A 9-Variable Lactate-to-Albumin Ratio-Driven Clinical Risk Score for Early Mortality Prediction in Very Elderly Mechanically Ventilated ICU Patients: Multicenter Derivation and External Validation

A 9-Variable Lactate-to-Albumin Ratio-Driven Clinical Risk Score for Early Mortality Prediction in Very Elderly Mechanically Ventilated ICU Patients: Multicenter Derivation and External Validation | 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 Article A 9-Variable Lactate-to-Albumin Ratio-Driven Clinical Risk Score for Early Mortality Prediction in Very Elderly Mechanically Ventilated ICU Patients: Multicenter Derivation and External Validation Chenxi Wang, Xiujuan Hu, Jiayu Liu, Guanhua Li, Li Zhang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9412320/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Background Very elderly patients (≥ 75 years) receiving invasive mechanical ventilation in the ICU face exceptionally high mortality, yet traditional severity scores such as SOFA frequently demonstrate limited discriminative performance in this frail population. We aimed to develop and externally validate a parsimonious, clinically actionable 9-variable Clinical Risk Score (Ventilation-CRS) driven by the lactate-to-albumin ratio (LAR), using only objective baseline data available at ICU admission. Methods This retrospective multicenter study included 5,941 patients from the MIMIC-IV database (derivation) and 1,293 patients from the eICU-CRD database (external validation). A 9-variable logistic regression model was constructed and directly compared with a 35-variable XGBoost model and conventional scores (SOFA and APS III) through discrimination, calibration, reclassification (IDI), and decision curve analysis. Results In the external validation cohort, the Ventilation-CRS achieved an AUC of 0.693 (95% CI 0.659–0.721), outperforming the 35-variable XGBoost model (AUC 0.665) and markedly surpassing the SOFA score (AUC 0.437). The model demonstrated acceptable calibration, the highest net clinical benefit on decision curve analysis, and effective risk stratification (observed mortality: 0.0% low-risk, 20.2% medium-risk, 45.0% high-risk in the derivation cohort). A bedside nomogram and risk-stratified decision framework were developed to facilitate early triage and goals-of-care discussions. Conclusion This parsimonious Ventilation-CRS, anchored by admission LAR and readily available physiologic variables, provides a transparent, generalizable, and clinically practical tool that outperforms both complex machine learning approaches and traditional scoring systems. It offers immediate bedside utility for mortality risk assessment and resource allocation in very elderly mechanically ventilated patients. Health sciences/Diseases Health sciences/Health care Health sciences/Medical research Health sciences/Risk factors Mechanical Ventilation Very Elderly Mortality Prediction Lactate-to-Albumin Ratio Clinical Risk Score Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1 Introduction The global population is rapidly aging, leading to a sharp rise in very elderly patients (aged ≥ 75 years) admitted to intensive care units (ICUs). In this population, the initiation of invasive mechanical ventilation (MV) marks a pivotal and high-stakes transition, carrying exceptionally high hospital mortality, accelerated functional decline, and heavy healthcare resource burden. Profound physiological frailty markedly impairs their ability to withstand acute hypoxic stress and systemic inflammation. Therefore, early and precise prognostic stratification based solely on objective data available at ICU admission is essential to guide individualized therapy and facilitate timely goals-of-care discussions 1 , 2 . Contemporary ICU decision-making still depends predominantly on traditional severity scores, including APACHE II, SOFA, and SAPS II. However, these instruments were developed primarily in younger, more heterogeneous ICU cohorts and frequently fail to account for the distinct vulnerabilities of very elderly patients receiving MV—particularly preexisting nutritional depletion, diminished physiologic reserve, and heightened frailty. Furthermore, their dependence on numerous complex variables severely restricts bedside usability and predictive accuracy in this frail, high-risk subgroup 3 – 5 . Although machine-learning techniques such as extreme gradient boosting and LASSO-regularized models have shown promise in mortality prediction, most high-dimensional algorithms are hindered by the “black-box” effect, overfitting, limited interpretability, and poor generalizability to the specific subpopulation of very elderly patients on invasive MV. Moreover, few studies have exclusively targeted patients aged ≥ 75 years receiving MV or conducted direct head-to-head comparisons of parsimonious, clinically grounded models against complex machine-learning approaches in large, multicenter cohorts. 6 , 7 . To address these critical gaps, we developed and externally validated a parsimonious, LAR-driven 9-variable Clinical Risk Score (Ventilation-CRS) specifically for very elderly patients receiving invasive MV in the ICU. The model was constructed using the full MIMIC-IV derivation cohort, guided by clinical experience and LASSO-regularized logistic regression to identify nine objective, routinely collected baseline physiologic and laboratory variables at ICU admission. This streamlined, transparent approach—externally validated in the independent eICU cohort—yielded a model that outperforms both conventional severity scores and high-dimensional machine-learning algorithms while preserving complete bedside interpretability and clinical plausibility. The present multicenter study therefore aimed to establish and validate this accurate, explainable tool to support individualized early decision-making in very elderly mechanically ventilated ICU patients 8 – 11 . 2 Methods 2.1 Study Design and Data Sources This retrospective, multicenter cohort investigation was conducted using data extracted from two prominent clinical repositories: the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.1) and the eICU Collaborative Research Database (eICU-CRD, version 2.0). The primary model derivation relied on the MIMIC-IV registry, which captures comprehensive electronic health records from 94,458 intensive care unit (ICU) admissions at the Beth Israel Deaconess Medical Center (BIDMC) spanning the years 2008 to 2022. To robustly assess the external validity and spatial generalizability of our predictive framework, the eICU-CRD was utilized as an independent validation cohort, aggregating de-identified patient data across 208 distinct hospitals throughout the United States. Because both datasets consist of strictly anonymized and encrypted health information, the Institutional Review Boards at both the Massachusetts Institute of Technology (MIT) and BIDMC granted an exemption from the requirement for individual informed consent. The principal investigator (Certification Access No. 65378269) successfully fulfilled the Collaborative Institutional Training Initiative (CITI) requisite to obtain authorized data access. Furthermore, all study protocols strictly adhered to the ethical principles outlined in the Declaration of Helsinki, with all data sourced legitimately through the PhysioNet platform. 2.2 Study Population To maintain rigorous methodological symmetry across both the derivation (MIMIC-IV) and external validation (eICU-CRD) cohorts, we applied identical, stringent eligibility criteria. The target population consisted of highly vulnerable geriatric patients, strictly defined as individuals aged 75 years or older, who necessitated invasive MV during their hospitalization. To ensure the integrity of baseline physiological assessments and to avoid statistical confounding from recurrent critical illness, only the primary (index) ICU admission for each patient was incorporated into the analysis; all subsequent readmissions were systematically excluded. Furthermore, we excluded patients with an ICU length of stay of less than 24 hours. This temporal threshold was deliberately established to guarantee sufficient data granularity for baseline physiological profiling and to exclude early transfers or immediate, irreversible mortalities where early prognostic modeling would yield limited actionable utility. Finally, records exhibiting substantial data missingness (>20%) in essential demographic or initial vital sign parameters were removed from the analytical cohort. The detailed, step-by-step patient attrition and final cohort assembly are meticulously delineated in the study flowchart ( Figure 1 ). 