV and RDW, while iron exhibited a moderate correlation with ferritin, as illustrated in Supplementary Material Figure SM1. Inclusion or exclusion of these parameters did not result in meaningful differences in the primary performance metrics (F1 score, MCC, and AUC-ROC) (ΔF1 score ± 0.008, p > 0.05). Therefore, to enhance model parsimony and simplify calculation in clinical practice, iron, vitamin B12, and folate were not included in the final formula. When the predefined threshold of 0.219 was applied, performance metrics were consistent across the test and external test sets. In sensitivity analyses conducted within ± 20% of this threshold, changes in the key metrics (sensitivity, specificity, F1 score, MCC) remained negligible. Moreover, 95% confidence intervals obtained through bootstrapping (B = 1000) indicated that these differences were not statistically significant. The regression formula, coefficients, and odds ratios of the final model are summarized in Tables 2 A and 2 B. Both the hematologic axis (E1) and the inflammatory axis (E2) were identified as significant independent predictors of 30-day emergency department revisit (E1: OR 2.06, 95% CI 2.00–2.13; E2: OR 2.31, 95% CI 2.17–2.46; both p < 0.001). Age and creatinine also emerged as strong predictors (Age: OR 4.23, 95% CI 4.07–4.40; Creatinine: OR 2.89, 95% CI 2.71–3.08), whereas an increase in AST levels was unexpectedly associated with the highest risk (OR 17.20, 95% CI 14.68–20.15; p < 0.001). Female sex, on the other hand, was found to be inversely associated with risk and demonstrated a protective effect compared with males (OR 0.35, 95% CI 0.33–0.37; p < 0.001). The logistic regression model demonstrated high classification performance in both the internal test set and the external validation set. In the test set, sensitivity was 85.2% (95% CI 84.2–86.1), specificity 84.9% (95% CI 84.3–85.5), and accuracy 84.9% (95% CI 84.5–85.5). Similarly, in the external test set, high sensitivity (87.9%, 95% CI 87.0–88.8) and specificity (84.2%, 95% CI 83.6–84.8) were obtained. The positive predictive value was 70.9% in the test set and 67.5% in the external set, whereas the negative predictive value was 93.0% and 94.9%, respectively (Table 3 ). Overall performance metrics of the model also demonstrated balanced success. The F1 score (0.77 in the test set; 0.76 in the external set) and Matthews correlation coefficient (0.67 in both sets) supported the consistency of classification between positive and negative classes. The diagnostic odds ratio (DOR) was 32.3 (95% CI 29.6–35.3) in the test set and 38.8 (95% CI 35.3–42.8) in the external set, indicating strong discriminatory power. Relative risk (RR) was 10.1 and 13.3, respectively, demonstrating that the probability of emergency department revisit was markedly higher in the high-risk group compared with the low-risk group (Table 3 ). The area under the receiver operating characteristic curve (AUC-ROC), reflecting discriminative ability, was 0.921 (95% CI 0.916–0.924) in the test set and 0.935 (95% CI 0.931–0.938) in the external set, demonstrating strong model discrimination. Similarly, the area under the precision–recall curve (AUC-PR) exceeded 0.82 in both sets (Fig. 2 ). Calibration analysis (Calibration-in-the-Large, CITL) demonstrated good agreement of the model in both the test and external test sets. In the test set, the intercept was 0.28 (95% CI 0.23–0.33) and the calibration slope was 0.92 (95% CI 0.89–0.94), with a CITL value of 0.34 (95% CI 0.29–0.38). In the external set, the intercept was 0.06 (95% CI 0.01–0.12), the calibration slope was 1.01 (95% CI 0.98–1.04), and the CITL was 0.06 (95% CI 0.01–0.11). These values indicate that the model was well calibrated in both internal and external validation (in Supplementary Material, Figure SM2 and Table SM2). To assess the clinical utility of the model, decision curve analysis (DCA) was performed. DCA illustrates the net benefit of the model across different probability thresholds in comparison with two extreme strategies: treating all patients as high risk (“treat-all”) and treating no patients as high risk (“treat-none”). This analysis highlights not only the statistical discriminative ability but also the contribution of the model to clinical decision-making. In our study, the model provided greater net benefit than either of the two extreme strategies across a wide range of probability thresholds in both the test and external test sets (Fig. 3 ). The added benefit was particularly evident within the 10–40% threshold range, suggesting that the model can be reliably applied in clinical practice for patients within the low-to-intermediate risk spectrum. The similar findings observed in the external validation set further support the generalizability and clinical validity of the model. With respect to calibration, the Brier score was 0.10 in the test set and 0.09 in the external set, with these low values indicating good model calibration. In addition, pseudo-R² values—including McFadden (0.46–0.51), Cox–Snell (0.43–0.45), and Nagelkerke (0.61–0.65)—suggested a high explanatory power of the model. Furthermore, likelihood ratios (PLR ≈ 5.6; NLR ≈ 0.14–0.17) supported the strong diagnostic performance of the model (Table 3 ). In addition, subgroup analyses were conducted by sex and age tertiles to assess the consistency of the model. While the overall performance metrics of the model were preserved across subgroups, differences in sensitivity and specificity values were observed between groups (in Supplementary Material, Table SM3). An exploratory, observational evaluation was performed for secondary clinical outcomes. No additional modeling was conducted, and no formal statistical inference was intended. For example, among patients classified as high risk by the model (Pred = 1), 30-day readmission (44–45%), 90-day transfusion (15%), and 180-day mortality (3%) rates were observed. In contrast, in the low-risk group (Pred = 0), the rates of these events were markedly lower (< 2%). Similarly, among patients who actually revisited the emergency department (ED = 1), these event rates also remained high (approximately 49%, 17%, and 3%). These findings suggest that the risk groups identified by the model were aligned with subsequent clinical trajectories (Supplementary Material, Table SM4). 4. DISCUSSION The appropriate observation window for emergency department (ED) reattendance is critical both for quality measurement and for the design of preventive interventions. Although the commonly used 72-hour cutoff is practical, large-population data have shown that revisit patterns follow a logarithmically declining curve, with a secondary peak occurring around day 9; thus, the 72-hour metric captures only a small fraction of the overall picture. In contrast, a 30-day follow-up period encompasses both early and relatively late revisits, providing a more realistic quality indicator, and this approach was therefore adopted in our study [ 1 ]. In the literature, 30-day emergency department reattendance rates are generally reported to range between 15% and 22% [ 1 , 12 , 13 ]. In our cohort, however, this rate was notably higher at approximately 30%. Such a discrepancy may be attributable to the patient profile, comorbidity burden, and regional healthcare utilization patterns characteristic of the tertiary academic hospital in which our study was conducted. Furthermore, as our analysis included only revisits following outpatient clinic encounters, this selection may have contributed to the elevated rate. The high reattendance rate further emphasizes the clinical importance of our model, underscoring the value of early risk stratification for healthcare planning and cost-effectiveness. The use of a parsimonious, laboratory-based model provides a strong rationale in terms of both accessibility and interpretability: routine blood tests (e.g., albumin, CRP, creatinine, hemogram parameters) can distinguish short-term mortality and adverse outcomes with high accuracy, and in some studies have even outperformed formal triage algorithms [ 14 ]. Our strategy integrated these biomarkers into biologically defined axes (hematologic–inflammatory) to enable risk stratification that can be easily explained in clinical practice. Moreover, the use of a threshold predetermined on the internal test dataset and subsequently applied to the test and external validation sets without recalibration is consistent with good reporting practice [ 10 , 15 ]. In our study, vitamin B12 and folate tests, which were initially included, did not significantly alter model performance. Deficiencies in B12 and folate are known to have characteristic hematologic manifestations. Specifically, mean corpuscular volume (MCV) increases in the presence of B12 or folate deficiency due to the occurrence of macrocytosis. Red cell distribution width (RDW) increases in these deficiencies as well, reflecting anisocytosis caused by the coexistence of large macrocytes and normocytes. The combination of elevated RDW and MCV is therefore suggestive of B12 or folate deficiency. Accordingly, rather than being directly included in the model, B12 and folate deficiencies may have been adequately represented through their hematologic surrogates, namely macrocytosis and anisocytosis (MCV and RDW) [ 16 – 18 ]. Iron testing, in contrast, is well known to exhibit diurnal variation and is strongly influenced by dietary intake. In healthy individuals, serum ferritin levels show a strong correlation with iron stores [ 19 ]. In our study, however, only a weak-to-moderate correlation was observed between iron and ferritin. Given that measured serum iron levels were less stable and that ferritin, as a more reliable marker, correlated with iron while also being interpretable in the logistic regression framework, ferritin was retained as the preferred variable for the final model. In our cohort, female sex was associated with a lower risk of reattendance, consistent with findings reported in large-scale datasets [ 20 , 21 ]. This difference may be influenced by a complex interplay of clinical and sociodemographic factors, including reasons for presentation, discharge planning, healthcare-seeking behaviors, and comorbidity profiles. While causal interpretations should be avoided, it should be cautiously noted that female sex may act as a protective factor at least in the general population [ 1 ]. From a biological perspective, albumin—a widely available and rapidly measured parameter—has been shown to serve as an independent predictor of 30-day mortality in the emergency department and to provide additional clinical value particularly in patients within the low-to-intermediate Sequential/Sepsis-related Organ Failure Assessment (SOFA) ranges [ 22 , 23 ]. Similarly, the neutrophil-to-lymphocyte ratio (NLR) and other hemogram-derived indices have been associated with short-term adverse outcomes across different acute patient populations [ 24 , 25 ]. Taken together, this body of evidence suggests that our axis-based approach represents more than a purely statistical construct and in fact reflects meaningful pathophysiological mechanisms. In the literature, methodological performance values should always be interpreted within the clinical context of the endpoint under study. For example, models predicting 30-day mortality among discharged patients using rich data sources (clinical plus textual information) have achieved AUCs in the range of ~ 0.94–0.97 [ 26 ]. In contrast, predicting reattendance is inherently more challenging due to its strong behavioral and organizational components; institution-specific models for 30-day postoperative ED-related readmissions have reported AUCs of ~ 0.85–0.89, while triage-based hospital admission predictions typically remain at ~ 0.82–0.83 [ 27 , 28 ]. Within this reference framework, the discriminative performance of our laboratory-only model—operating with a prespecified threshold—together with its consistency in F1/MCC metrics (0.77/0.66), suggests a low-friction and scalable solution for clinical practice. Moreover, despite addressing a more difficult endpoint, our method achieved an AUC-ROC of 0.92/0.94 in the test and external validation sets, respectively, which represents a remarkably high level of performance. In our study, the markedly strong association of AST levels with the risk of reattendance (OR ≈ 17) was a striking and somewhat unexpected finding that warrants explanation. Although AST is primarily a liver-derived enzyme, it is a non-specific biomarker that can be elevated in various pathophysiological conditions, including muscle injury, hemolysis, drug exposure, or metabolic stress. Thus, elevated AST levels may in fact reflect an underlying comorbidity burden or systemic tissue injury. Nevertheless, the magnitude of the observed association cannot be fully accounted for by biological plausibility alone and may have been influenced by measurement-related variation or cohort-specific characteristics. Therefore, this unexpected and pronounced association between AST and reattendance risk should be retested in independent validation cohorts and further investigated with respect to underlying biological mechanisms. Interpretation of calibration findings is critical for implementation. Even when CITL is positively shifted in the test set, local prevalence or logit differences can be practically addressed through intercept recalibration, provided that the slope remains close to 1; this represents the TRIPOD-compliant first step toward field adaptation. In such cases, rather than focusing on the absolute interpretation of predicted probabilities, calibration can be strengthened through threshold-based decision-making (classification) combined with local intercept adjustment [ 15 , 29 ]. This approach provides a practical starting point for future studies aiming to further refine model implementation. Decision curve analysis (DCA) demonstrates the net benefit of a model compared with “treat-all” or “treat-none” strategies across clinically meaningful probability thresholds. In line with the foundational methodology of DCA and contemporary interpretation guidelines, the positive net benefit observed particularly within the 10–40% range provides rational evidence for interventions that may be effective at low-to-intermediate risk thresholds, such as early post-discharge telephone follow-up (within 48–72 hours), home care coordination, expedited appointment scheduling (≤ 7 days), and medication adherence counseling. Such strategies may not only reduce adverse events that could otherwise progress to life-threatening conditions before ED reattendance but also improve cost-effectiveness from a healthcare expenditure perspective [ 30 , 31 ]. From a public health perspective, overcrowding and demand volatility in the ED pose major challenges for resource management. Systematic reviews and real-time prediction models developed for revisit forecasting and patient flow management have demonstrated that hourly-to-daily uncertainty can be reduced [ 32 ]. When combined with individual-level risk stratification, these approaches enable proactive targeting of resources. Our model may facilitate focused monitoring and referral of high-risk subgroups, thereby supporting earlier interventions for potentially preventable revisits [ 33 ]. Although the primary objective of this study was to predict 30-day emergency department reattendance, exploratory and descriptive analyses revealed that patients classified as high risk by the model also exhibited markedly higher rates of hospitalization, transfusion, and mortality. These findings suggest that the model may reflect not only reattendance but also subsequent clinical trajectories. However, these results are hypothesis-generating in nature and fall outside the main scope of our study. Future logistic regression or multivariable modeling efforts incorporating these secondary outcomes may further enhance the clinical utility of the model. Limitations This study was based on routine laboratory results obtained on the day of outpatient presentation in adults presenting outside the emergency department. Therefore, the findings cannot be directly generalized to patients presenting initially to the ED or those requiring hospitalization. In the primary analyses, a complete-case approach was applied, meaning that cases with missing values for any variable required by the model were excluded. As laboratory testing in the outpatient setting is often contingent on the clinical condition, missingness may not have been completely random, raising the possibility of selection bias. Nonetheless, our decision not to apply imputation provided practical simplicity for model use and interpretation; sensitivity analyses in future studies targeting broader patient groups may better assess this risk The “external test” cohort was a held-out sample from the same institution; thus, performance may vary across different laboratory platforms, patient populations, and healthcare delivery settings. Accordingly, local intercept calibration and threshold optimization should be performed before implementation. Ideally, multicenter, prospective validation and impact studies are needed to assess generalizability and real-world utility. Furthermore, as the model relies solely on laboratory data, it does not incorporate clinical context such as symptoms, vital signs, or reasons for presentation, which may contribute to loss of specificity in certain cases. Conclusion Using an interpretable model based on routine laboratory measurements, the risk of 30-day emergency department reattendance can be stratified with clinically meaningful accuracy and a net benefit profile confirmed by decision curve analysis. With reinforcement of calibration through local intercept adjustment and threshold selection informed by stakeholder input, the model holds potential to evolve into a practical decision-support tool for proactive post-discharge monitoring and resource planning. One notable aspect of our study is that the model not only predicted reattendance but also identified risk groups that were meaningfully associated with clinical outcomes such as mortality, hospitalization, and transfusion. This finding highlights that biological stress axes reflect not only healthcare utilization but also patient prognosis. In light of these results, the AXIS-2 model could be piloted in specific high-risk populations, such as elderly patients with a high comorbidity burden, individuals with frequent outpatient visits, and patients with chronic inflammatory or hematologic conditions. Early risk stratification in these groups may facilitate more targeted implementation of post-discharge follow-up programs (e.g., telephone monitoring, home care coordination, or expedited appointments). In this way, the model has the potential not only to enhance patient safety at the individual level but also to reduce ED overcrowding and healthcare costs at the system level. Declarations Appendix / Supplementary Material Additional tables and figures supporting the findings of this study are provided in the Supplementary Material (Table SM1-SM4, Figure SM1–SM2). These materials include reference ranges, correlation analyses, calibration plots, and subgroup results. Ethics approval and consent to participate This study was conducted in accordance with the Declaration of Helsinki and approved by the Non-Interventional Clinical Research Ethics Committee of the İzmir Tepecik Training and Research Hospital, University of Health Sciences Türkiye (Approval No: 2025/08-19, dated 11 September 2025). The requirement for informed consent was waived by the ethics committee because the study used retrospective, anonymized laboratory data with no identifiable personal information. Consent for publication Not applicable. No individual person’s data in any form (including images, videos, or identifiable information) are included in this manuscript . Availability of data and materials The datasets generated and analyzed during the current study are not publicly available due to institutional data protection regulations but are available from the corresponding author (FD) upon reasonable request. All key summary tables and model coefficients are included in this published article and its supplementary information files. Competing interests The authors declare that they have no competing interests. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Authors’ contributions FD (Ferhat Demirci) obtained ethics approval, conceptualized the study, curated and analyzed the data, developed the methodology, performed the statistical analysis, and drafted the original manuscript. AD (Aylin Demirci) contributed to data collection and participated in the review and editing of the manuscript. Both authors read and approved the final version of the manuscript. Acknowledgements The authors express their sincere gratitude to Prof. Dr. Savaş Yakan, Chief Physician of İzmir Tepecik Training and Research Hospital, and Prof. Dr. Ayfer Çolak, Head of the Department of Medical Biochemistry, for their valuable support and encouragement throughout this study. Footnotes Not applicable. References Rising KL, Victor TW, Hollander JE, Carr BG. Patient Returns to the Emergency Department: The Time-to-return Curve. Academic Emergency Medicine. 2014 Aug;21(8):864–71. Frąckiewicz J, Włodarek D, Brzozowska A, Wierzbicka E, Słowińska MA, Wądołowska L, et al. Hematological parameters and all-cause mortality: a prospective study of older people. Aging Clin Exp Res. 2018 May 29;30(5):517–26. Kim S, Lee K, Kim I, Jung S, Kim MJ. Red cell distribution width and early mortality in elderly patients with severe sepsis and septic shock. Clin Exp Emerg Med. 2015 Sep;2(3):155–61. Liu X, Shen Y, Wang H, Ge Q, Fei A, Pan S. Prognostic Significance of Neutrophil-to-Lymphocyte Ratio in Patients with Sepsis: A Prospective Observational Study. Mediators Inflamm. 2016;2016:8191254. Liu Y, Gao Y, Liang B, Liang Z. The prognostic value of C-reactive protein to albumin ratio in patients with sepsis: a systematic review and meta-analysis. Aging Male. 2023 Dec;26(1):2261540. Brossard C, Goetz C, Catoire P, Cipolat L, Guyeux C, Gil Jardine C, et al. Predicting emergency department admissions using a machine-learning algorithm: a proof of concept with retrospective study. BMC Emerg Med. 2025 Jan 6;25(1):3. Raita Y, Goto T, Faridi MK, Brown DFM, Camargo CA, Hasegawa K. Emergency department triage prediction of clinical outcomes using machine learning models. Crit Care. 2019 Feb 22;23(1):64. Collins GS, Reitsma JB, Altman DG, Moons K. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD Statement. BMC Med. 2015;13(1):1. Wolff RF, Moons KGM, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies. Ann Intern Med. 2019 Jan 1;170(1):51–8. Vickers AJ, Elkin EB. Decision Curve Analysis: A Novel Method for Evaluating Prediction Models. Medical Decision Making. 2006 Nov 1;26(6):565–74. Van Calster B, McLernon DJ, van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. BMC Med. 2019 Dec 16;17(1):230. Kim H, Han SJ, Lee JH, Lim J, Moon S do, Moon H, et al. A Descriptive Study of Emergency Department Visits Within 30 Days of Discharge. Ann Geriatr Med Res. 2021 Dec 31;25(4):245–51. Brown JF, Fu J. Emergency Department Avoidance by Transgender Persons: Another Broken Thread in the “Safety Net” of Emergency Medicine Care. Ann Emerg Med. 2014 Jun;63(6):721–2. Kristensen M, Iversen AKS, Gerds TA, Østervig R, Linnet JD, Barfod C, et al. Routine blood tests are associated with short term mortality and can improve emergency department triage: a cohort study of >12,000 patients. Scand J Trauma Resusc Emerg Med. 2017 Dec 28;25(1):115. Vickers AJ, van Calster B, Steyerberg EW. A simple, step-by-step guide to interpreting decision curve analysis. Diagn Progn Res. 2019 Dec 4;3(1):18. Maner BS, Killeen RB, Moosavi L. Mean Corpuscular Volume. 2025. Epstein-Peterson ZD, Chokshi I, Barrow B, Lobaugh S, Devlin S, Fenelus M, et al. Laboratory evaluation of folate deficiency among inpatients with cancer. Int J Lab Hematol. 2021 Aug;43(4):O164–7. Rajashekar RB, Patel S, Kulkarrni P. Discriminant Functions in the Diagnosis of Vitamin B12 Deficiency Anemia, the Value of RDW-SD: An Analytical Study. National Journal of Laboratory Medicine. 2017;6(1):PO01–6. Rohr M, Brandenburg V, Brunner-La Rocca HP. How to diagnose iron deficiency in chronic disease: A review of current methods and potential marker for the outcome. Eur J Med Res. 2023 Jan 9;28(1):15. Tsai IT, Sun CK, Chang CS, Lee KH, Liang CY, Hsu CW. Characteristics and outcomes of patients with emergency department revisits within 72 hours and subsequent admission to the intensive care unit. Tzu Chi Med J. 2016;28(4):151–6. Wang HC, Fang CC, Huang CH, Gao JW, Chen JH, Tsai CL. Factors associated with overall and high-risk return visits to the emergency department: a vital sign trajectory approach. BMC Emerg Med. 2025 Apr 12;25(1):57. Turcato G, Zaboli A, Sibilio S, Fanni Canelles MF, Rella E, Giudiceandrea A, et al. Prognostic Role of Serum Albumin in Predicting 30-Day Mortality in Patients with Infections in Emergency Department: A Prospective Study. J Clin Med. 2023 May 13;12(10):3447. Turcato G, Zaboli A, Sibilio S, Mian M, Brigo F. The Clinical Utility of Albumin with Sequential Organ Failure Assessment (SOFA) in Improving 30-Day Mortality Prediction in Patients with Infection in the Emergency Department. J Clin Med. 2023 Dec 14;12(24):7676. Soulaiman SE, Dopa D, Raad ABT, Hasan W, Ibrahim N, Hasan AY, et al. Cohort retrospective study: the neutrophil to lymphocyte ratio as an independent predictor of outcomes at the presentation of the multi-trauma patient. Int J Emerg Med. 2020 Dec 4;13(1):5. Galardo G, Crisanti L, Gentile A, Cornacchia M, Iatomasi F, Egiddi I, et al. Neutrophil to lymphocyte ratio (NLR) and short-term mortality risk in elderly acute medical patients admitted to a University Hospital Emergency Department. Intern Emerg Med. 2025 Mar;20(2):553–62. Barash Y, Soffer S, Grossman E, Tau N, Sorin V, BenDavid E, et al. Alerting on mortality among patients discharged from the emergency department: a machine learning model. Postgrad Med J. 2022 Mar 1;98(1157):166–71. Parker CA, Liu N, Wu SX, Shen Y, Lam SSW, Ong MEH. Predicting hospital admission at the emergency department triage: A novel prediction model. Am J Emerg Med. 2019 Aug;37(8):1498–504. Mišić V V., Gabel E, Hofer I, Rajaram K, Mahajan A. Machine Learning Prediction of Postoperative Emergency Department Hospital Readmission. Anesthesiology. 2020 May;132(5):968–80. Van Calster B, McLernon DJ, van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. BMC Med. 2019 Dec 16;17(1):230. Lukanski A, Watters S, Bilderback AL, Buchanan D, Hodges JC, Burwell D, et al. Implementing a Discharge Follow-up Phone Call Program Reduces Readmission Rates in an Integrated Health System. Journal for Healthcare Quality. 2023 Nov;45(6):315–23. Bilicki DJ, Reeves MJ. Outpatient Follow-Up Visits to Reduce 30-Day All-Cause Readmissions for Heart Failure, COPD, Myocardial Infarction, and Stroke: A Systematic Review and Meta-Analysis. Prev Chronic Dis. 2024 Sep 26;21:240138. Asheim A, Bache-Wiig Bjørnsen LP, Næss-Pleym LE, Uleberg O, Dale J, Nilsen SM. Real-time forecasting of emergency department arrivals using prehospital data. BMC Emerg Med. 2019 Dec 5;19(1):42. Silva E, Pereira MF, Vieira JT, Ferreira‐Coimbra J, Henriques M, Rodrigues NF. Predicting hospital emergency department visits accurately: A systematic review. Int J Health Plann Manage. 