2.3 Data Extraction and Missing Value Handling Clinical parameters, vital signs, and laboratory diagnostics were acquired through Structured Query Language (SQL) scripts. To exclusively capture the patient's initial physiological derangement and maintain the model's immediate early-warning utility, all continuous or repeated variables were restricted to the first available (baseline) values recorded upon ICU admission. This approach ensures that the Ventilation-CRS provides actionable prognostic information at the earliest clinical juncture, without the requirement for a 24-hour observation window. The Lactate-to-Albumin Ratio (LAR), a pivotal biomarker of coupled microcirculatory and nutritional deficit, was mathematically derived by dividing the baseline blood lactate concentration (mmol/L) by the serum albumin level (g/dL). Given the retrospective nature of real-world electronic health records, data missingness is inevitable. As a general methodological heuristic, candidate variables exhibiting a missingness rate exceeding 30% across the cohorts were considered for exclusion to preserve data robustness. However, a strict, a priori clinical exception was implemented. Variables deemed fundamentally indispensable for the assessment of acute respiratory failure and tissue hypoperfusion-specifically Lactate (missing in 42.6% of the eICU cohort), Albumin (missing in 35.4% of the MIMIC cohort), SaO 2 (34.8% in MIMIC), and Neutrophil counts (34.2% in MIMIC)-were purposefully retained. Excluding these cores homeostatic markers based solely on an arbitrary statistical threshold would severely compromise the biological validity and prognostic depth of the study. Assuming the data were missing at random (MAR)-a standard assumption in critical care registries where test frequency fluctuates based on clinical acuity-we employed the Multiple Imputation by Chained Equations (MICE) algorithm to robustly handle the missing values. This sophisticated predictive mean matching approach generated five parallel imputed datasets, minimizing potential bias while securely preserving the physiological architecture of our predictive framework. The robustness of including these high-missingness variables was subsequently confirmed through complete-case sensitivity analysis (see Section 3.6). Model robustness was further evaluated by complete-case sensitivity analysis restricted to patients with complete data for all nine variables ( n = 5,941; Supplementary Table S1) . 2.4 Statistical Analysis and Model Evaluation Baseline continuous variables, exhibiting predominantly non-normal distributions, were reported as medians accompanied by interquartile ranges (IQR) and compared utilizing the Mann-Whitney U test. Categorical variables were expressed as absolute frequencies with percentages and assessed via the Chi-square test. The parsimonious Ventilation-CRS was formulated using multivariable logistic regression within the MIMIC-IV derivation cohort, with variable inclusion strictly guided by a priori clinical reasoning and baseline statistical significance. To rigorously construct a high-dimensional comparative baseline, an advanced machine learning pipeline was simultaneously developed. The derivation data was partitioned (70:30 ratio) to train an Extreme Gradient Boosting (XGBoost) classifier. To systematically mitigate overfitting, 40 candidate features were initially regularized using the Least Absolute Shrinkage and Selection Operator (LASSO) with an L1 penalty, successfully isolating a subset of 35 optimal predictors for XGBoost training. To strictly circumvent potential collinearity between arterial oxygen saturation (SaO 2 ) and peripheral capillary oxygen saturation (SpO₂), which represent identical physiological constructs, SpO₂ was pre-emptively excluded prior to LASSO regularization. The risk-stratification thresholds (20% for high-risk) were pre-defined based on clinical utility: the low-risk threshold targets candidates for early liberation from MV, while the high-risk threshold acts as a clinical trigger for escalated multidisciplinary review and proactive Goals of Care (GOC) discussions. The eICU-CRD dataset served as the independent external validation cohort. Discriminative capacity across all models (Ventilation-CRS, XGBoost, SOFA, and APS III) was quantified using the Area Under the Receiver Operating Characteristic curve (AUC), with formal pairwise statistical comparisons executed via the DeLong method. Calibration fidelity was evaluated visually through calibration curves and quantitatively using the Brier score. To evaluate tangible clinical utility, Decision Curve Analysis (DCA) was deployed to estimate the net clinical benefit across a continuum of threshold probabilities. Furthermore, the Integrated Discrimination Improvement (IDI) index was computed to assess the exact reclassification superiority of the Ventilation-CRS over legacy severity scores. To decrypt the "black-box" nature of predictive modeling and provide bedside transparency, SHapley Additive exPlanations (SHAP) values were extracted to visualize the individualized pathophysiological drivers behind specific mortality predictions. Finally, rigorous subgroup sensitivity analyses (presented via forest plots) were conducted across distinct chronic comorbidity profiles and advanced age strata to confirm spatial and demographic generalizability. To robustly validate the internal stability of the logistic regression-based Ventilation-CRS and account for potential model overfitting, internal validation was performed using 1,000 bootstrap resamples on the derivation cohort to compute the optimism-corrected AUC and its 95% confidence intervals (CI). All analytical procedures and visualizations were executed utilizing Python (version 3.9) and R software (version 4.2.2). A two-tailed p -value of less than 0.05 was considered to denote statistical significance. 3 Results 3.1 Participant Selection and Flow The detailed flowing screening process for study participants from both distinct multicenter databases is horizontally delineated in Figure 1 . In the derivation cohort from the MIMIC-IV database (v3.1), we initiated the selection with a comprehensive pool of 94,458 distinct patients. The first round of pragmatic screening simultaneously applied the predefined inclusion and exclusion criteria, focusing strictly on geriatric patients (Age ≥ 75 years), their first documented ICU admission, and a required ICU length of stay of at least 24 hours. This initial wave of screening refined the potential population to 16,329 individuals. Subsequently, a critical second-round precise refinement was performed to include only those who underwent invasive mechanical ventilation during their hospitalization. This robust two-step selection process culminated in a final derivation cohort of 5,941 patients. To maintain perfect methodological symmetry for external validation, an identical hierarchical screening pathway was implemented within the eICU-CRD database (v2.0), which initially comprised 200,859 patients. Following the same first-round parameters for age, first ICU admission, and stay duration, a stratified population of 28,810 patients was obtained. The identical second-round requirement for invasive mechanical ventilation therapy ultimately defined a highly validated and stable external cohort of 1,293 patients. 3.2 Baseline Characteristics of the Study Populations A total of 7,234 mechanically ventilated patients were included: 5,941 in the MIMIC-IV derivation cohort (overall mortality, 22.3%) and 1,293 in the eICU external validation cohort (overall mortality, 32.3%). Baseline demographic and clinical characteristics are summarized in Table 1 . Across both cohorts, non-survivors were significantly older and presented with worse admission physiology, characterized by profound hypoxia (lower PaO 2 ), exacerbated acidemia (lower pH), and pronounced multi-organ dysfunction (all p <0.05). Notably, biomarkers reflecting the intersection of acute hypoxic stress and baseline nutritional depletion were highly discriminative: non-survivors exhibited significantly higher lactate and lower albumin levels, yielding a markedly elevated LAR in both the derivation (0.7 vs 0.4, p < 0.001) and validation (1.0 vs 0.6, p < 0.001) cohorts. Severity of illness, as measured by the APS III score, was consistently higher among non-survivors. However, the atypically low median SOFA score observed in the MIMIC-IV cohort (median, 2.0) highlights the inherent limitations of sub-score documentation (e.g., Glasgow Coma Scale) in retrospective electronic health records, reinforcing the necessity of laboratory-driven prognostic models. The overlapping median SOFA scores (median 2.0 for both groups) further illustrate its limited early discriminative power in this specific population. Table 1. Baseline Demographic and Clinical Characteristics of Mechanically Ventilated Patients Stratified by Hospital Mortality. Variable MIMIC p -value eICU p -value Survived ( n = 4615) Non-survived ( n = 1326) Survived ( n = 875) Non-survived ( n = 418) Demographics Age, years 81.1 [77.8-85.1] 83.0 [79.0-87.5] <0.001 80.0 [77.0-84.0] 82.0 [78.0-86.0] <0.001 Male, n (%) 2579 (55.9%) 685 (51.7%) 0.007 435 (49.8%) 224 (53.6%) 0.228 Weight, kg 81.0 [68.9-93.9] 76.2 [64.0-90.7] <0.001 74.8 [63.4-88.4] 74.6 [62.0-87.6] 0.499 Comorbidities, n (%) CHF 1663 (36.0%) 592 (44.6%) <0.001 138 (15.8%) 75 (17.9%) 0.366 COPD 1273 (27.6%) 389 (29.3%) 0.223 106 (12.1%) 49 (11.7%) 0.911 Diabetes 1491 (32.3%) 432 (32.6%) 0.878 136 (15.5%) 71 (17.0%) 0.561 Liver disease 44 (1.0%) 42 (3.2%) <0.001 8 (0.9%) 10 (2.4%) 0.062 Renal disease 1242 (26.9%) 443 (33.4%) <0.001 114 (13.0%) 54 (12.9%) 1 Vital Signs SBP, mmHg 119.0 [104.0-136.0] 121.0 [103.0-140.0] 0.024 128.0 [108.0-149.5] 120.0 [101.0-141.0] 0.002 DBP, mmHg 59.0 [51.0-70.0] 62.5 [52.0-76.0] <0.001 67.0 [56.0-79.5] 64.0 [52.0-78.5] 0.092 HR, bpm 80.0 [72.0-90.0] 88.0 [74.0-102.0] <0.001 85.0 [73.0-99.0] 86.5 [73.0-108.0] 0.031 RR, bpm 16.0 [14.0-20.0] 20.0 [16.0-24.0] <0.001 18.0 [15.0-21.0] 20.0 [16.0-24.0] <0.001 SpO 2 , % 100.0 [97.0-100.0] 98.0 [95.0-100.0] <0.001 99.0 [96.0-100.0] 98.0 [95.0-100.0] 0.003 Respiratory & Gas Balance PaO 2 , mmHg 243.0 [100.0-382.0] 99.5 [61.0-183.8] <0.001 118.7 [78.0-218.0] 107.0 [70.0-180.0] 0.01 FiO 2 , % 54.0 [45.6-61.1] 51.4 [43.8-62.5] 0.2 60.0 [40.0-100.0] 60.0 [40.0-100.0] 0.026 PaCO 2 , mmHg 40.0 [36.0-45.0] 41.0 [35.0-49.0] 0.355 39.0 [33.4-46.1] 40.2 [33.0-48.0] 0.477 pH 7.39 [7.34-7.44] 7.36 [7.27-7.42] <0.001 7.39 [7.32-7.44] 7.36 [7.27-7.42] <0.001 BE, mmol/L 0.0 [-2.0-2.0] -2.0 [-6.0-1.0] <0.001 0.1 [-3.7-3.0] -1.8 [-6.0-2.6] <0.001 SaO2, % 97.0 [95.0–98.0] 95.0 [83.0–98.0] < 0.001 98.0 [95.3–99.9] 98.0 [94.0–99.7] 0.048 Laboratory Findings Lactate, mmol/L 1.4 [1.1-2.0] 1.9 [1.3-3.2] <0.001 1.7 [1.1-2.7] 2.3 [1.4-4.0] <0.001 Albumin, g/dL 3.4 [2.8-3.8] 2.9 [2.5-3.3] <0.001 3.0 [2.5-3.4] 2.6 [2.2-3.1] <0.001 LAR 0.4 [0.3-0.6] 0.7 [0.4-1.2] <0.001 0.6 [0.4-1.0] 1.0 [0.5-1.7] <0.001 LY, ×10⁹/L 1.3 [0.8-1.9] 0.9 [0.5-1.4] <0.001 1.0 [0.6-1.5] 1.0 [0.6-1.4] 0.147 WBC , ×10⁹/L 9.6 [7.1-13.3] 11.7 [8.3-16.4] <0.001 10.6 [7.8-14.5] 11.9 [8.6-17.8] <0.001 Plt, ×10⁹/L 182.0 [137.0-237.0] 188.0 [136.0-259.0] 0.008 189.0 [143.0-242.0] 182.5 [140.8-246.0] 0.931 Hb, g/dL 10.9 [9.2-12.4] 10.7 [9.4-12.2] 0.317 11.2 [9.5-12.5] 10.8 [9.3-12.2] 0.05 Cr, mg/dL 1.0 [0.8-1.4] 1.2 [0.9-1.8] <0.001 1.1 [0.8-1.6] 1.3 [0.9-1.9] <0.001 BUN, mg/dL 21.0 [16.0-31.0] 28.0 [19.0-45.0] <0.001 23.0 [16.2-35.0] 28.0 [19.0-44.0] <0.001 Glu, mg/dL 126.0 [104.0-158.0] 140.0 [112.0-183.0] <0.001 130.0 [104.0-169.0] 143.0 [108.0-195.2] 0.002 Na mmol/L 139.0 [137.0-142.0] 139.0 [136.0-142.0] 0.073 139.0 [137.0-142.0] 138.0 [135.0-141.8] 0.004 K, mmol/L 4.1 [3.8-4.5] 4.2 [3.8-4.7] <0.001 4.0 [3.7-4.5] 4.2 [3.7-4.6] 0.012 Cl, mmol/L 105.0 [101.0-109.0] 104.0 [100.0-108.0] <0.001 105.0 [101.0-109.0] 105.0 [100.0-109.0] 0.15 HCO 3 , mmol/L 24.0 [21.0-26.0] 22.0 [19.0-25.5] <0.001 24.0 [21.0-27.0] 23.0 [19.4-27.0] <0.001 AG, mmol/L 13.0 [11.0-16.0] 15.0 [13.0-18.0] <0.001 10.0 [7.0-12.4] 10.0 [8.0-14.0] 0.015 TBil, mg/dL 0.6 [0.4-0.9] 0.7 [0.4-1.1] <0.001 0.6 [0.4-0.9] 0.7 [0.5-1.2] <0.001 ALT, U/L 20.0 [14.0-34.0] 26.0 [15.0-63.5] <0.001 22.0 [14.0-38.0] 26.0 [16.0-55.8] <0.001 AST, U/L 28.0 [20.0-47.0] 43.0 [25.0-101.0] <0.001 26.0 [19.0-49.0] 39.0 [23.2-90.0] <0.001 ALP, U/L 72.0 [55.0-97.0] 82.0 [61.0-118.5] <0.001 72.0 [56.0-93.0] 80.0 [64.0-105.0] <0.001 INR 1.2 [1.1-1.4] 1.3 [1.1-1.6] <0.001 1.2 [1.1-1.4] 1.3 [1.1-1.7] <0.001 Clinical Scores SOFA score 2.0 [1.0-4.0] 2.0 [0.0-4.0] 0.018 9.0 [7.0-12.0] 10.0 [8.0-13.0] <0.001 APS III score 42.0 [32.0-56.0] 60.0 [45.0-79.0] <0.001 58.0 [43.0-79.0] 74.0 [55.8-97.0] <0.001 Notes: Data are presented as median [interquartile range] for continuous variables and No. (%) for categorical variables. Differences between survival groups were compared using the Mann-Whitney U test for continuous variables and the Chi-square test for categorical variables. 3.3 Construction and Validation of the Ventilation-CRS Model The Ventilation-CRS model was constructed using the initial values of nine pre-selected, objective clinical variables recorded upon ICU admission. By prioritizing the first available data points following admission, the model ensures maximum utility for early bedside triage. The selection of these nine variables was primarily guided by a priori clinical reasoning focusing on respiratory, metabolic, and systemic stability. This was further corroborated by preliminary univariable screenings in the MIMIC-IV cohort, ensuring each included marker was both biologically plausible and statistically significant. Multivariable logistic regression analysis ( Table 2 ) identified the LAR as the most potent independent predictor of in-hospital mortality (OR, 1.816; 95% CI, 1.606–2.053; p < 0.001). Other parameters, including age, glucose, neutrophil count, AST, INR, and respiratory rate, were also identified as significant risk factors. Conversely, baseline oxygen saturation (SaO 2 ) demonstrated a significant protective effect (OR, 0.977; 95% CI, 0.971–0.983; p < 0.001). The predictive performance and clinical utility of the Ventilation-CRS model are comprehensively illustrated in Figure 2 . The model demonstrated satisfactory discriminative ability, yielding an apparent area under the curve (AUC) of 0.753 in the MIMIC-IV derivation cohort. Crucially, to rigorously account for potential model overfitting and ensure internal stability, we performed an internal validation using 1,000 bootstrap resamples. This stringent analysis revealed a minimal optimism of 0.0019, yielding an optimism-corrected AUC of 0.752 (95% CI, 0.738–0.765). This indicates highly robust internal consistency with negligible overfitting. Furthermore, the model maintained a stable AUC of 0.693 in the completely independent and highly heterogeneous eICU external validation cohort ( Figure 2A ). Calibration analysis in the external cohort revealed acceptable agreement between the predicted probabilities and observed mortality, with a calibration slope of 0.82 and an intercept of 0.22 ( Figure 2B ). Decision curve analysis (DCA) confirmed that the Ventilation-CRS model provided a consistent net clinical benefit across a wide and clinically relevant range of threshold probabilities compared to the default "treat-all" or "treat-none" strategies ( Figure 2C ). Furthermore, the model effectively stratified patients in the external cohort into low ( 20%) risk categories. The observed mortality rates exhibited a clear stepwise increase: 0.0% for the low-risk, 20.2% for the medium-risk, and 45.0% for the high-risk groups ( Figure 2D ). Notably, the absence of mortality (0.0%) in the strictly defined low-risk group accurately reflects the inherently high baseline acuity of patients requiring mechanical ventilation, further validating the model's reliability in safely identifying true high-risk populations for actionable bedside interventions. Although Na exhibited marginal statistical significance in the multivariable model ( p = 0.093), it was deliberately retained in the final Ventilation-CRS because sodium balance is a fundamental homeostatic marker in the geriatric population, where electrolyte derangements are strongly associated with underlying frailty and acute mortality. Excluding this core physiological parameter based strictly on a p -value threshold would diminish the model's clinical grounding and face validity. In the complete-case sensitivity analysis ( n = 5,941), all coefficients showed minimal changes (<6%) compared with the primary multiple imputation model ( Supplementary Table S1 ). Notably, the LAR ratio remained the most potent predictor with an identical odds ratio of 1.816. Table 2. Multivariable Logistic Regression Analysis of the 9-Variable Ventilation-CRS Model. Variable Coef OR (95% CI) p -value Age, y 0.0523 1.054 (1.040-1.067) < 0.001 LAR 0.5965 1.816 (1.606-2.053) < 0.001 SaO 2 , % -0.023 0.977 (0.971-0.983) < 0.001 Glu, mg/dL 0.0019 1.002 (1.001-1.003) < 0.001 Na, mmol/L -0.0124 0.988 (0.974-1.002) 0.093 NE,×10 9 /L 0.0389 1.040 (1.027-1.052) < 0.001 AST, U/L 0.0006 1.001 (1.000-1.001) < 0.001 INR 0.1203 1.128 (1.019-1.249) 0.02 RR, bpm 0.0628 1.065 (1.053-1.077) < 0.001 3.4 Diminishing Returns of Machine Learning Complexity To rigorously evaluate whether incorporating a larger number of variables and utilizing complex, non-linear algorithms could yield superior prognostic performance, we developed an advanced machine learning pipeline. The MIMIC-IV derivation cohort was randomly partitioned into a training set ( n = 4158, 70%) and an internal validation set ( n = 1783, 30%), while the eICU dataset served as the external validation cohort ( N = 1293). Initially, 39 candidate clinical features (excluding peripheral oxygen saturation to strictly prevent multicollinearity with arterial oxygen saturation) were introduced into a Least Absolute Shrinkage and Selection Operator (LASSO) regression model. The L1 penalty effectively regularized the model, eliminating redundant variables and identifying a subset of 35 optimal predictors ( Figure 3A) . Subsequently, an extreme gradient boosting (XGBoost) model was trained using these 35 features. Variable importance analysis, quantified by the Gain metric, revealed the top contributors to the XGBoost model ( Table 3, Figure 3B ). Notably, the LAR maintained a top-tier prognostic weight, ranking as the third most important predictor (Gain = 10.32) across the entire high-dimensional feature space, surpassed only by PaO 2 and respiratory rate, further confirming its robust biological and prognostic relevance in mechanically ventilated patients. Despite the sophisticated algorithmic architecture and the inclusion of 35 features, the