2023 Jul 10;38(4):904–17. Tables Table 1. Descriptive statistics of the datasets Features Unit Train Test Test Set External Test Set P value Gender All n 55,431 (%100) 18,477 (%100) 18,478 (%100) >0.05 Male n 22,742 (%41.03) 7,562 (%40.93) 7,605 (%41.16) >0.05 Female n 32,689 (%58.97) 10,915 (%59.07) 10,873 (%58.84) >0.05 Age All years 52.93±19.46 53.32±19.32 52.90±19.46 >0.05 Male years 54.06±19.56 54.40±19.31 53.99±19.54 >0.05 Female years 52.14±19.36 52.57±19.28 52.14±19.36 >0.05 Hemoglobin g/dL 13.00±1.99 12.99±2.00 13.00±1.99 >0.05 White blood cell 10³/µL 8.22±9.44 8.17±8.44 8.21±9.61 >0.05 Neutrophil 10³/µL 4.79±3.45 4.78±3.13 4.76±3.14 >0.05 Lymphocyte 10³/µL 2.66±8.23 2.61±7.32 2.66±8.46 >0.05 Platelet 10³/µL 266.86±109.05 266.82±110.65 266.08±108.91 >0.05 Neutrophil/ Lymphocyte Ratio 2.55 ± 2.43 2.54 ± 2.31 2.55±2.57 >0.05 Platelet/Lymphocyte Ratio 140.87±93.27 141.05±94.12 140.43±91.37 >0.05 Mean Corpuscular Volume fL 86.14±7.75 86.08±7.79 86.09±7.74 >0.05 Red Distribution Width % 14.79±2.68 14.82±2.76 14.78±2.68 >0.05 C-Reactive Protein mg/L 10.61±24.95 11.12±27.56 10.81±26.21 >0.05 Albumin g/L 42.62±14.62 42.36±6.72 42.41±9.37 0.017 - Test Ferritin ng/mL 107.58±277.35 114.06±345.11 105.04±239.16 0.0058 - test Iron µg/dL 173.90±149.41 172.06±148.56 174.39±157.42 >0.05 B12 pg/mL 296.90±244.84 297.69±242.88 298.97±245.70 >0.05 Folate ng/mL 8.68±4.44 8.65±4.40 8.65±4.40 >0.05 Creatinine mg/dL 1.01±0.79 1.02±0.84 1.02±0.83 >0.05 Aspartate Aminotransferase U/L 25.45±55.41 25.61±52.61 26.88±250.71 >0.05 Alanine Aminotransferase U/L 24.96±52.43 25.39±63.52 24.74±51.38 >0.05 EMERGENCY30 0 40,594 (%73.23) 12,899 (%69.81) 13,459 (%72.84) * Test 1 14,837 (%26.77) 5,578 (%30.19) 5,019 (%27.16) * Test *For groups 0 and 1: in the Train vs Test comparison, p = 2.25×10⁻¹⁹; in Train vs External, p = 0.298; and in Test vs External, p = 1.35×10⁻¹⁰. Table 2. Final Logistic Regression Model, Biological Axis Components, and Multivariable Results A. Final Logistic Regression Model and Biological Axis Components Regression Formula logit(p) = β₀ + β₁E1_score + β₂E2_score + β₃Age + β₄Creatinine + β₅ALT + β₆AST + Σβg·1(Gender category) p = 1 / (1 + e −logit(p) ) E1 (Hematology) z(HGB)^−, z(LYM)^−, z(MCV)^+, z(NEU)^+, z(PLT)^+, z(RDW)^+, z(WBC)^+ → mean E2 (Inflammation) z(CRP)^+, z(Ferritin)^+, z(Albumin)^−, z(NLR)^+, z(PLR)^+ → mean B. Multivariable Logistic Regression Results Feature B (beta) OR (95% CI) p-value Intercept -0.959 0.38 (0.37 – 0.40) <0.001 E1 score 0.724 2.06 (2.00 – 2.13) <0.001 E2 score 0.838 2.31 (2.17 – 2.46) <0.001 Age 1.442 4.23 (4.07 – 4.40) <0.001 Creatinine 1.061 2.89 (2.71 – 3.08) <0.001 ALT 0.361 1.44 (1.32 – 1.56) <0.001 AST 2.845 17.20 (14.68 – 20.15) <0.001 Female (ref=male) -1.053 0.35 (0.33 – 0.37) <0.001 HGB: Hemoglobin; LYM: Lymphocyte count; MCV: Mean corpuscular volume; NEU: Neutrophil count; PLT: Platelet count; RDW: Red cell distribution width; WBC: White blood cell count; CRP: C-reactive protein; NLR: Neutrophil-to-lymphocyte ratio; PLR: Platelet-to-lymphocyte ratio. When calculating Z-scores, the (+) sign indicates that an increase in the relevant parameter is effective in increasing the risk of emergency room visits, while the (-) sign indicates that the increase is effective in reducing the risk. The superscript (^) sign indicates that the variable is used in the model with this risk aspect. Continuous variables are standardized (z-score); ORs are for a 1 SD increase. Table 3. Logistic regression performance results Test Set External Test Set TP 4752 4414 FP 1948 2128 FN 826 605 TN 10951 11331 Sensitivity 0,852 [0.842, 0.861] 0,879 [0.870, 0.888] Specificity 0,849 [0.843, 0.855] 0,842 [0.836, 0.848] PPV 0,709 [0.699, 0.720] 0,675 [0.665, 0.685] NPV 0,930 [0.925, 0.934] 0,949 [0.945, 0.953] PLR 5,641 [5.414, 5.866] 5,562 [5.370, 5.780] NLR 0,174 [0.164, 0.186] 0,143 [0.133, 0.154] Accuracy 0,850 [0.845, 0.855] 0,852 [0.847, 0.857] F1 Score 0,774 [0.766, 0.782] 0,764 [0.756, 0.772] MCC 0,669 [0.658, 0.680] 0,671 [0.661, 0.681] AUC-ROC 0,921 [0.916, 0.924] 0,935 [0.931, 0.938] AUC-PR 0,817 [0.805, 0.828] 0,835 [0.824, 0.846] Diagnostic Odds Ratio 32,342 [29.611, 35.324] 38,848 [35.265, 42.796] Relative Risk 10,112 [9.452, 10.819] 13,311 [12.295, 14.412] Brier Score 0,102 [0,099, 0,105] 0,087 [0,085, 0,089] R 2 _McFadden 0,461 [0.447, 0.475] 0,512 [0.500, 0.524] R 2 _CoxSnell 0,431 [0.421, 0.441] 0,451 [0.442, 0.460] R 2 _Nagelkerke 0,611 [0.597, 0.624] 0,654 [0.642, 0.665] TP: True Positive, FP: False Positive, FN: False Negative, TN: True Negative, PPV: Positive Predictive Value, NPV: Negative Predictive Value, PLR: Positive Likelihood Ratio, NLR: Negative Likelihood Ratio, Accuracy: Overall Correct Classification Rate, F1 Score: Harmonic Mean of Precision and Recall, MCC: Matthews Correlation Coefficient, AUC-ROC: Area Under the Receiver Operating Characteristic Curve, AUC-PR: Area Under the Precision–Recall Curve. Additional Declarations No competing interests reported. Supplementary Files SM.docx Cite Share Download PDF Status: Published Journal Publication published 13 Feb, 2026 Read the published version in BMC Emergency Medicine → Version 1 posted Editorial decision: Revision requested 19 Jan, 2026 Reviews received at journal 13 Jan, 2026 Reviews received at journal 10 Jan, 2026 Reviewers agreed at journal 06 Jan, 2026 Reviewers agreed at journal 05 Jan, 2026 Reviewers agreed at journal 22 Dec, 2025 Reviews received at journal 19 Dec, 2025 Reviewers agreed at journal 29 Nov, 2025 Reviewers agreed at journal 26 Nov, 2025 Reviewers invited by journal 05 Nov, 2025 Editor invited by journal 04 Nov, 2025 Editor assigned by journal 01 Nov, 2025 Submission checks completed at journal 01 Nov, 2025 First submitted to journal 29 Oct, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7980356","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":545652575,"identity":"5c826f10-ad20-47d9-af9a-e89d5acdac66","order_by":0,"name":"Ferhat Demirci","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4ElEQVRIiWNgGAWjYDCCAwwGDCDExsz/8AGQz8NHvBb2HmYDkBY24rSAAM8ZNgkQTVAL3+3DGx/dKLgnzyeRe6zya46dDBsD88NHN/BokTyXVmycY1Bs2CaRl3Zbdlsy0GFsxsY5eLQYnOExk84xSEhgk0gwuy25jRmohYdNmoAW898wLcWS2+qJ0mLGDNbCc8aM8eO2w4S1SJ5hKwY5zLCNvS1ZmnHbcR42ZgJ+4TvDvPFzzp8Eeflm5oMff26rtudnb374GJ8WFMDMAyaJVQ4CjD9IUT0KRsEoGAUjBgAAvdpBL5N0qWoAAAAASUVORK5CYII=","orcid":"","institution":"University of Health Sciences","correspondingAuthor":true,"prefix":"","firstName":"Ferhat","middleName":"","lastName":"Demirci","suffix":""},{"id":545652576,"identity":"8985d963-ffc2-417e-8f0a-bfe7a564d345","order_by":1,"name":"Aylin Demirci","email":"","orcid":"","institution":"University of Health Sciences","correspondingAuthor":false,"prefix":"","firstName":"Aylin","middleName":"","lastName":"Demirci","suffix":""}],"badges":[],"createdAt":"2025-10-29 13:23:34","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7980356/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7980356/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12873-026-01496-w","type":"published","date":"2026-02-13T15:59:22+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":96248122,"identity":"cd7255e7-25f0-4250-93b3-473815d13620","added_by":"auto","created_at":"2025-11-19 07:28:05","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":471113,"visible":true,"origin":"","legend":"","description":"","filename":"figrler.docx","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/50053bd6a013b57b60e8500d.docx"},{"id":96077884,"identity":"1d7009fa-0683-4182-9137-bced3ff3e04e","added_by":"auto","created_at":"2025-11-17 10:54:58","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":168004,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/d713a863e7ddd74b9d41182e.docx"},{"id":96077900,"identity":"42e53aa2-e485-400f-91f7-ce49f4fa5505","added_by":"auto","created_at":"2025-11-17 10:55:01","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":29927,"visible":true,"origin":"","legend":"","description":"","filename":"Tables.docx","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/133a5387df4eca8d1adffe4d.docx"},{"id":96077885,"identity":"37a0d3e4-c8c0-4135-a2eb-81ca5ca11727","added_by":"auto","created_at":"2025-11-17 10:54:58","extension":"json","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5929,"visible":true,"origin":"","legend":"","description":"","filename":"d60cf93b73e845b2b15302e121a248e2.json","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/793a665647a67e6dc6df4ee5.json"},{"id":96077901,"identity":"f4dab02f-204a-4250-9d0c-d7d7783df2ac","added_by":"auto","created_at":"2025-11-17 10:55:01","extension":"docx","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":690260,"visible":true,"origin":"","legend":"","description":"","filename":"SM.docx","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/fa2c3fe204ddbdf125cceaa4.docx"},{"id":96077897,"identity":"7b6d6fe5-375f-454a-aa66-e21d3504bc6b","added_by":"auto","created_at":"2025-11-17 10:55:01","extension":"xml","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":134312,"visible":true,"origin":"","legend":"","description":"","filename":"d60cf93b73e845b2b15302e121a248e21enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/2d6c3f98f635122698d642e4.xml"},{"id":96077887,"identity":"925f40df-ebcd-485e-b5e4-ec1d75f0efc8","added_by":"auto","created_at":"2025-11-17 10:54:59","extension":"jpeg","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":412100,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/51e96b79f2657c407dec2d9f.jpeg"},{"id":96077903,"identity":"8422502c-87e7-4463-9633-833f286896b9","added_by":"auto","created_at":"2025-11-17 10:55:02","extension":"jpeg","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":420135,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/ea3dec14600ed8af008ae48b.jpeg"},{"id":96077881,"identity":"0fce6c15-7119-4967-b10e-80f5c3636560","added_by":"auto","created_at":"2025-11-17 10:54:56","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":69277,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/c9283d093041c11b1a0b822f.png"},{"id":96077880,"identity":"ae3cac60-4bb6-4261-8b36-92cd2aff41ce","added_by":"auto","created_at":"2025-11-17 10:54:55","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":70958,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/e03bf4f8b976527b1563a6ac.png"},{"id":96077889,"identity":"959bd4b0-82ea-43aa-97b4-e419485cd76f","added_by":"auto","created_at":"2025-11-17 10:54:59","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":273152,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/16b3338ec2bcf2a295a0c428.png"},{"id":96249606,"identity":"54ce88a5-4a62-46cc-adc7-182ea59b812f","added_by":"auto","created_at":"2025-11-19 07:35:38","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":97787,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/c0bc2f0feadeedfdac6592d1.png"},{"id":96077875,"identity":"95550ed6-ba4c-45f8-8fe3-c4cce347e649","added_by":"auto","created_at":"2025-11-17 10:54:54","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":21113,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/a75acda71a33a8533531d97a.png"},{"id":96077886,"identity":"4c4dd174-9181-4cd6-8962-4e1e514d1f21","added_by":"auto","created_at":"2025-11-17 10:54:58","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":21775,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/1b51ebc35b26f097b3a2a9ea.png"},{"id":96077891,"identity":"f800165d-639f-4f8f-8b40-8217cf7bdc06","added_by":"auto","created_at":"2025-11-17 10:55:00","extension":"xml","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":131957,"visible":true,"origin":"","legend":"","description":"","filename":"d60cf93b73e845b2b15302e121a248e21structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/937d9af3ba43c4f60e4e24a6.xml"},{"id":96077902,"identity":"543e8cfe-06b4-4637-91ab-6529c5d53bd3","added_by":"auto","created_at":"2025-11-17 10:55:01","extension":"html","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":141307,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/2ff06a8c84411e2989b6faba.html"},{"id":96077892,"identity":"68100e02-49c0-4360-8c77-6102ffb62cc6","added_by":"auto","created_at":"2025-11-17 10:55:00","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":125064,"visible":true,"origin":"","legend":"\u003cp\u003eThe Standards for Reporting Diagnostic Accuracy diagram\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/38ea6f90180dd585c8024721.png"},{"id":96248226,"identity":"0283afdd-1a80-42a6-9d1a-d9d6690a9bab","added_by":"auto","created_at":"2025-11-19 07:28:11","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":74294,"visible":true,"origin":"","legend":"\u003cp\u003eTest set and External Test Set AUC-ROC and AUC-PR plots\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/e417c02ebbfe38acf2b28473.png"},{"id":96077893,"identity":"1e103785-103d-452e-bf30-7dea0dcb3e81","added_by":"auto","created_at":"2025-11-17 10:55:00","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":43699,"visible":true,"origin":"","legend":"\u003cp\u003eDecision curve analysis of the logistic regression model in the test and external test datasets.\u003c/p\u003e\n\u003cp\u003eNB: Net benefit (y-axis) represents the clinical value of the model at a given threshold probability (x-axis) compared with “treat-all” and “treat-none” strategies. The blue line (Model NB) indicates the logistic regression model, the orange dashed line (Treat-all) represents intervention for all patients, and the green dotted line (Treat-none) represents no intervention. DCA: Decision Curve Analysis.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/e32932e82e91ba840dc8b4a3.png"},{"id":102785989,"identity":"bd9afbf8-617c-4c57-b572-85352c75ade2","added_by":"auto","created_at":"2026-02-16 16:11:22","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1066273,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/4789071d-4c19-4432-bd79-e1ae065cbb08.pdf"},{"id":96077876,"identity":"452c8453-01fe-4478-9596-fb1dd13e8d49","added_by":"auto","created_at":"2025-11-17 10:54:54","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":690260,"visible":true,"origin":"","legend":"","description":"","filename":"SM.docx","url":"https://assets-eu.researchsquare.com/files/rs-7980356/v1/7493e3932766f9b9f914166d.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Routine Laboratory Data for Predicting 30-Day Emergency Department Revisits: The AXIS-2 Risk Score","fulltext":[{"header":"1. INTRODUCTION","content":"\u003cp\u003eA substantial proportion of patients presenting to outpatient clinics return to the emergency department within a short period of time. This phenomenon not only increases healthcare utilization but also reflects an elevated risk of morbidity and mortality. Therefore, early prediction of emergency department revisit risk using laboratory data obtained at the time of the outpatient encounter carries critical clinical importance. Similarly, among patients discharged from the emergency department, 30-day revisit is common and represents a key quality indicator. While the conventional 72-hour metric fails to capture many events, the 30-day revisit rate has been reported to be approximately 20% in previous studies [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn the context of revisits, anemia and inflammation emerge as two principal biological axes. Anemia contributes to adverse prognosis through alterations in erythrocyte indices and a reduction in oxygen-carrying capacity. In particular, parameters such as hemoglobin (HGB), red cell distribution width (RDW), and mean corpuscular volume (MCV) have been shown to be associated with short-term clinical outcomes [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Inflammation, on the other hand, can be readily monitored through parameters such as elevated C-reactive protein (CRP), decreased albumin, and an increased neutrophil-to-lymphocyte ratio (NLR). These markers have been found to be valuable predictors of short-term mortality and revisit risk in numerous studies [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn recent years, machine learning models have demonstrated promising results in predicting emergency department visits [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. However, the interpretability of these models remains limited, making their widespread adoption as decision-support tools by clinicians challenging. Therefore, there is a need for scoring systems that not only provide strong predictive performance but are also computationally feasible and transparent. In this context, multivariable models such as logistic regression are widely used, as they offer interpretability and transparency. For studies of this nature, the TRIPOD [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] and PROBAST [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] frameworks are recommended for reporting and risk of bias assessment. Moreover, calibration metrics\u0026mdash;including calibration-in-the-large (CITL) and calibration slope\u0026mdash;should always be reported and recalibrated when necessary. To illustrate clinical utility in a threshold-based manner, Decision Curve Analysis, which relies on the net benefit approach, is commonly advocated [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn this study, we aimed to develop a novel risk score, termed AXIS-2 (Anemia\u0026ndash;Inflammation Two-Axis Composite Index). The objectives were to evaluate the predictive performance of the AXIS-2 score in forecasting 30-day emergency department revisits among outpatients, to test its validity in an independent dataset, and to present it as a transparent risk assessment tool that may contribute to clinical decision-making. Additionally, this study investigated comparative outcomes in patients identified by AXIS-2 as being at high risk of revisit, including hospitalization rates, transfusion requirements, and mortality during follow-up.