machine learning model exhibited classic signs of severe overfitting and poor external generalizability. While the XGBoost model achieved an exceptionally high AUC of 0.980 in the training set and 0.795 in the internal validation set, its discriminative performance sharply declined to an AUC of 0.665 in the eICU external validation cohort ( Figure 3C1 ). Furthermore, external calibration was suboptimal, demonstrating notable deviation from the ideal line with a calibration slope of 0.55 and an intercept of -0.47 ( Figure 3C2 ). Compared to the 9-variable clinical experience model (Ventilation-CRS, external AUC = 0.693) developed in Section 3.2, the heavily parameterized 35-variable XGBoost model completely failed to improve predictive accuracy in unseen data. These findings strongly demonstrate a "diminishing return of complexity" in this clinical context, highlighting that a parsimonious model strictly grounded in fundamental, objective physiological parameters offers superior robustness, stability, and clinical utility across heterogeneous ICU settings. Table 3. XGBoost Predictor Importance Ranking Based on Gain Values. Predictor Gain Value Importance Rank PaO 2 18.92723 1 RR 10.39721 2 LAR 10.32179 3 LY 7.963913 4 ALT 7.109571 5 BUN 6.049428 6 AST 6.035837 7 BE 5.792546 8 FiO2 5.764657 9 NE 5.756567 10 pH 5.635104 11 INR 5.547956 12 Age 5.301125 13 WBC 5.271259 14 HR 5.054357 15 Notes: The table presents the predictors selected by the LASSO regression TOP 15 (35 in total), ranked by their relative contribution (Gain value) to the XGBoost model's predictive accuracy. 3.5 Model Performance, Head-to-Head Comparison, and Clinical Utility To thoroughly evaluate the robustness and clinical value of the Ventilation-CRS model, we conducted a head-to-head comparison against the high-dimensional XGBoost model and two universally established clinical scoring systems (SOFA and APS III) using the independent eICU validation cohort. The comprehensive performance metrics are summarized in Table 4 . In terms of discrimination, the 9-variable Ventilation-CRS model demonstrated the highest predictive accuracy, achieving an AUC of 0.693 (95% CI, 0.659-0.721). It significantly outperformed both the traditional SOFA score (AUC = 0.437; p < 0.001) and the APS III score (AUC = 0.651; p < 0.001) ( Figure 4A ). Notably, the SOFA score exhibited an AUC below 0.50, indicating a complete failure to discriminate mortality risk within this specific mechanically ventilated geriatric cohort. Furthermore, the parsimonious Ventilation-CRS model surpassed the complex 35-variable XGBoost model (AUC = 0.665) in the unseen external data, strongly reinforcing our previous observation regarding the diminishing returns of machine learning complexity and highlighting the superior generalizability of our clinical model. Regarding calibration, the Ventilation-CRS model achieved the lowest Brier score (0.2049) among all evaluated models. The calibration curve ( Figure 4D ) visually confirmed that the Ventilation-CRS probabilities aligned most closely with the ideal reference line, whereas the XGBoost model and traditional scores exhibited varying degrees of overestimation or underestimation. The Integrated Discrimination Improvement (IDI) analysis corroborated these findings, demonstrating that the Ventilation-CRS model provided a massive reclassification improvement over the SOFA score (IDI = -0.1035 in favor of CRS). While XGBoost showed nominal fluctuations in reclassification (IDI = 0.0234), its overall clinical reliability was compromised by inferior calibration and discrimination. Clinical utility was further evaluated via Decision Curve Analysis (DCA) ( Figure 4B ). The Ventilation-CRS model consistently provided the highest net clinical benefit across the entire clinically relevant threshold probability range (0.05–0.50). Conversely, the SOFA score offered no clinical benefit, plunging below the "treat-none" reference line almost immediately. Finally, risk stratification accuracy was critically assessed against the overall baseline mortality (32.3%) of the external cohort ( Figure 4C ). The Ventilation-CRS model effectively separated patients into logical trajectories, consistently distinguishing those above and below the baseline average. Notably, in the high-acuity external eICU cohort (median SOFA = 9.0), no patients (n = 0) met the ultra-low-risk threshold (<5%), reflecting the model's conservative and safe orientation when faced with extremely ill geriatric populations. In contrast, the Ventilation-CRS demonstrated a clear and reliable prognostic gradient in the medium-risk (5%–20%, observed mortality 20.2%) and high-risk (>20%, observed mortality 45.0%) strata. In stark contrast, traditional scoring systems demonstrated dangerous clinical misclassification. For instance, patients classified as "low risk" ( 20%) by SOFA exhibited a mortality rate of only 29.0%, which was even lower than the baseline mortality of the entire cohort. These results collectively highlight the Ventilation-CRS model as a highly accurate, reliable, and clinically safe decision-support tool that avoids the perilous underestimation of mortality risk in vulnerable older adults. Table 4. Head-to-Head Comparison of Model Performance in the External Validation Cohort (eICU). Model AUC (95% CI) Brier Score IDI vs CRS p -value (DeLong) Ventilation-CRS 0.693 (0.659-0.721) 0.2049 Ref Ref XGBoost 0.665 (0.634-0.693) 0.2144 0.0234 ns SOFA 0.437 (0.403-0.471) 0.2336 -0.1035 < 0.001 APS III 0.651 (0.616-0.690) 0.2085 0.0048 0.05 for the DeLong test compared to the Ventilation-CRS model. IDI values were calculated relative to the Ventilation-CRS model; a negative value for SOFA indicates significantly inferior reclassification performance compared to our parsimonious model. Note regarding risk stratification: The 0.0% observed mortality in the low-risk group (<5%) was recorded in the MIMIC-IV derivation cohort; in the higher-acuity external eICU cohort, no patients met this threshold ( n = 0). 3.6 Clinical Significance, Interpretability, and Sensitivity Analysis To bridge the gap between population-level statistical metrics and individualized bedside decision-making, we employed SHapley Additive exPlanations (SHAP) to interpret the Ventilation-CRS model. This method visualizes the positive or negative contribution of each physiological parameter to the final mortality risk for individual patients. Figure 5 displays three representative SHAP waterfall plots for patients stratified into high (~50%), medium (~15%), and the lowest predicted risk categories in the external validation cohort. The base value represents the cohort's average predicted risk, with red bars pushing the individual risk higher and blue bars pulling it lower. For instance, in the high-risk patient ( Figure 5A ), an elevated LAR and advanced age served as the primary drivers of increased mortality risk. Conversely, in the lowest-risk patient ( Figure 5C ), robust physiological reserves-characterized by a healthy respiratory rate, stable SaO 2 , and normal sodium levels-acted as strong protective factors, effectively reducing the predicted mortality to 5.2%. This individualized transparency allows clinicians to identify the precise physiological derangements driving a specific patient's deterioration. In the complete-case sensitivity analysis ( n = 5,941), all coefficients showed minimal changes (<6%) compared with the primary multiple imputation model ( Supplementary Table S1 ). Notably, the LAR ratio remained the most potent predictor with an identical odds ratio of 1.816. This consistency across both imputed and non-imputed datasets confirms the robust structural integrity of the Ventilation-CRS. To rigorously evaluate the generalizability of the Ventilation-CRS model across highly heterogeneous ICU populations, we conducted a comprehensive sensitivity analysis assessing the model's discriminative performance (AUC) across major chronic comorbidities and advanced age subgroups. As illustrated in the forest plot ( Figure 6 ), the predictive accuracy of the Ventilation-CRS model remained remarkably robust and consistent. Regardless of whether patients presented with concomitant congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD), diabetes, or renal disease, the subgroup AUCs tightly clustered around the overall cohort AUC line (0.693, red dashed line). Importantly, the model maintained its robust performance across different geriatric strata, achieving an AUC of 0.694 in patients aged 75-84 years and remaining stable at 0.677 in the extremely elderly group (≥85 years). These findings confirm that the Ventilation-CRS model accurately captures fundamental acute physiological stress signals, demonstrating reliable prognostic utility independent of baseline chronic disease phenotypes or extreme age. Furthermore, given the high missing rates of baseline lactate and albumin, we performed a stringent complete-case sensitivity analysis ( n = 5,941) to verify that multiple imputation did not introduce significant bias. All coefficients showed minimal changes (<6%), and the LAR remained the strongest predictor with an identical odds ratio of 1.816 ( Supplementary Table S1 ). The overall model performance remained robust, confirming that the missing-at-random (MAR) assumption was valid and that the high prognostic weight of LAR is biologically authentic rather than a statistical artifact of imputation. 3.7 Clinical Translation: Bedside Nomogram and Risk-Based Decision Framework To facilitate the seamless integration of the Ventilation-CRS model into daily clinical practice, we developed two highly actionable bedside tools: a prognostic nomogram and a risk-stratified decision-making framework. The 9-variable nomogram ( Figure 7 ) provides an intuitive visual interface for individualized risk assessment, enabling clinicians to derive a precise probability of in-hospital death based on initial baseline values recorded at the onset of mechanical ventilation. This bedside-ready tool eliminates the need for complex computational devices, providing immediate prognostic clarity at the earliest clinical juncture. By locating a patient’s specific values for the nine objective predictors-including Age, SaO 2 , respiratory rate, Sodium, Glucose, AST, INR, Neutrophil count, and LAR-and projecting them onto the standardized points scale, a cumulative "Total Points" score can be calculated and mapped directly to the final mortality risk axis. Beyond mere risk quantification, we further proposed a structured clinical decision-making framework to translate these statistical predictions into specific management strategies ( Figure 8 ). Based on the model’s validated thresholds, patients are triaged into three distinct risk trajectories. For the Low-Risk group (predicted mortality ≤ 5%, observed 0.0%), clinical triggers such as stable SaO 2 and lower respiratory rates indicate a high potential for successful liberation from the ventilator; therefore, management should prioritize early spontaneous breathing trials and extubation. Conversely, the High-Risk group (predicted mortality > 20%, observed 45.0%) represents the most critical cohort, where triggers such as significantly elevated LAR, advanced age (≥ 85 years), or severe coagulopathy necessitate aggressive care escalation. In this high-acuity setting, the framework serves as a vital clinical trigger for multidisciplinary review and early proactive family communication regarding the goals of care. By linking objective physiological markers to targeted management pathways, this integrated system ensures that the Ventilation-CRS model functions as a comprehensive decision-support tool for optimizing outcomes in older critically ill patients. 4 Discussion A notable finding in this study was the clear diminishing returns of increasing model complexity. Despite achieving an exceptionally high AUC of 0.980 in the training set, the 35-variable XGBoost model exhibited substantial performance degradation in the independent external eICU cohort (AUC 0.665). In contrast, our parsimonious 9-variable Ventilation-CRS maintained superior discriminative ability (AUC 0.693) with markedly better calibration and clinical net benefit. These results illustrate that, in the heterogeneous population of very elderly patients receiving invasive mechanical ventilation, overly complex machine-learning algorithms often capture database-specific noise rather than generalizable physiology. This supports a return to the principle of parsimony in clinical prediction modeling 12 , 13 . Head-to-head comparison further revealed the limited discriminative performance of the SOFA score in this specific cohort (AUC 0.437). This finding is consistent with known challenges in applying traditional organ failure-based scores to very elderly patients, where factors such as sedation, incomplete Glasgow Coma Scale documentation, and baseline frailty may substantially underestimate risk. In contrast, the Ventilation-CRS provided a clearer prognostic gradient across risk strata and avoided dangerous underestimation of mortality, particularly in high-acuity populations. To facilitate rapid translation into clinical practice, we developed a bedside nomogram and a structured risk-stratified decision-making framework. The nomogram enables clinicians to calculate individualized in-hospital mortality probability using only nine objective admission variables, without requiring complex computation. The accompanying framework links predefined risk categories (low, medium, and high) to specific management recommendations, such as prioritizing early spontaneous breathing trials and extubation in low-risk patients, and triggering multidisciplinary review with proactive goals-of-care discussions in high-risk patients. By connecting readily available physiologic markers to actionable pathways, these tools offer immediate bedside utility for early triage, resource allocation, and ethical decision-making in older critically ill patients. Several limitations should be acknowledged. First, as a retrospective study using large multicenter databases, missing data were handled with multiple imputation; however, complete-case sensitivity analyses confirmed the robustness of the key findings, including the dominant role of the lactate-to-albumin ratio (LAR). Second, both derivation and validation cohorts were derived from Western healthcare systems, which may limit generalizability to other regions and ethnic groups; external validation in Asian or other diverse populations would be valuable 14 , 15 . Third, the model relies solely on baseline admission data and does not incorporate dynamic changes during the ICU stay. Future prospective, multicenter studies are warranted to evaluate the real-world impact of implementing the Ventilation-CRS on clinical decision-making, resource utilization, and patient-centered outcomes 16 , 17 . In conclusion, the Ventilation-CRS represents a practical step toward simplicity in geriatric critical care 18 – 20 .By integrating only nine objective, routinely collected markers at ICU admission—with the lactate-to-albumin ratio as the strongest driver—this parsimonious and transparent model safely outperforms both high-dimensional machine-learning approaches and conventional scoring systems such as SOFA 21 – 23 . Our findings demonstrate that clinically grounded, parsimonious tools leveraging initial baseline data can provide superior bedside safety and decision support for the most vulnerable elderly mechanically ventilated population. The accompanying nomogram and risk-stratified framework further empower clinicians to optimize early triage, resource allocation, and goals-of-care discussions without the need for complex computation 24 , 25 . 5 Conclusion The Ventilation-CRS represents a paradigm shift toward simplicity in geriatric critical care. By utilizing only nine objective markers routinely collected at the time of ICU admission, this parsimonious model offers a high-performance, transparent, and stable prognostic tool that safely outperforms both high-dimensional machine-learning algorithms and traditional scoring systems. Our findings demonstrate that clinically-grounded models leveraging initial baseline data can provide superior bedside safety and decision support for the most vulnerable elderly MV population. By linking objective physiological markers to targeted management pathways, this framework empowers clinicians to optimize early triage, resource allocation, and ethical goals-of-care discussions without the need for complex computation. Declarations Data Availability Statement The data that support the findings of this study are available from the PhysioNet repository (MIMIC‑IV and eICU‑CRD databases). These data are available under a license and are not publicly available. Data can be obtained from the corresponding author upon reasonable request, subject to approval from PhysioNet, completion of the required CITI training, and signing of the data use agreement. Conflict of Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Ethics statement The creation and public release of this database were approved by the Institutional Review Boards (IRBs) of the Massachusetts Institute of Technology (Cambridge, MA) and Beth Israel Deaconess Medical Center (Boston, MA). As the MIMIC-IV v3.1 database consists entirely of de-identified, retrospective, publicly available clinical data, this study was determined to be exempt from both full ethical review and the requirement for informed consent. Author Contributions C.W. (Chenxi Wang): Conceptualization, Methodology, Software, Data curation, Formal analysis, Visualization, Writing-original draft. X.H. (Xiujuan Hu) and J.L. (Jiayu Liu): Investigation, Resources. G.L. (Guanhua Li): Project administration, Supervision. L.Z. (Li Zhang): Conceptualization, Supervision, Validation, Writing-review & editing. All authors critically reviewed the manuscript and approved the final submitted version. Funding Tianjin Key Medical Discipline Construction Project, No. TJYXZDXK-3-032C. Acknowledgments The authors gratefully acknowledge the Computational Physiology Laboratory (LCP) at the Massachusetts Institute of Technology (MIT) for its sustained stewardship and maintenance of the MIMIC-IV v3.1 database. References Cao, M. et al. Population-wise incidence and outcomes of patients requiring invasive and non-invasive mechanical ventilation in China: a nationwide retrospective analysis by age, sex, and comorbidity. Ann. Intensive Care . 