\u003c/p\u003e"},{"header":"2. MATERIALS AND METHODS","content":"\u003cp\u003e2.1 \u0026nbsp;Study Population / Subjects\u003c/p\u003e\n\u003cp\u003eThis study was conducted at SBÜ İzmir Tepecik Training and Research Hospital, a tertiary care academic center, and its affiliated service building. The primary dataset included patients who presented on an outpatient basis, excluding emergency department visits, between January 1, 2015, and August 31, 2025.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInclusion criteria:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e1.\u0026nbsp; \u0026nbsp;Patients aged \u0026gt;18 years.\u003c/p\u003e\n\u003cp\u003e2.\u0026nbsp; \u0026nbsp;Patients presenting to non-emergency outpatient clinics of our hospital between January 1, 2015, and August 31, 2025 (only the first eligible visit was considered).\u003c/p\u003e\n\u003cp\u003e3.\u0026nbsp; \u0026nbsp;Patients for whom the following laboratory tests were requested completely and without missing values: hemoglobin (HGB), white blood cell count (WBC), absolute neutrophil count (NEU), absolute lymphocyte count (LYM), platelet count (PLT), red cell distribution width (RDW), mean corpuscular volume (MCV), C-reactive protein (CRP), vitamin B12, folate, ferritin, iron, albumin, creatinine, alanine aminotransferase (ALT), and aspartate aminotransferase (AST).\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Exclusion criteria:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e1. Patients with outpatient visits other than the first eligible outpatient encounter during the study period (to avoid duplicate results).İlk uygun poliklinik başvurusu öncesi aynı gün acil başvurusu olan hastalar\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e2. Patients with an emergency department visit on the same day prior to the first eligible outpatient encounter.\u003c/p\u003e\n\u003cp\u003e3. Patients younger than 18 years of age.\u003c/p\u003e\n\u003cp\u003e4. Patients whose residential address was transferred outside of İzmir in the Central Population Management System of Turkey (MERNIS) registry within 30 days.\u003c/p\u003e\n\u003cp\u003e5. Patients with missing laboratory test results.\u003c/p\u003e\n\u003cp\u003e6. Patients whose laboratory test results did not contain numeric data.\u003c/p\u003e\n\u003cp\u003e7. Patients who did not provide a sample within 2 hours after the outpatient visit.\u003c/p\u003e\n\u003cp\u003e8. Patients with a lymphocyte count of zero, precluding calculation of NLR and PLR.\u003c/p\u003e\n\u003cp\u003e9. Patients with more than one protocol number recorded on the same day.\u003c/p\u003e\n\u003cp\u003e10. Pregnant patients.\u003c/p\u003e\n\u003cp\u003e11. Patients with active oncologic disease or hematological malignancy.\u003c/p\u003e\n\u003cp\u003e12. Forensic cases.\u003c/p\u003e\n\u003cp\u003e13. Patients diagnosed with acute coronary syndrome (ACS), pulmonary embolism (PE), aortic dissection/aneurysm rupture, subarachnoid hemorrhage/stroke-TIA, or acute massive gastrointestinal bleeding (as these conditions require immediate spontaneous emergency department admission).\u003c/p\u003e\n\u003cp\u003eWhole blood samples were collected into K2-EDTA tubes by phlebotomy nurses and analyzed within 30 minutes using a Sysmex XN-1000 analyzer (Kobe, Japan). Complete blood count (CBC) reagents and calibrators were obtained from the respective manufacturer, certified, and officially registered products, while internal quality control materials were supplied by Sysmex Corporation (Kobe, Japan). Serum for biochemical assays, including C-reactive protein (CRP), ferritin, iron, albumin, creatinine, alanine aminotransferase (ALT), and aspartate aminotransferase (AST), was obtained from plain tubes after centrifugation and analyzed within 2 hours on a Beckman Coulter AU-5800 analyzer (California, USA). All reagents and calibrators for biochemical measurements were provided by the respective manufacturer, certified, and officially registered, with internal quality control materials supplied by Bio-Rad Corporation (California, USA).\u003c/p\u003e\n\u003cp\u003e2.2 \u0026nbsp;Study Design\u003c/p\u003e\n\u003cp\u003ePrior to the initiation of the study, approval was obtained from the Non-Interventional Ethics Committee of SBÜ İzmir Tepecik Training and Research Hospital (Decision Date 11.09.2025 and No: 2025 / 08-19). To ensure data confidentiality, all patient identifiers were anonymized. A dataset including age, sex, complete blood count (CBC) parameters—white blood cell count (WBC), neutrophil count (NEU), lymphocyte count (LYM), platelet count (PLT), hemoglobin (HGB), red cell distribution width (RDW), and mean corpuscular volume (MCV)—as well as calculated ratios such as the neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR), and biochemical tests including C-reactive protein (CRP), vitamin B12, folate, ferritin, iron, albumin, creatinine, alanine aminotransferase (ALT), and aspartate aminotransferase (AST), was constructed from the records of 131,867 patients and transferred to Microsoft Excel 2021 (USA) for analysis.\u003c/p\u003e\n\u003cp\u003eAmong the numerous tests included in the complete blood count and basic biochemistry panel, we selected parameters with established prognostic relevance supported by the literature and high accessibility in routine clinical practice. These variables were grouped into two axes: Axis-1 (E1), representing indicators related to anemia and erythropoiesis (Hb, LYM, MCV, NEU, PLT, RDW, WBC), and Axis-2 (E2), reflecting inflammation and catabolism (CRP, ferritin, albumin, NLR, PLR). In this way, the model was designed to capture a biologically meaningful structure that integrates two fundamental processes in a comprehensive manner. Reference ranges for the laboratory tests included in the study are provided in Supplementary Material, Table SM1.\u003c/p\u003e\n\u003cp\u003eAfter applying the exclusion criteria, the final dataset comprised 92,386 patients. This dataset was transferred to the Python environment (version 3.11, USA) for regression analysis. Non-numeric categorical variables were converted into numerical values as follows: male = 1, female = 2; 30-day emergency department revisit (Emergency30) = 0/1. The distribution of Emergency30 events was examined to assess class balance in the cleaned dataset. To prevent overfitting and preserve class proportions during dataset partitioning, a stratified sampling approach was employed. In addition, a class-weighting method was applied during model development to mitigate class imbalance. Following data preprocessing, the dataset was randomly divided into three parts using stratified sampling based on the binary outcome variable while maintaining the original class distribution: 60% training set, 20% internal test set (hereafter referred to as the test set), and 20% external test set. The number of patients allocated to each dataset (training, test, external test) is detailed in a flow diagram (Figure 1), in accordance with the TRIPOD guidelines.\u003c/p\u003e\n\u003cp\u003e2.3 \u0026nbsp;Data Preprocessing and\u0026nbsp;Training of Machine Learning Algorithms\u003c/p\u003e\n\u003cp\u003eThe cleaned dataset was processed in Python 3.11 (USA) using both Kaggle Notebook (USA) and Google Colab (USA) platforms. Logistic regression (LR) was employed for classification, with\u0026nbsp;\u003cem\u003eclass_weight = \"balanced\"\u003c/em\u003e applied to mitigate class imbalance. The input set consisted of Axis-1 (hematology axis), Axis-2 (inflammation axis), sex, age, creatinine, ALT, and AST variables. The axes were defined by z-scoring the corresponding laboratory variables based on the training set, multiplying each variable by a sign reflecting its clinical risk direction (i.e., positive for risk-increasing markers and negative for protective ones), and calculating the composite axis score as the mean of these terms. Continuous variables were standardized using the “StandardScaler” function, and categorical variables were encoded using one-hot encoding. Stratified sampling was used during dataset partitioning to preserve class proportions across the sets.\u003c/p\u003e\n\u003cp\u003eThe model was trained on the development (training) dataset and internally validated using an independent internal test set derived from the same source population. To improve the calibration of predicted probabilities, isotonic calibration was applied, ensuring better agreement between predicted probabilities and observed outcomes. The optimal decision threshold was determined by maximizing the Youden J index on the internal test set. The finalized model, together with this fixed threshold, was subsequently evaluated on an external validation dataset to assess its generalizability and real-world applicability.\u003c/p\u003e\n\u003cp\u003eIn all analyses, the same predefined threshold derived from the internal test set was used for evaluation of both the internal and external validation datasets, without further recalibration.\u003c/p\u003e\n\u003cp\u003e2.4 \u0026nbsp;Performance Evaluation\u003c/p\u003e\n\u003cp\u003eThe logistic regression model was comprehensively evaluated using both the test set and the external test set. To ensure the reliability and clinical interpretability of the model, a multidimensional set of performance metrics was employed.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClassification Performance Metrics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e· Area under the receiver operating characteristic curve (AUC-ROC)\u003c/p\u003e\n\u003cp\u003e· Area under the precision–recall curve (AUC-PR)\u003c/p\u003e\n\u003cp\u003e· Confusion matrix analysis at the Youden J threshold\u003c/p\u003e\n\u003cp\u003e· Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, and F1 score\u003c/p\u003e\n\u003cp\u003e· Matthews correlation coefficient (MCC)\u003c/p\u003e\n\u003cp\u003e· Likelihood ratios (LR⁺/LR⁻) and diagnostic odds ratio (DOR)\u003c/p\u003e\n\u003cp\u003e· Relative risk (RR) with 95% confidence intervals (CIs)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCalibration and Clinical Utility Metrics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e· Calibration-in-the-large (CITL) and calibration slope (with 95% CIs)\u003c/p\u003e\n\u003cp\u003e· Calibration plots comparing predicted and observed risks\u003c/p\u003e\n\u003cp\u003e· Decision Curve Analysis (DCA) to evaluate net clinical benefit across a range of probability thresholds\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eValidation Results\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e· The optimal decision threshold was determined by maximizing the Youden J index on the internal test set and then directly applied to the external validation set without any further adjustment.\u003c/p\u003e\n\u003cp\u003e· All primary performance metrics were reported with 95% confidence intervals obtained through bootstrapping.\u003c/p\u003e\n\u003cp\u003e· Additional analyses included threshold sweeps (±20% around the Youden J point) and subgroup evaluations (sex, age tertiles, anemia status) to enhance the clinical interpretability of the model.\u003c/p\u003e\n\u003cp\u003eThis structured and multidimensional evaluation process demonstrated both the discriminative ability and the clinical applicability of the logistic regression model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.5 \u0026nbsp;Statistical Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eComparisons across the three datasets (training, test, and external test) were performed using one-way ANOVA for continuous variables and the Pearson chi-square test for categorical variables. For variables with statistically significant differences, post-hoc pairwise comparisons were conducted to identify the source of variation. Continuous variables were standardized using z-scores, and odds ratios (ORs) were reported per one standard deviation increase. All statistical analyses were conducted in Python (v3.11) using the \u003cem\u003epandas\u003c/em\u003e, \u003cem\u003escipy\u003c/em\u003e, and \u003cem\u003estatsmodels\u003c/em\u003e libraries, with two-sided p-values \u0026lt;0.05 considered statistically significant.\u003c/p\u003e"},{"header":"3. RESULTS","content":"\u003cp\u003eThe baseline characteristics of the study population, including demographic data, hemogram parameters, and relevant biochemical variables, are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, stratified by training, test, and external test sets.\u003c/p\u003e\u003cp\u003eIn comparisons across groups, no significant differences were observed in sex and age distributions (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05). Similarly, the majority of hematological parameters did not differ significantly between the datasets. However, statistically significant differences were identified in albumin (p\u0026thinsp;=\u0026thinsp;0.017) and ferritin (p\u0026thinsp;=\u0026thinsp;0.0058) levels, which were attributable to variations in the test set. In addition, the outcome variable Emergency30 differed significantly across the training, test, and external test sets (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), with this difference also driven by the test set.\u003c/p\u003e\u003cp\u003eIn the training dataset, the mean age of patients was 52.93\u0026thinsp;\u0026plusmn;\u0026thinsp;19.46 years, with males having a mean age of 54.06\u0026thinsp;\u0026plusmn;\u0026thinsp;19.56 years and females 52.14\u0026thinsp;\u0026plusmn;\u0026thinsp;19.36 years. The sex distribution consisted of 58.97% females and 41.03% males. The test and external test sets demonstrated similar age and sex distributions; however, the rate of emergency department revisits was higher in the test set compared with the other two sets. Potential linear associations among variables included in the modeling were assessed using Pearson correlation, while monotonic associations were evaluated using Spearman correlation. The results and visualizations of these analyses are provided in the Supplementary Material (Figure SM1).