15 (1), 120. 10.1186/s13613-025-01537-w (2025). Mahran, G. S. K., El-aty, N. S. A., Abd-Elhamed, A. G., Ali, M. M. A. & Kasasbeh, M. A. M. Comparison of baseline characteristics and outcomes in ICU patients with endotracheal tubes versus tracheostomies: a prospective observational study. Nurs. Crit. Care . 31 (2), e70352. 10.1111/nicc.70352 (2026). Nguyen, B. H. et al. Long-term mortality in patients with chronic obstructive pulmonary disease requiring acute non-invasive ventilation with and without obstructive sleep apnoea. BMJ Open. Respir Res. 11 (1), e002496. 10.1136/bmjresp-2024-002496 (2024). Tekin, B. et al. Comparison of scoring systems: SOFA, APACHE-II, LODS, MODS, and SAPS-II in critically ill elderly sepsis patients. J. Infect. Dev. Ctries. 18 (1), 122–130. 10.3855/jidc.18526 (2024). Zhou, J., Luo, X. Y., Li, H. L., Shi, G. Z. & Chen, G. Q. Comparison of the predictive value of APACHE II, SOFA, SAPS II, GCS and GCS-P scores for in-hospital mortality in critically ill patients after craniotomy: a retrospective cohort study in a Chinese tertiary hospital. BMJ Open. 16 (2), e101867. 10.1136/bmjopen-2025-101867 (2026). Li, H. et al. A machine learning-based prediction of hospital mortality in mechanically ventilated ICU patients. PLoS One . 19 (9), e0309383. 10.1371/journal.pone.0309383 (2024). Cao, Z. et al. Association between the platelet-albumin-bilirubin score and all-cause mortality in ICU-admitted heart failure patients: a retrospective cohort analysis and machine learning-based prognostic modeling. Front. Cardiovasc. Med. 12 , 1622554. 10.3389/fcvm.2025.1622554 (2025). Wei, J. et al. Nomogram predictive model for in-hospital mortality risk in elderly ICU patients with urosepsis. BMC Infect. Dis. 24 (1), 442. 10.1186/s12879-024-09319-8 (2024). Zeng, Y., Tan, N., He, X., Peng, S. & Qiu, E. Development of a nomogram to predict in-hospital mortality of trauma patients in the ICU: an analysis of the MIMIC-IV database. Sci. Rep. 16 (1), 6802. 10.1038/s41598-026-38251-x (2026). Hu, Y., Zhong, Y., Fan, F., Shen, Y. & Jiang, C. Development and validation of a bedside nomogram for predicting 30-day mortality in patients with type 1 hepatorenal syndrome: a retrospective cohort study. Med. (Baltim). 105 (8), e47640. 10.1097/MD.0000000000047640 (2026). Shao, D., He, L., Zhong, D., Li, K. & Zhang, X. Association between the geriatric nutritional risk index and 28-day mortality in critically ill sepsis-associated pneumonia patients: retrospective study based on two cohorts. Front. Nutr. 12 , 1698973. 10.3389/fnut.2025.1698973 (2026). Mu, Y., Lou, F., Feng, G. & Huang, M. Lactate to albumin ratio predicts one year mortality in ICU patients receiving invasive mechanical ventilation. Sci. Rep. 16 (1), 5590. 10.1038/s41598-026-36028-w (2026). Oh, T. K. & Song, I. A. Trained intensivist coverage and survival outcomes in critically ill patients: a nationwide cohort study in South Korea. Ann. Intensive Care . 13 (1), 4. 10.1186/s13613-023-01100-5 (2023). Liu, X. et al. Association of lactate to albumin ratio with the severity and prognosis of patients with sepsis admitted to the emergency intensive care unit: a prospective cohort study. Med. (Baltim). 105 (2), e47064. 10.1097/MD.0000000000047064 (2026). Kokkoris, S. et al. Lactate to albumin ratio and mortality in patients with severe coronavirus disease-2019 admitted to an intensive care unit. J. Clin. Med. 13 (23), 7106. 10.3390/jcm13237106 (2024). Mesinovic, M., Watkinson, P. & Zhu, T. Explainable machine learning for predicting ICU mortality in myocardial infarction patients using pseudo-dynamic data. Sci. Rep. 15 (1), 27887. 10.1038/s41598-025-13299-3 (2025). Du, J. et al. Stress hyperglycemia ratio improves machine learning-based mortality risk prediction in critically ill COVID-19 patients: a multicenter retrospective study. BMC Infect. Dis. 26 (1), 314. 10.1186/s12879-025-12376-2 (2026). Hsu, C. H. et al. Prognostic nomogram for elderly patients with acute respiratory failure receiving invasive mechanical ventilation: a nationwide population-based cohort study in Taiwan. Sci. Rep. 10 (1), 13161. 10.1038/s41598-020-70130-x (2020). Hu, X. et al. Development of a nomogram to predict in-ICU mortality of elderly patients with sepsis-associated liver injury: an analysis of the MIMIC-IV database. Front. Med. (Lausanne) . 12 , 1516853. 10.3389/fmed.2025.1516853 (2025). Li, X. et al. Analysis of prognostic risk factors and risk management measures for patients with ischemic stroke and bloodstream infection based on machine learning. Front. Cell. Infect. Microbiol. 15 , 1715309. 10.3389/fcimb.2025.1715309 (2026). Gan, W., Chen, Z., Tao, Z. & Li, W. Constructing a nomogram model to estimate the risk of ventilator-associated pneumonia for elderly patients in the intensive care unit. Adv. Respir Med. 92 (1), 77–88. 10.3390/arm92010010 (2024). Liu, X. et al. GRACE-ICU: a multimodal nomogram-based approach for illness severity assessment of older adults in the ICU. NPJ Digit. Med. 8 (1), 519. 10.1038/s41746-025-01875-w (2025). Liu, J., Yuan, Y. & Liang, Y. Development and validation of a nomogram for predicting the need for mechanical ventilation in patients undergoing high-flow nasal cannula therapy: a retrospective study based on the MIMIC-IV database. Br. J. Hosp. Med. (Lond) . 87 (2), 50719. 10.31083/BJHM50719 (2026). Kong, G., Lin, K. & Hu, Y. Using machine learning methods to predict in-hospital mortality of sepsis patients in the ICU. BMC Med. Inf. Decis. Mak. 20 (1), 251. 10.1186/s12911-020-01271-2 (2020). Bosch, N. A., Law, A. C., Rucci, J. M., Peterson, D. & Walkey, A. J. Predictive validity of the sequential organ failure assessment score versus claims-based scores among critically ill patients. Ann. Am. Thorac. Soc. 19 (6), 1072–1076. 10.1513/AnnalsATS.202111-1251RL (2022). Additional Declarations No competing interests reported. Supplementary Files SS.docx STROBEChecklist.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 24 Apr, 2026 Editor assigned by journal 22 Apr, 2026 Editor invited by journal 20 Apr, 2026 Submission checks completed at journal 18 Apr, 2026 First submitted to journal 18 Apr, 2026 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-9412320","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":634142319,"identity":"3e86296e-c2ee-4cf5-a0bd-927277f008c4","order_by":0,"name":"Chenxi Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA50lEQVRIiWNgGAWjYBACNvbmgw8+/LCR45c/kPggoaKGsBY+nmPJhjN70owlZzA8Nnhw5hhhLXISOWrSPGyHEzfcYHwm+bCFmQiH8ZxhkODhSUvccLs5rSKxgY2Bv707gYBfeg8YSFjYGM+8cyztRuIOGQaJM2c3ELDlXEKCAU+abN+BHKCWM2wMBhK5BLRI5BgcSGA7zNhwIP9bQWIbM1FaDBsOsB1WnHAjIY2BOC3AQGZsBAVyz4FkiYQzx3gI+kW+vfn47z+gqGRvSPz4o6JGjr+9F78WDMBDmvJRMApGwSgYBVgBAI6rUW3iv76uAAAAAElFTkSuQmCC","orcid":"","institution":"Tianjin Chest Hospital","correspondingAuthor":true,"prefix":"","firstName":"Chenxi","middleName":"","lastName":"Wang","suffix":""},{"id":634142326,"identity":"5cc5ce4e-3663-40eb-9444-18aba6fcd743","order_by":1,"name":"Xiujuan Hu","email":"","orcid":"","institution":"Tianjin Chest Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xiujuan","middleName":"","lastName":"Hu","suffix":""},{"id":634142328,"identity":"573c8e0b-6c95-4b2a-baf8-3404c179fffc","order_by":2,"name":"Jiayu Liu","email":"","orcid":"","institution":"Tianjin Chest Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jiayu","middleName":"","lastName":"Liu","suffix":""},{"id":634142329,"identity":"3b05d296-6abb-4caf-87df-413828e0088b","order_by":3,"name":"Guanhua Li","email":"","orcid":"","institution":"Tianjin Chest Hospital","correspondingAuthor":false,"prefix":"","firstName":"Guanhua","middleName":"","lastName":"Li","suffix":""},{"id":634142331,"identity":"3af18eac-351b-4ad1-9445-232e488b3042","order_by":4,"name":"Li Zhang","email":"","orcid":"","institution":"Tianjin Chest Hospital","correspondingAuthor":false,"prefix":"","firstName":"Li","middleName":"","lastName":"Zhang","suffix":""}],"badges":[],"createdAt":"2026-04-14 08:08:33","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9412320/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9412320/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108837928,"identity":"908c95e6-2d06-411f-a1ab-78cfa1dfe53f","added_by":"auto","created_at":"2026-05-09 00:18:44","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":277411,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlowchart of Study Population Selection.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Figure1Flowchart.png","url":"https://assets-eu.researchsquare.com/files/rs-9412320/v1/450c33aaf6c9e1be6dba0927.png"},{"id":108977524,"identity":"d596da68-1fe5-44ee-b802-6d223b57596b","added_by":"auto","created_at":"2026-05-11 11:31:59","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":555038,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eValidation and Clinical Utility of the Ventilation-CRS Model in the Derivation and External Validation Cohorts.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNotes\u003c/strong\u003e: (A) ROC Curves: Receiver operating characteristic curves demonstrating the discriminative performance of the Ventilation-CRS model in the derivation cohort (MIMIC-IV, AUC = 0.753) and the external validation cohort (eICU, AUC = 0.693).\u003c/p\u003e\n\u003cp\u003e(B) Calibration Plot: Calibration analysis for the external eICU cohort, comparing predicted mortality probabilities with observed mortality (Slope = 0.82, Intercept = 0.22). The dotted line represents perfect calibration.\u003c/p\u003e\n\u003cp\u003e(C) Decision Curve Analysis: DCA illustrating the clinical net benefit of the Ventilation-CRS model compared to \"treat-all\" and \"treat-none\" strategies across various threshold probabilities.\u003c/p\u003e\n\u003cp\u003e(D) Risk Stratification in the External Cohort. Distribution of observed mortality rates across predefined risk categories: low (\u0026lt;5%), medium (5–20%), and high (\u0026gt;20%). Notably, in the high-acuity eICU cohort, no patients (\u003cem\u003en\u003c/em\u003e = 0) met the ultra-low-risk (\u0026lt;5%) threshold, reflecting the inherently high baseline severity of very elderly patients requiring mechanical ventilation and the model's conservative safety profile.\u003c/p\u003e","description":"","filename":"Figure2VentilationCRS.png","url":"https://assets-eu.researchsquare.com/files/rs-9412320/v1/2f9dbb4a5cf164663d2d88a5.png"},{"id":108977232,"identity":"93b7290d-fa87-42c9-9baf-cc08b8efa8d2","added_by":"auto","created_at":"2026-05-11 11:30:59","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":754235,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMachine Learning Pipeline and Performance Evaluation Demonstrating Diminishing Returns of Complexity.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNotes: \u003c/strong\u003e(A) LASSO Coefficient Profiles: The path of coefficients for the 39 candidate clinical features evaluated by the LASSO regression model over varying penalization parameters (Log Alpha). The vertical dotted line indicates the optimal lambda that selected 35 non-zero features.\u003c/p\u003e\n\u003cp\u003e(B) XGBoost Top 15 Predictor Importance: Bar chart illustrating the top 15 most influential predictors in the XGBoost model, ranked by their Gain values. LAR ranks third, demonstrating high prognostic weight even among dozens of variables.\u003c/p\u003e\n\u003cp\u003e(C1) XGBoost ROC Curves: Receiver operating characteristic curves for the 35-variable XGBoost model. The model shows severe overfitting, with a training AUC of 0.980, dropping to an internal validation AUC of 0.795, and an external validation AUC of 0.665.\u003c/p\u003e\n\u003cp\u003e(C2) XGBoost Calibration (External): Calibration plot for the XGBoost model in the external eICU cohort, showing suboptimal agreement between predicted and observed mortality (Slope = 0.55, Intercept = -0.47).\u003c/p\u003e","description":"","filename":"Figure3MachineLearningComplexity.png","url":"https://assets-eu.researchsquare.com/files/rs-9412320/v1/5116cd443d7162f2f0d09692.png"},{"id":109204427,"identity":"a1c78da9-79bf-437c-9b09-4c520a2386a3","added_by":"auto","created_at":"2026-05-13 14:59:41","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":753620,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComprehensive Head-to-Head Comparison of Model Performance and Clinical Utility in the External Cohort.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNotes\u003c/strong\u003e: (A) ROC Curve Comparison: Receiver operating characteristic curves demonstrating the superior discrimination of the Ventilation-CRS model (AUC = 0.693) compared to XGBoost (AUC = 0.665), APS III (AUC = 0.651), and SOFA. Note the failure of the SOFA score (AUC = 0.437).\u003c/p\u003e\n\u003cp\u003e(B) Decision Curve Analysis: DCA comparing the net clinical benefit of the four models. The Ventilation-CRS model provides the highest net benefit across all clinically relevant threshold probabilities.\u003c/p\u003e\n\u003cp\u003e(C) Risk Stratification Comparison. Distribution of observed mortality across predicted low (\u0026lt;5%), medium (5–20%), and high (\u0026gt;20%) risk strata. The red dashed line denotes the cohort’s baseline mortality (32.3%). While traditional scores exhibited catastrophic misclassification (e.g., APS III yielding 50.0% mortality in the low-risk group), the Ventilation-CRS maintained a robust prognostic gradient. Of note, no patients (\u003cem\u003en\u003c/em\u003e = 0) met the ultra-low-risk threshold in this high-severity population, underscoring the model's conservative safety profile.\u003c/p\u003e\n\u003cp\u003e(D) Calibration Curves Comparison: Calibration plots showing the agreement between predicted probabilities and observed mortality. The Ventilation-CRS model aligns most closely with the ideal dotted line, reflecting its superior calibration reliability (Brier Score = 0.2049).\u003c/p\u003e","description":"","filename":"Figure4HeadtoHeadComparisonFinal.png","url":"https://assets-eu.researchsquare.com/files/rs-9412320/v1/2f45a6056df9ac402d8aa88e.png"},{"id":108837932,"identity":"a40bf10a-ee68-433f-80b0-1bfa0f1930ee","added_by":"auto","created_at":"2026-05-09 00:18:44","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":367520,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIndividualized Interpretability of the Ventilation-CRS Model via SHAP Waterfall Plots.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNotes\u003c/strong\u003e: Representative SHAP plots for patients with (A) High predicted risk, (B) Medium predicted risk, and (C) the Lowest predicted risk within the external eICU cohort. Red arrows indicate clinical features that increase the predicted mortality risk, whereas blue arrows denote features that decrease the risk relative to the baseline expected value.\u003c/p\u003e","description":"","filename":"Figure5SHAPAnalysis.png","url":"https://assets-eu.researchsquare.com/files/rs-9412320/v1/e86b406d35efe39f4d4193bb.png"},{"id":108977370,"identity":"eb9ad4b2-5238-4095-adeb-621b02985e5b","added_by":"auto","created_at":"2026-05-11 11:31:33","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":342691,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSubgroup Sensitivity Analysis of the Ventilation-CRS Model.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNotes\u003c/strong\u003e: Forest plot demonstrating the discriminative performance (AUC and 95% CIs) of the Ventilation-CRS model across major clinical comorbidities and advanced age subgroups (75-84 years vs. ≥85 years) in the external cohort. The vertical red dashed line denotes the overall model AUC (0.693), illustrating the model's high stability and generalizability across heterogeneous patient subpopulations.\u003c/p\u003e","description":"","filename":"Figure6SensitivityAnalysis.png","url":"https://assets-eu.researchsquare.com/files/rs-9412320/v1/7b4341a76950e40b57110f27.png"},{"id":108837934,"identity":"7be6c8f7-1de0-4a7d-830b-0b9cf476aff6","added_by":"auto","created_at":"2026-05-09 00:18:44","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":227731,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePrognostic Nomogram for Predicting In-Hospital Mortality in Older Mechanically Ventilated Patients.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNotes:\u003c/strong\u003e The nomogram is used by summing the points assigned to each of the nine clinical variables. Locate the value of each variable on its respective axis, draw a vertical line upward to the \"Points\" axis to determine the score, and then sum these scores to find the \"Total Points.\" The corresponding \"Mortality Risk\" indicates the individualized probability of in-hospital death. The nomogram was derived from the multivariable logistic regression model in the MIMIC-IV cohort\u003c/p\u003e","description":"","filename":"Figure7NomogramStandalone.png","url":"https://assets-eu.researchsquare.com/files/rs-9412320/v1/3eea89c7a85f13ea6e5e795d.png"},{"id":108837935,"identity":"23c1f2c0-cc31-4749-a93c-224ec4cf33a6","added_by":"auto","created_at":"2026-05-09 00:18:44","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":468369,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClinical Decision-Making Framework Based on Ventilation-CRS Risk Stratification.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNotes\u003c/strong\u003e: A hierarchical pathway translating the model's risk scores into bedside management strategies. The framework categorizes patients into three risk strata, each defined by specific [Clinical Triggers] (biological and physiological markers) and matched with targeted [Management] recommendations (clinical actions).