\u003c/p\u003e\u003cp\u003eFollowing these baseline comparisons, a multivariable logistic regression model was developed to predict short-term emergency department revisits, which constituted the primary objective of the study. The model incorporated hematologic (E1) and inflammatory (E2) scores, defined along two biologically meaningful axes, together with age, sex, creatinine, ALT, and AST. In defining the axes, coefficients for variables known to be inversely associated with risk\u0026mdash;such as hemoglobin (HGB), albumin, and lymphocytes (LYM)\u0026mdash;were assigned negative signs, whereas parameters associated with increased risk (e.g., NEU, RDW, CRP) were assigned positive signs. In this way, a biologically grounded and interpretable logistic regression framework was constructed (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA).\u003c/p\u003e\u003cp\u003eVitamin B12 and folate levels showed strong correlations with MCV and RDW, while iron exhibited a moderate correlation with ferritin, as illustrated in Supplementary Material Figure SM1. Inclusion or exclusion of these parameters did not result in meaningful differences in the primary performance metrics (F1 score, MCC, and AUC-ROC) (ΔF1 score\u0026thinsp;\u0026plusmn;\u0026thinsp;0.008, p\u0026thinsp;\u0026gt;\u0026thinsp;0.05). Therefore, to enhance model parsimony and simplify calculation in clinical practice, iron, vitamin B12, and folate were not included in the final formula.\u003c/p\u003e\u003cp\u003eWhen the predefined threshold of 0.219 was applied, performance metrics were consistent across the test and external test sets. In sensitivity analyses conducted within \u0026plusmn;\u0026thinsp;20% of this threshold, changes in the key metrics (sensitivity, specificity, F1 score, MCC) remained negligible. Moreover, 95% confidence intervals obtained through bootstrapping (B\u0026thinsp;=\u0026thinsp;1000) indicated that these differences were not statistically significant.\u003c/p\u003e\u003cp\u003eThe regression formula, coefficients, and odds ratios of the final model are summarized in Tables\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA and \u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB. Both the hematologic axis (E1) and the inflammatory axis (E2) were identified as significant independent predictors of 30-day emergency department revisit (E1: OR 2.06, 95% CI 2.00\u0026ndash;2.13; E2: OR 2.31, 95% CI 2.17\u0026ndash;2.46; both p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Age and creatinine also emerged as strong predictors (Age: OR 4.23, 95% CI 4.07\u0026ndash;4.40; Creatinine: OR 2.89, 95% CI 2.71\u0026ndash;3.08), whereas an increase in AST levels was unexpectedly associated with the highest risk (OR 17.20, 95% CI 14.68\u0026ndash;20.15; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Female sex, on the other hand, was found to be inversely associated with risk and demonstrated a protective effect compared with males (OR 0.35, 95% CI 0.33\u0026ndash;0.37; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e\u003cp\u003eThe logistic regression model demonstrated high classification performance in both the internal test set and the external validation set. In the test set, sensitivity was 85.2% (95% CI 84.2\u0026ndash;86.1), specificity 84.9% (95% CI 84.3\u0026ndash;85.5), and accuracy 84.9% (95% CI 84.5\u0026ndash;85.5). Similarly, in the external test set, high sensitivity (87.9%, 95% CI 87.0\u0026ndash;88.8) and specificity (84.2%, 95% CI 83.6\u0026ndash;84.8) were obtained. The positive predictive value was 70.9% in the test set and 67.5% in the external set, whereas the negative predictive value was 93.0% and 94.9%, respectively (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eOverall performance metrics of the model also demonstrated balanced success. The F1 score (0.77 in the test set; 0.76 in the external set) and Matthews correlation coefficient (0.67 in both sets) supported the consistency of classification between positive and negative classes. The diagnostic odds ratio (DOR) was 32.3 (95% CI 29.6\u0026ndash;35.3) in the test set and 38.8 (95% CI 35.3\u0026ndash;42.8) in the external set, indicating strong discriminatory power. Relative risk (RR) was 10.1 and 13.3, respectively, demonstrating that the probability of emergency department revisit was markedly higher in the high-risk group compared with the low-risk group (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe area under the receiver operating characteristic curve (AUC-ROC), reflecting discriminative ability, was 0.921 (95% CI 0.916\u0026ndash;0.924) in the test set and 0.935 (95% CI 0.931\u0026ndash;0.938) in the external set, demonstrating strong model discrimination. Similarly, the area under the precision\u0026ndash;recall curve (AUC-PR) exceeded 0.82 in both sets (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eCalibration analysis (Calibration-in-the-Large, CITL) demonstrated good agreement of the model in both the test and external test sets. In the test set, the intercept was 0.28 (95% CI 0.23\u0026ndash;0.33) and the calibration slope was 0.92 (95% CI 0.89\u0026ndash;0.94), with a CITL value of 0.34 (95% CI 0.29\u0026ndash;0.38). In the external set, the intercept was 0.06 (95% CI 0.01\u0026ndash;0.12), the calibration slope was 1.01 (95% CI 0.98\u0026ndash;1.04), and the CITL was 0.06 (95% CI 0.01\u0026ndash;0.11). These values indicate that the model was well calibrated in both internal and external validation (in Supplementary Material, Figure SM2 and Table SM2).\u003c/p\u003e\u003cp\u003eTo assess the clinical utility of the model, decision curve analysis (DCA) was performed. DCA illustrates the net benefit of the model across different probability thresholds in comparison with two extreme strategies: treating all patients as high risk (\u0026ldquo;treat-all\u0026rdquo;) and treating no patients as high risk (\u0026ldquo;treat-none\u0026rdquo;). This analysis highlights not only the statistical discriminative ability but also the contribution of the model to clinical decision-making.\u003c/p\u003e\u003cp\u003eIn our study, the model provided greater net benefit than either of the two extreme strategies across a wide range of probability thresholds in both the test and external test sets (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The added benefit was particularly evident within the 10\u0026ndash;40% threshold range, suggesting that the model can be reliably applied in clinical practice for patients within the low-to-intermediate risk spectrum. The similar findings observed in the external validation set further support the generalizability and clinical validity of the model.\u003c/p\u003e\u003cp\u003eWith respect to calibration, the Brier score was 0.10 in the test set and 0.09 in the external set, with these low values indicating good model calibration. In addition, pseudo-R\u0026sup2; values\u0026mdash;including McFadden (0.46\u0026ndash;0.51), Cox\u0026ndash;Snell (0.43\u0026ndash;0.45), and Nagelkerke (0.61\u0026ndash;0.65)\u0026mdash;suggested a high explanatory power of the model. Furthermore, likelihood ratios (PLR\u0026thinsp;\u0026asymp;\u0026thinsp;5.6; NLR\u0026thinsp;\u0026asymp;\u0026thinsp;0.14\u0026ndash;0.17) supported the strong diagnostic performance of the model (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eIn addition, subgroup analyses were conducted by sex and age tertiles to assess the consistency of the model. While the overall performance metrics of the model were preserved across subgroups, differences in sensitivity and specificity values were observed between groups (in Supplementary Material, Table SM3).\u003c/p\u003e\u003cp\u003eAn exploratory, observational evaluation was performed for secondary clinical outcomes. No additional modeling was conducted, and no formal statistical inference was intended. For example, among patients classified as high risk by the model (Pred\u0026thinsp;=\u0026thinsp;1), 30-day readmission (44\u0026ndash;45%), 90-day transfusion (15%), and 180-day mortality (3%) rates were observed. In contrast, in the low-risk group (Pred\u0026thinsp;=\u0026thinsp;0), the rates of these events were markedly lower (\u0026lt;\u0026thinsp;2%). Similarly, among patients who actually revisited the emergency department (ED\u0026thinsp;=\u0026thinsp;1), these event rates also remained high (approximately 49%, 17%, and 3%). These findings suggest that the risk groups identified by the model were aligned with subsequent clinical trajectories (Supplementary Material, Table SM4).\u003c/p\u003e"},{"header":"4. DISCUSSION","content":"\u003cp\u003eThe appropriate observation window for emergency department (ED) reattendance is critical both for quality measurement and for the design of preventive interventions. Although the commonly used 72-hour cutoff is practical, large-population data have shown that revisit patterns follow a logarithmically declining curve, with a secondary peak occurring around day 9; thus, the 72-hour metric captures only a small fraction of the overall picture. In contrast, a 30-day follow-up period encompasses both early and relatively late revisits, providing a more realistic quality indicator, and this approach was therefore adopted in our study [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn the literature, 30-day emergency department reattendance rates are generally reported to range between 15% and 22% [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. In our cohort, however, this rate was notably higher at approximately 30%. Such a discrepancy may be attributable to the patient profile, comorbidity burden, and regional healthcare utilization patterns characteristic of the tertiary academic hospital in which our study was conducted. Furthermore, as our analysis included only revisits following outpatient clinic encounters, this selection may have contributed to the elevated rate. The high reattendance rate further emphasizes the clinical importance of our model, underscoring the value of early risk stratification for healthcare planning and cost-effectiveness.\u003c/p\u003e\u003cp\u003eThe use of a parsimonious, laboratory-based model provides a strong rationale in terms of both accessibility and interpretability: routine blood tests (e.g., albumin, CRP, creatinine, hemogram parameters) can distinguish short-term mortality and adverse outcomes with high accuracy, and in some studies have even outperformed formal triage algorithms [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Our strategy integrated these biomarkers into biologically defined axes (hematologic\u0026ndash;inflammatory) to enable risk stratification that can be easily explained in clinical practice. Moreover, the use of a threshold predetermined on the internal test dataset and subsequently applied to the test and external validation sets without recalibration is consistent with good reporting practice [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn our study, vitamin B12 and folate tests, which were initially included, did not significantly alter model performance. Deficiencies in B12 and folate are known to have characteristic hematologic manifestations. Specifically, mean corpuscular volume (MCV) increases in the presence of B12 or folate deficiency due to the occurrence of macrocytosis. Red cell distribution width (RDW) increases in these deficiencies as well, reflecting anisocytosis caused by the coexistence of large macrocytes and normocytes. The combination of elevated RDW and MCV is therefore suggestive of B12 or folate deficiency. Accordingly, rather than being directly included in the model, B12 and folate deficiencies may have been adequately represented through their hematologic surrogates, namely macrocytosis and anisocytosis (MCV and RDW) [\u003cspan additionalcitationids=\"CR17\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Iron testing, in contrast, is well known to exhibit diurnal variation and is strongly influenced by dietary intake. In healthy individuals, serum ferritin levels show a strong correlation with iron stores [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. In our study, however, only a weak-to-moderate correlation was observed between iron and ferritin. Given that measured serum iron levels were less stable and that ferritin, as a more reliable marker, correlated with iron while also being interpretable in the logistic regression framework, ferritin was retained as the preferred variable for the final model.\u003c/p\u003e\u003cp\u003eIn our cohort, female sex was associated with a lower risk of reattendance, consistent with findings reported in large-scale datasets [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. This difference may be influenced by a complex interplay of clinical and sociodemographic factors, including reasons for presentation, discharge planning, healthcare-seeking behaviors, and comorbidity profiles. While causal interpretations should be avoided, it should be cautiously noted that female sex may act as a protective factor at least in the general population [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eFrom a biological perspective, albumin\u0026mdash;a widely available and rapidly measured parameter\u0026mdash;has been shown to serve as an independent predictor of 30-day mortality in the emergency department and to provide additional clinical value particularly in patients within the low-to-intermediate Sequential/Sepsis-related Organ Failure Assessment (SOFA) ranges [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Similarly, the neutrophil-to-lymphocyte ratio (NLR) and other hemogram-derived indices have been associated with short-term adverse outcomes across different acute patient populations [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Taken together, this body of evidence suggests that our axis-based approach represents more than a purely statistical construct and in fact reflects meaningful pathophysiological mechanisms.\u003c/p\u003e\u003cp\u003eIn the literature, methodological performance values should always be interpreted within the clinical context of the endpoint under study. For example, models predicting 30-day mortality among discharged patients using rich data sources (clinical plus textual information) have achieved AUCs in the range of ~\u0026thinsp;0.94\u0026ndash;0.97 [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. In contrast, predicting reattendance is inherently more challenging due to its strong behavioral and organizational components; institution-specific models for 30-day postoperative ED-related readmissions have reported AUCs of ~\u0026thinsp;0.85\u0026ndash;0.89, while triage-based hospital admission predictions typically remain at ~\u0026thinsp;0.82\u0026ndash;0.83 [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Within this reference framework, the discriminative performance of our laboratory-only model\u0026mdash;operating with a prespecified threshold\u0026mdash;together with its consistency in F1/MCC metrics (0.77/0.66), suggests a low-friction and scalable solution for clinical practice. Moreover, despite addressing a more difficult endpoint, our method achieved an AUC-ROC of 0.92/0.94 in the test and external validation sets, respectively, which represents a remarkably high level of performance.\u003c/p\u003e\u003cp\u003eIn our study, the markedly strong association of AST levels with the risk of reattendance (OR\u0026thinsp;\u0026asymp;\u0026thinsp;17) was a striking and somewhat unexpected finding that warrants explanation. Although AST is primarily a liver-derived enzyme, it is a non-specific biomarker that can be elevated in various pathophysiological conditions, including muscle injury, hemolysis, drug exposure, or metabolic stress. Thus, elevated AST levels may in fact reflect an underlying comorbidity burden or systemic tissue injury. Nevertheless, the magnitude of the observed association cannot be fully accounted for by biological plausibility alone and may have been influenced by measurement-related variation or cohort-specific characteristics. Therefore, this unexpected and pronounced association between AST and reattendance risk should be retested in independent validation cohorts and further investigated with respect to underlying biological mechanisms.\u003c/p\u003e\u003cp\u003eInterpretation of calibration findings is critical for implementation. Even when CITL is positively shifted in the test set, local prevalence or logit differences can be practically addressed through intercept recalibration, provided that the slope remains close to 1; this represents the TRIPOD-compliant first step toward field adaptation. In such cases, rather than focusing on the absolute interpretation of predicted probabilities, calibration can be strengthened through threshold-based decision-making (classification) combined with local intercept adjustment [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. This approach provides a practical starting point for future studies aiming to further refine model implementation.\u003c/p\u003e\u003cp\u003eDecision curve analysis (DCA) demonstrates the net benefit of a model compared with \u0026ldquo;treat-all\u0026rdquo; or \u0026ldquo;treat-none\u0026rdquo; strategies across clinically meaningful probability thresholds. In line with the foundational methodology of DCA and contemporary interpretation guidelines, the positive net benefit observed particularly within the 10\u0026ndash;40% range provides rational evidence for interventions that may be effective at low-to-intermediate risk thresholds, such as early post-discharge telephone follow-up (within 48\u0026ndash;72 hours), home care coordination, expedited appointment scheduling (\u0026le;\u0026thinsp;7 days), and medication adherence counseling. Such strategies may not only reduce adverse events that could otherwise progress to life-threatening conditions before ED reattendance but also improve cost-effectiveness from a healthcare expenditure perspective [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eFrom a public health perspective, overcrowding and demand volatility in the ED pose major challenges for resource management. Systematic reviews and real-time prediction models developed for revisit forecasting and patient flow management have demonstrated that hourly-to-daily uncertainty can be reduced [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. When combined with individual-level risk stratification, these approaches enable proactive targeting of resources. Our model may facilitate focused monitoring and referral of high-risk subgroups, thereby supporting earlier interventions for potentially preventable revisits [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eAlthough the primary objective of this study was to predict 30-day emergency department reattendance, exploratory and descriptive analyses revealed that patients classified as high risk by the model also exhibited markedly higher rates of hospitalization, transfusion, and mortality. These findings suggest that the model may reflect not only reattendance but also subsequent clinical trajectories. However, these results are hypothesis-generating in nature and fall outside the main scope of our study. Future logistic regression or multivariable modeling efforts incorporating these secondary outcomes may further enhance the clinical utility of the model.\u003c/p\u003e\u003cp\u003e\u003cb\u003eLimitations\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThis study was based on routine laboratory results obtained on the day of outpatient presentation in adults presenting outside the emergency department. Therefore, the findings cannot be directly generalized to patients presenting initially to the ED or those requiring hospitalization. In the primary analyses, a complete-case approach was applied, meaning that cases with missing values for any variable required by the model were excluded. As laboratory testing in the outpatient setting is often contingent on the clinical condition, missingness may not have been completely random, raising the possibility of selection bias. Nonetheless, our decision not to apply imputation provided practical simplicity for model use and interpretation; sensitivity analyses in future studies targeting broader patient groups may better assess this risk\u003c/p\u003e\u003cp\u003eThe \u0026ldquo;external test\u0026rdquo; cohort was a held-out sample from the same institution; thus, performance may vary across different laboratory platforms, patient populations, and healthcare delivery settings. Accordingly, local intercept calibration and threshold optimization should be performed before implementation. Ideally, multicenter, prospective validation and impact studies are needed to assess generalizability and real-world utility. Furthermore, as the model relies solely on laboratory data, it does not incorporate clinical context such as symptoms, vital signs, or reasons for presentation, which may contribute to loss of specificity in certain cases.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eUsing an interpretable model based on routine laboratory measurements, the risk of 30-day emergency department reattendance can be stratified with clinically meaningful accuracy and a net benefit profile confirmed by decision curve analysis. With reinforcement of calibration through local intercept adjustment and threshold selection informed by stakeholder input, the model holds potential to evolve into a practical decision-support tool for proactive post-discharge monitoring and resource planning.\u003c/p\u003e\u003cp\u003eOne notable aspect of our study is that the model not only predicted reattendance but also identified risk groups that were meaningfully associated with clinical outcomes such as mortality, hospitalization, and transfusion. This finding highlights that biological stress axes reflect not only healthcare utilization but also patient prognosis.\u003c/p\u003e\u003cp\u003eIn light of these results, the AXIS-2 model could be piloted in specific high-risk populations, such as elderly patients with a high comorbidity burden, individuals with frequent outpatient visits, and patients with chronic inflammatory or hematologic conditions. Early risk stratification in these groups may facilitate more targeted implementation of post-discharge follow-up programs (e.g., telephone monitoring, home care coordination, or expedited appointments). In this way, the model has the potential not only to enhance patient safety at the individual level but also to reduce ED overcrowding and healthcare costs at the system level.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAppendix / Supplementary Material\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAdditional tables and figures supporting the findings of this study are provided in the Supplementary Material (Table SM1-SM4, Figure SM1–SM2). These materials include reference ranges, correlation analyses, calibration plots, and subgroup results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was conducted in accordance with the Declaration of Helsinki and approved by the Non-Interventional Clinical Research Ethics Committee of the İzmir Tepecik Training and Research Hospital, University of Health Sciences Türkiye (Approval No: 2025/08-19, dated 11 September 2025). The requirement for informed consent was waived by the ethics committee because the study used retrospective, anonymized laboratory data with no identifiable personal information.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable. No individual person’s data in any form (including images, videos, or identifiable information) are included in this manuscript\u003cstrong\u003e.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and analyzed during the current study are not publicly available due to institutional data protection regulations but are available from the corresponding author (FD) upon reasonable request. All key summary tables and model coefficients are included in this published article and its supplementary information files.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors’ contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFD (Ferhat Demirci) obtained ethics approval, conceptualized the study, curated and analyzed the data, developed the methodology, performed the statistical analysis, and drafted the original manuscript. AD (Aylin Demirci) contributed to data collection and participated in the review and editing of the manuscript. Both authors read and approved the final version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors express their sincere gratitude to Prof. Dr. Savaş Yakan, Chief Physician of İzmir Tepecik Training and Research Hospital, and Prof. Dr. Ayfer Çolak, Head of the Department of Medical Biochemistry, for their valuable support and encouragement throughout this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFootnotes\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eRising KL, Victor TW, Hollander JE, Carr BG. Patient Returns to the Emergency Department: The Time-to-return Curve. Academic Emergency Medicine. 2014 Aug;21(8):864\u0026ndash;71. \u003c/li\u003e\n\u003cli\u003eFrąckiewicz J, Włodarek D, Brzozowska A, Wierzbicka E, Słowińska MA, Wądołowska L, et al. Hematological parameters and all-cause mortality: a prospective study of older people. Aging Clin Exp Res. 2018 May 29;30(5):517\u0026ndash;26. \u003c/li\u003e\n\u003cli\u003eKim S, Lee K, Kim I, Jung S, Kim MJ. Red cell distribution width and early mortality in elderly patients with severe sepsis and septic shock. Clin Exp Emerg Med. 2015 Sep;2(3):155\u0026ndash;61. \u003c/li\u003e\n\u003cli\u003eLiu X, Shen Y, Wang H, Ge Q, Fei A, Pan S. Prognostic Significance of Neutrophil-to-Lymphocyte Ratio in Patients with Sepsis: A Prospective Observational Study. Mediators Inflamm. 2016;2016:8191254. \u003c/li\u003e\n\u003cli\u003eLiu Y, Gao Y, Liang B, Liang Z. The prognostic value of C-reactive protein to albumin ratio in patients with sepsis: a systematic review and meta-analysis. Aging Male. 2023 Dec;26(1):2261540. \u003c/li\u003e\n\u003cli\u003eBrossard C, Goetz C, Catoire P, Cipolat L, Guyeux C, Gil Jardine C, et al. Predicting emergency department admissions using a machine-learning algorithm: a proof of concept with retrospective study. BMC Emerg Med. 2025 Jan 6;25(1):3. \u003c/li\u003e\n\u003cli\u003eRaita Y, Goto T, Faridi MK, Brown DFM, Camargo CA, Hasegawa K. Emergency department triage prediction of clinical outcomes using machine learning models. Crit Care. 2019 Feb 22;23(1):64. \u003c/li\u003e\n\u003cli\u003eCollins GS, Reitsma JB, Altman DG, Moons K. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD Statement. BMC Med. 2015;13(1):1. \u003c/li\u003e\n\u003cli\u003eWolff RF, Moons KGM, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies. Ann Intern Med. 2019 Jan 1;170(1):51\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eVickers AJ, Elkin EB. Decision Curve Analysis: A Novel Method for Evaluating Prediction Models. Medical Decision Making. 2006 Nov 1;26(6):565\u0026ndash;74. \u003c/li\u003e\n\u003cli\u003eVan Calster B, McLernon DJ, van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. BMC Med. 2019 Dec 16;17(1):230. \u003c/li\u003e\n\u003cli\u003eKim H, Han SJ, Lee JH, Lim J, Moon S do, Moon H, et al. A Descriptive Study of Emergency Department Visits Within 30 Days of Discharge. Ann Geriatr Med Res. 2021 Dec 31;25(4):245\u0026ndash;51. \u003c/li\u003e\n\u003cli\u003eBrown JF, Fu J. Emergency Department Avoidance by Transgender Persons: Another Broken Thread in the \u0026ldquo;Safety Net\u0026rdquo; of Emergency Medicine Care. Ann Emerg Med. 2014 Jun;63(6):721\u0026ndash;2. \u003c/li\u003e\n\u003cli\u003eKristensen M, Iversen AKS, Gerds TA, \u0026Oslash;stervig R, Linnet JD, Barfod C, et al. Routine blood tests are associated with short term mortality and can improve emergency department triage: a cohort study of \u0026amp;gt;12,000 patients. Scand J Trauma Resusc Emerg Med. 2017 Dec 28;25(1):115. \u003c/li\u003e\n\u003cli\u003eVickers AJ, van Calster B, Steyerberg EW. A simple, step-by-step guide to interpreting decision curve analysis. Diagn Progn Res. 2019 Dec 4;3(1):18. \u003c/li\u003e\n\u003cli\u003eManer BS, Killeen RB, Moosavi L. Mean Corpuscular Volume. 2025. \u003c/li\u003e\n\u003cli\u003eEpstein-Peterson ZD, Chokshi I, Barrow B, Lobaugh S, Devlin S, Fenelus M, et al. Laboratory evaluation of folate deficiency among inpatients with cancer. Int J Lab Hematol. 2021 Aug;43(4):O164\u0026ndash;7. \u003c/li\u003e\n\u003cli\u003eRajashekar RB, Patel S, Kulkarrni P. Discriminant Functions in the Diagnosis of Vitamin B12 Deficiency Anemia, the Value of RDW-SD: An Analytical Study. National Journal of Laboratory Medicine. 2017;6(1):PO01\u0026ndash;6. \u003c/li\u003e\n\u003cli\u003eRohr M, Brandenburg V, Brunner-La Rocca HP. How to diagnose iron deficiency in chronic disease: A review of current methods and potential marker for the outcome. Eur J Med Res. 2023 Jan 9;28(1):15. \u003c/li\u003e\n\u003cli\u003eTsai IT, Sun CK, Chang CS, Lee KH, Liang CY, Hsu CW. Characteristics and outcomes of patients with emergency department revisits within 72 hours and subsequent admission to the intensive care unit. Tzu Chi Med J. 2016;28(4):151\u0026ndash;6. \u003c/li\u003e\n\u003cli\u003eWang HC, Fang CC, Huang CH, Gao JW, Chen JH, Tsai CL. Factors associated with overall and high-risk return visits to the emergency department: a vital sign trajectory approach. BMC Emerg Med. 2025 Apr 12;25(1):57. \u003c/li\u003e\n\u003cli\u003eTurcato G, Zaboli A, Sibilio S, Fanni Canelles MF, Rella E, Giudiceandrea A, et al. Prognostic Role of Serum Albumin in Predicting 30-Day Mortality in Patients with Infections in Emergency Department: A Prospective Study. J Clin Med. 2023 May 13;12(10):3447. \u003c/li\u003e\n\u003cli\u003eTurcato G, Zaboli A, Sibilio S, Mian M, Brigo F. The Clinical Utility of Albumin with Sequential Organ Failure Assessment (SOFA) in Improving 30-Day Mortality Prediction in Patients with Infection in the Emergency Department. J Clin Med. 2023 Dec 14;12(24):7676. \u003c/li\u003e\n\u003cli\u003eSoulaiman SE, Dopa D, Raad ABT, Hasan W, Ibrahim N, Hasan AY, et al. Cohort retrospective study: the neutrophil to lymphocyte ratio as an independent predictor of outcomes at the presentation of the multi-trauma patient. Int J Emerg Med. 2020 Dec 4;13(1):5. \u003c/li\u003e\n\u003cli\u003eGalardo G, Crisanti L, Gentile A, Cornacchia M, Iatomasi F, Egiddi I, et al. Neutrophil to lymphocyte ratio (NLR) and short-term mortality risk in elderly acute medical patients admitted to a University Hospital Emergency Department. Intern Emerg Med. 2025 Mar;20(2):553\u0026ndash;62. \u003c/li\u003e\n\u003cli\u003eBarash Y, Soffer S, Grossman E, Tau N, Sorin V, BenDavid E, et al. Alerting on mortality among patients discharged from the emergency department: a machine learning model. Postgrad Med J. 2022 Mar 1;98(1157):166\u0026ndash;71. \u003c/li\u003e\n\u003cli\u003eParker CA, Liu N, Wu SX, Shen Y, Lam SSW, Ong MEH. Predicting hospital admission at the emergency department triage: A novel prediction model. Am J Emerg Med. 2019 Aug;37(8):1498\u0026ndash;504. \u003c/li\u003e\n\u003cli\u003eMi\u0026scaron;ić V V., Gabel E, Hofer I, Rajaram K, Mahajan A. Machine Learning Prediction of Postoperative Emergency Department Hospital Readmission. Anesthesiology. 2020 May;132(5):968\u0026ndash;80. \u003c/li\u003e\n\u003cli\u003eVan Calster B, McLernon DJ, van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. BMC Med. 2019 Dec 16;17(1):230. \u003c/li\u003e\n\u003cli\u003eLukanski A, Watters S, Bilderback AL, Buchanan D, Hodges JC, Burwell D, et al. Implementing a Discharge Follow-up Phone Call Program Reduces Readmission Rates in an Integrated Health System. Journal for Healthcare Quality. 2023 Nov;45(6):315\u0026ndash;23. \u003c/li\u003e\n\u003cli\u003eBilicki DJ, Reeves MJ. Outpatient Follow-Up Visits to Reduce 30-Day All-Cause Readmissions for Heart Failure, COPD, Myocardial Infarction, and Stroke: A Systematic Review and Meta-Analysis. Prev Chronic Dis. 2024 Sep 26;21:240138. \u003c/li\u003e\n\u003cli\u003eAsheim A, Bache-Wiig Bj\u0026oslash;rnsen LP, N\u0026aelig;ss-Pleym LE, Uleberg O, Dale J, Nilsen SM. Real-time forecasting of emergency department arrivals using prehospital data. BMC Emerg Med. 2019 Dec 5;19(1):42. \u003c/li\u003e\n\u003cli\u003eSilva E, Pereira MF, Vieira JT, Ferreira‐Coimbra J, Henriques M, Rodrigues NF. Predicting hospital emergency department visits accurately: A systematic review. Int J Health Plann Manage. 2023 Jul 10;38(4):904\u0026ndash;17. \u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1. Descriptive statistics of the datasets\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eFeatures\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eUnit\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003eTrain Test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003eTest Set\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003eExternal Test Set\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003eP value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003eGender\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 58px;\"\u003e\n \u003cp\u003eAll\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003en\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e55,431 (%100)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e18,477 (%100)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e18,478 (%100)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 58px;\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003en\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e22,742 (%41.03)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e7,562 (%40.93)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e7,605 (%41.16)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 58px;\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003en\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e32,689 (%58.97)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e10,915 (%59.07)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e10,873 (%58.84)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 58px;\"\u003e\n \u003cp\u003eAll\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eyears\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e52.93\u0026plusmn;19.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e53.32\u0026plusmn;19.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e52.90\u0026plusmn;19.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 58px;\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eyears\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e54.06\u0026plusmn;19.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e54.40\u0026plusmn;19.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e53.99\u0026plusmn;19.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 58px;\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eyears\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e52.14\u0026plusmn;19.36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e52.57\u0026plusmn;19.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e52.14\u0026plusmn;19.36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eHemoglobin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eg/dL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e13.00\u0026plusmn;1.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e12.99\u0026plusmn;2.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e13.00\u0026plusmn;1.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eWhite blood cell\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e10\u0026sup3;/\u0026micro;L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e8.22\u0026plusmn;9.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e8.17\u0026plusmn;8.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e8.21\u0026plusmn;9.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eNeutrophil\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e10\u0026sup3;/\u0026micro;L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e4.79\u0026plusmn;3.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e4.78\u0026plusmn;3.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e4.76\u0026plusmn;3.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eLymphocyte\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e10\u0026sup3;/\u0026micro;L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e2.66\u0026plusmn;8.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e2.61\u0026plusmn;7.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e2.66\u0026plusmn;8.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003ePlatelet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e10\u0026sup3;/\u0026micro;L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e266.86\u0026plusmn;109.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e266.82\u0026plusmn;110.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e266.08\u0026plusmn;108.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eNeutrophil/ Lymphocyte Ratio\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e2.55 \u0026plusmn; 2.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e2.54 \u0026plusmn; 2.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e2.55\u0026plusmn;2.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003ePlatelet/Lymphocyte Ratio\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e140.87\u0026plusmn;93.27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e141.05\u0026plusmn;94.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e140.43\u0026plusmn;91.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eMean Corpuscular Volume\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003efL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e86.14\u0026plusmn;7.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e86.08\u0026plusmn;7.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e86.09\u0026plusmn;7.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eRed Distribution Width\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e14.79\u0026plusmn;2.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e14.82\u0026plusmn;2.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e14.78\u0026plusmn;2.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eC-Reactive Protein\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003emg/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e10.61\u0026plusmn;24.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e11.12\u0026plusmn;27.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e10.81\u0026plusmn;26.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eAlbumin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eg/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e42.62\u0026plusmn;14.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e42.36\u0026plusmn;6.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e42.41\u0026plusmn;9.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.017 - Test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eFerritin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eng/mL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e107.58\u0026plusmn;277.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e114.06\u0026plusmn;345.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e105.04\u0026plusmn;239.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e0.0058 - test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eIron\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026micro;g/dL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e173.90\u0026plusmn;149.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e172.06\u0026plusmn;148.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e174.39\u0026plusmn;157.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eB12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003epg/mL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e296.90\u0026plusmn;244.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e297.69\u0026plusmn;242.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e298.97\u0026plusmn;245.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eFolate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eng/mL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e8.68\u0026plusmn;4.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e8.65\u0026plusmn;4.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e8.65\u0026plusmn;4.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eCreatinine\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003emg/dL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e1.01\u0026plusmn;0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e1.02\u0026plusmn;0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e1.02\u0026plusmn;0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eAspartate Aminotransferase\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eU/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e25.45\u0026plusmn;55.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e25.61\u0026plusmn;52.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e26.88\u0026plusmn;250.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eAlanine Aminotransferase\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eU/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e24.96\u0026plusmn;52.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e25.39\u0026plusmn;63.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e24.74\u0026plusmn;51.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003eEMERGENCY30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 58px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e40,594 (%73.23)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e12,899 (%69.81)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e13,459 (%72.84)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e* Test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 58px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 148px;\"\u003e\n \u003cp\u003e14,837 (%26.77)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 118px;\"\u003e\n \u003cp\u003e5,578 (%30.19)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e5,019 (%27.16)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62px;\"\u003e\n \u003cp\u003e* Test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e*For groups 0 and 1: in the Train vs Test comparison, p = 2.25\u0026times;10⁻\u0026sup1;⁹; in Train vs External, p = 0.298; and in Test vs External, p = 1.35\u0026times;10⁻\u0026sup1;⁰.\u003c/p\u003e\n\u003cp\u003eTable 2. Final Logistic Regression Model, Biological Axis Components, and Multivariable Results\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" valign=\"top\" style=\"width: 604px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eA. Final Logistic Regression Model and Biological Axis Components\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003eRegression Formula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 491px;\"\u003e\n \u003cp\u003e\u003cstrong\u003elogit(p) = \u0026beta;₀ + \u0026beta;₁E1_score + \u0026beta;₂E2_score + \u0026beta;₃Age + \u0026beta;₄Creatinine + \u0026beta;₅ALT + \u0026beta;₆AST + \u0026Sigma;\u0026beta;g\u0026middot;1(Gender category)\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003ep = 1 / (1 + e\u003csup\u003e\u0026minus;logit(p)\u003c/sup\u003e)\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eE1 (Hematology)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" valign=\"bottom\" style=\"width: 491px;\"\u003e\n \u003cp\u003ez(HGB)^\u0026minus;, z(LYM)^\u0026minus;, z(MCV)^+, z(NEU)^+, z(PLT)^+, z(RDW)^+, z(WBC)^+ \u0026nbsp;\u0026rarr; mean\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003eE2 (Inflammation)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" valign=\"bottom\" style=\"width: 491px;\"\u003e\n \u003cp\u003ez(CRP)^+, z(Ferritin)^+, z(Albumin)^\u0026minus;, z(NLR)^+, z(PLR)^+ \u0026nbsp;\u0026rarr; mean\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" style=\"width: 604px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eB. \u0026nbsp; Multivariable Logistic Regression Results\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFeature\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eB (beta)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOR (95% CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eIntercept\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e-0.959\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e0.38 (0.37 \u0026ndash; 0.40)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eE1 score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e0.724\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e2.06 (2.00 \u0026ndash; 2.13)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eE2 score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e0.838\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e2.31 (2.17 \u0026ndash; 2.46)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e1.442\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e4.23 (4.07 \u0026ndash; 4.40)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eCreatinine\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e1.061\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e2.89 (2.71 \u0026ndash; 3.08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eALT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e0.361\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e1.44 (1.32 \u0026ndash; 1.56)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eAST\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e2.845\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e17.20 (14.68 \u0026ndash; 20.15)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eFemale (ref=male)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e-1.053\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e0.35 (0.33 \u0026ndash; 0.37)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 38px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 151px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eHGB: Hemoglobin; LYM: Lymphocyte count; MCV: Mean corpuscular volume; NEU: Neutrophil count; PLT: Platelet count; RDW: Red cell distribution width; WBC: White blood cell count; CRP: C-reactive protein; NLR: Neutrophil-to-lymphocyte ratio; PLR: Platelet-to-lymphocyte ratio.\u003c/p\u003e\n\u003cp\u003eWhen calculating Z-scores, the (+) sign indicates that an increase in the relevant parameter is effective in increasing the risk of emergency room visits, while the (-) sign indicates that the increase is effective in reducing the risk. The superscript (^) sign indicates that the variable is used in the model with this risk aspect.\u003c/p\u003e\n\u003cp\u003eContinuous variables are standardized (z-score); ORs are for a 1 SD increase.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Table 3. Logistic regression performance results\u003c/p\u003e\n\u003cdiv align=\"center\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003eTest Set\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eExternal Test Set\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eTP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e4752\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e4414\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eFP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e1948\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e2128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eFN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e826\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e605\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eTN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e10951\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e11331\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,852\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.842, 0.861]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,879\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.870, 0.888]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,849\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.843, 0.855]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,842\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.836, 0.848]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003ePPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,709\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.699, 0.720]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,675\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.665, 0.685]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,930\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.925, 0.934]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,949\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.945, 0.953]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003ePLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e5,641\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[5.414, 5.866]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e5,562\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[5.370, 5.780]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eNLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,174\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.164, 0.186]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,143\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.133, 0.154]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,850\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.845, 0.855]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,852\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.847, 0.857]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eF1 Score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,774\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.766, 0.782]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,764\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.756, 0.772]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eMCC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,669\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.658, 0.680]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,671\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.661, 0.681]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eAUC-ROC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,921\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.916, 0.924]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,935\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.931, 0.938]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eAUC-PR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,817\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.805, 0.828]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,835\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.824, 0.846]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eDiagnostic Odds Ratio\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e32,342\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[29.611, 35.324]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e38,848\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[35.265, 42.796]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eRelative Risk\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e10,112\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[9.452, 10.819]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e13,311\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[12.295, 14.412]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 151px;\"\u003e\n \u003cp\u003eBrier Score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,102\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0,099, 0,105]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,087\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0,085, 0,089]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 151px;\"\u003e\n \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e_McFadden\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,461\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.447, 0.475]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,512\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.500, 0.524]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 151px;\"\u003e\n \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e_CoxSnell\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,431\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.421, 0.441]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,451\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.442, 0.460]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 151px;\"\u003e\n \u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e_Nagelkerke\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0,611\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e[0.597, 0.624]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0,654\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0.642, 0.665]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eTP: True Positive, FP: False Positive, FN: False Negative, TN: True Negative, PPV: Positive Predictive Value, NPV: Negative Predictive Value, PLR: Positive Likelihood Ratio, NLR: Negative Likelihood Ratio, Accuracy: Overall Correct Classification Rate, F1 Score: Harmonic Mean of Precision and Recall, MCC: Matthews Correlation Coefficient, AUC-ROC: Area Under the Receiver Operating Characteristic Curve, AUC-PR: Area Under the Precision\u0026ndash;Recall Curve.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-emergency-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"emmd","sideBox":"Learn more about [BMC Emergency Medicine](http://bmcemergmed.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/emmd","title":"BMC Emergency Medicine","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7980356/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7980356/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground:\u003c/h2\u003e\u003cp\u003eThirty-day emergency department (ED) revisit is a major quality indicator reflecting morbidity and healthcare burden. Laboratory data obtained during outpatient encounters may capture underlying biological stress axes. This study aimed to develop and externally validate an interpretable logistic regression model\u0026mdash;the AXIS-2 (Anemia\u0026ndash;Inflammation Two-Axis Composite Index)\u0026mdash;for predicting 30-day ED revisits using routine laboratory parameters.\u003c/p\u003e\u003ch2\u003eMethods:\u003c/h2\u003e\u003cp\u003eThis retrospective study included 92,386 adult outpatients who visited a tertiary academic hospital between January 2015 and August 2025. Laboratory variables were biologically grouped into two axes: Axis-1 (hematologic indicators of anemia and erythropoiesis: hemoglobin, lymphocyte count, mean corpuscular volume, neutrophil count, platelet count, red cell distribution width, and white blood cell count) and Axis-2 (inflammatory and catabolic markers: C-reactive protein, ferritin, albumin, neutrophil-to-lymphocyte ratio, and platelet-to-lymphocyte ratio). A multivariable logistic regression model was trained using stratified sampling and isotonic calibration. Model performance was assessed in both internal and external test datasets using discrimination (AUC-ROC, AUC-PR), classification (F1 score, MCC, sensitivity, specificity), calibration, and decision curve analysis.\u003c/p\u003e\u003ch2\u003eResults:\u003c/h2\u003e\u003cp\u003eThe 30-day revisit rate was 30%. Both hematologic (Axis-1) and inflammatory (Axis-2) axes were independent predictors of revisit risk (Axis-1 OR 2.06; Axis-2 OR 2.31; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The model showed excellent discrimination (AUC-ROC 0.921 in the test set; 0.935 in the external set) and balanced classification accuracy (\u0026asymp;\u0026thinsp;85% for both sensitivity and specificity). Negative predictive value reached 95%, and calibration metrics indicated strong concordance between predicted and observed probabilities. Decision curve analysis demonstrated clear net benefit within the 10\u0026ndash;40% probability range. High-risk patients also exhibited higher 30-day readmission (\u0026asymp;\u0026thinsp;45%), transfusion (\u0026asymp;\u0026thinsp;15%), and 180-day mortality (\u0026asymp;\u0026thinsp;3%) rates compared with the low-risk group.\u003c/p\u003e\u003ch2\u003eConclusions:\u003c/h2\u003e\u003cp\u003eAXIS-2 is a transparent, laboratory-based model that accurately predicts 30-day emergency revisits and offers interpretable outputs suitable for integration into clinical workflows. Its strong calibration and decision-analytic benefit support its use as a cost-efficient tool for post-discharge monitoring and healthcare resource optimization.\u003c/p\u003e","manuscriptTitle":"Routine Laboratory Data for Predicting 30-Day Emergency Department Revisits: The AXIS-2 Risk Score","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-17 10:54:28","doi":"10.21203/rs.3.rs-7980356/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-01-20T04:27:08+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-01-14T02:25:58+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-01-10T21:20:46+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"301634378895216464469571336785025573482","date":"2026-01-07T01:45:42+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"329198224518809202581477588935138098782","date":"2026-01-05T11:24:58+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"213926377301900617999621334609072032743","date":"2025-12-22T18:18:37+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-19T11:49:11+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"89412013004089804758162265117566177698","date":"2025-11-29T11:11:59+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"235790818025665161715809047053493168871","date":"2025-11-27T04:38:22+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-11-05T23:28:46+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-11-04T07:31:14+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-11-01T05:36:54+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-11-01T05:36:18+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Emergency Medicine","date":"2025-10-29T13:21:49+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-emergency-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"emmd","sideBox":"Learn more about [BMC Emergency Medicine](http://bmcemergmed.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/emmd","title":"BMC Emergency Medicine","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a7a396ef-97f5-4c75-a0a8-9a06e3553fa0","owner":[],"postedDate":"November 17th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-02-16T16:08:35+00:00","versionOfRecord":{"articleIdentity":"rs-7980356","link":"https://doi.org/10.1186/s12873-026-01496-w","journal":{"identity":"bmc-emergency-medicine","isVorOnly":false,"title":"BMC Emergency Medicine"},"publishedOn":"2026-02-13 15:59:22","publishedOnDateReadable":"February 13th, 2026"},"versionCreatedAt":"2025-11-17 10:54:28","video":"","vorDoi":"10.1186/s12873-026-01496-w","vorDoiUrl":"https://doi.org/10.1186/s12873-026-01496-w","workflowStages":[]},"version":"v1","identity":"rs-7980356","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7980356","identity":"rs-7980356","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}