\u003c/p\u003e","description":"","filename":"Figure8SCAPStyleRestored.png","url":"https://assets-eu.researchsquare.com/files/rs-9412320/v1/b9858fcf5940f0712d30e096.png"},{"id":109206641,"identity":"c23ec6fd-14e8-4098-88df-8b8b911e00e7","added_by":"auto","created_at":"2026-05-13 15:14:27","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3750853,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9412320/v1/b5b47f90-9dbb-4526-ab7f-2f87fc64cca7.pdf"},{"id":108977364,"identity":"d81a3e3a-46ac-4182-a19c-c266bb6cc9f7","added_by":"auto","created_at":"2026-05-11 11:31:32","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":19617,"visible":true,"origin":"","legend":"","description":"","filename":"SS.docx","url":"https://assets-eu.researchsquare.com/files/rs-9412320/v1/530f1240cfc20fa4c86900d9.docx"},{"id":109067860,"identity":"00cba481-cec9-4afd-b17a-648977cfe26f","added_by":"auto","created_at":"2026-05-12 10:02:01","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":20885,"visible":true,"origin":"","legend":"","description":"","filename":"STROBEChecklist.docx","url":"https://assets-eu.researchsquare.com/files/rs-9412320/v1/760c0dce9b0a20454a8a514a.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"A 9-Variable Lactate-to-Albumin Ratio-Driven Clinical Risk Score for Early Mortality Prediction in Very Elderly Mechanically Ventilated ICU Patients: Multicenter Derivation and External Validation","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eThe global population is rapidly aging, leading to a sharp rise in very elderly patients (aged\u0026thinsp;\u0026ge;\u0026thinsp;75 years) admitted to intensive care units (ICUs). In this population, the initiation of invasive mechanical ventilation (MV) marks a pivotal and high-stakes transition, carrying exceptionally high hospital mortality, accelerated functional decline, and heavy healthcare resource burden. Profound physiological frailty markedly impairs their ability to withstand acute hypoxic stress and systemic inflammation. Therefore, early and precise prognostic stratification based solely on objective data available at ICU admission is essential to guide individualized therapy and facilitate timely goals-of-care discussions \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eContemporary ICU decision-making still depends predominantly on traditional severity scores, including APACHE II, SOFA, and SAPS II. However, these instruments were developed primarily in younger, more heterogeneous ICU cohorts and frequently fail to account for the distinct vulnerabilities of very elderly patients receiving MV\u0026mdash;particularly preexisting nutritional depletion, diminished physiologic reserve, and heightened frailty. Furthermore, their dependence on numerous complex variables severely restricts bedside usability and predictive accuracy in this frail, high-risk subgroup\u003csup\u003e\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eAlthough machine-learning techniques such as extreme gradient boosting and LASSO-regularized models have shown promise in mortality prediction, most high-dimensional algorithms are hindered by the \u0026ldquo;black-box\u0026rdquo; effect, overfitting, limited interpretability, and poor generalizability to the specific subpopulation of very elderly patients on invasive MV. Moreover, few studies have exclusively targeted patients aged\u0026thinsp;\u0026ge;\u0026thinsp;75 years receiving MV or conducted direct head-to-head comparisons of parsimonious, clinically grounded models against complex machine-learning approaches in large, multicenter cohorts.\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eTo address these critical gaps, we developed and externally validated a parsimonious, LAR-driven 9-variable Clinical Risk Score (Ventilation-CRS) specifically for very elderly patients receiving invasive MV in the ICU. The model was constructed using the full MIMIC-IV derivation cohort, guided by clinical experience and LASSO-regularized logistic regression to identify nine objective, routinely collected baseline physiologic and laboratory variables at ICU admission. This streamlined, transparent approach\u0026mdash;externally validated in the independent eICU cohort\u0026mdash;yielded a model that outperforms both conventional severity scores and high-dimensional machine-learning algorithms while preserving complete bedside interpretability and clinical plausibility. The present multicenter study therefore aimed to establish and validate this accurate, explainable tool to support individualized early decision-making in very elderly mechanically ventilated ICU patients \u003csup\u003e\u003cspan additionalcitationids=\"CR9 CR10\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e"},{"header":"2 Methods","content":"\u003ch2\u003e2.1\u0026nbsp; \u0026nbsp; \u0026nbsp; Study Design and Data Sources\u003c/h2\u003e\n\u003cp\u003eThis retrospective, multicenter cohort investigation was conducted using data extracted from two prominent clinical repositories: the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.1) and the eICU Collaborative Research Database (eICU-CRD, version 2.0). The primary model derivation relied on the MIMIC-IV registry, which captures comprehensive electronic health records from 94,458 intensive care unit (ICU) admissions at the Beth Israel Deaconess Medical Center (BIDMC) spanning the years 2008 to 2022. To robustly assess the external validity and spatial generalizability of our predictive framework, the eICU-CRD was utilized as an independent validation cohort, aggregating de-identified patient data across 208 distinct hospitals throughout the United States.\u003c/p\u003e\n\u003cp\u003eBecause both datasets consist of strictly anonymized and encrypted health information, the Institutional Review Boards at both the Massachusetts Institute of Technology (MIT) and BIDMC granted an exemption from the requirement for individual informed consent. The principal investigator (Certification Access No. 65378269) successfully fulfilled the Collaborative Institutional Training Initiative (CITI) requisite to obtain authorized data access. Furthermore, all study protocols strictly adhered to the ethical principles outlined in the Declaration of Helsinki, with all data sourced legitimately through the PhysioNet platform.\u003c/p\u003e\n\u003ch2\u003e2.2 Study Population\u003c/h2\u003e\n\u003cp\u003eTo maintain rigorous methodological symmetry across both the derivation (MIMIC-IV) and external validation (eICU-CRD) cohorts, we applied identical, stringent eligibility criteria. The target population consisted of highly vulnerable geriatric patients, strictly defined as individuals aged 75 years or older, who necessitated invasive MV during their hospitalization.\u003c/p\u003e\n\u003cp\u003eTo ensure the integrity of baseline physiological assessments and to avoid statistical confounding from recurrent critical illness, only the primary (index) ICU admission for each patient was incorporated into the analysis; all subsequent readmissions were systematically excluded. Furthermore, we excluded patients with an ICU length of stay of less than 24 hours. This temporal threshold was deliberately established to guarantee sufficient data granularity for baseline physiological profiling and to exclude early transfers or immediate, irreversible mortalities where early prognostic modeling would yield limited actionable utility. Finally, records exhibiting substantial data missingness (\u0026gt;20%) in essential demographic or initial vital sign parameters were removed from the analytical cohort. The detailed, step-by-step patient attrition and final cohort assembly are meticulously delineated in the study flowchart (\u003cstrong\u003eFigure 1\u003c/strong\u003e).\u003c/p\u003e\n\u003ch2\u003e2.3\u0026nbsp; \u0026nbsp; \u0026nbsp; Data Extraction and Missing Value Handling\u003c/h2\u003e\n\u003cp\u003eClinical parameters, vital signs, and laboratory diagnostics were acquired through Structured Query Language (SQL) scripts. To exclusively capture the patient\u0026apos;s initial physiological derangement and maintain the model\u0026apos;s immediate early-warning utility, all continuous or repeated variables were restricted to the first available (baseline) values recorded upon ICU admission. This approach ensures that the Ventilation-CRS provides actionable prognostic information at the earliest clinical juncture, without the requirement for a 24-hour observation window. The Lactate-to-Albumin Ratio (LAR), a pivotal biomarker of coupled microcirculatory and nutritional deficit, was mathematically derived by dividing the baseline blood lactate concentration (mmol/L) by the serum albumin level (g/dL).\u003c/p\u003e\n\u003cp\u003eGiven the retrospective nature of real-world electronic health records, data missingness is inevitable. As a general methodological heuristic, candidate variables exhibiting a missingness rate exceeding 30% across the cohorts were considered for exclusion to preserve data robustness. However, a strict, a priori clinical exception was implemented. Variables deemed fundamentally indispensable for the assessment of acute respiratory failure and tissue hypoperfusion-specifically Lactate (missing in 42.6% of the eICU cohort), Albumin (missing in 35.4% of the MIMIC cohort), SaO\u003csub\u003e2\u003c/sub\u003e (34.8% in MIMIC), and Neutrophil counts (34.2% in MIMIC)-were purposefully retained.\u003c/p\u003e\n\u003cp\u003eExcluding these cores homeostatic markers based solely on an arbitrary statistical threshold would severely compromise the biological validity and prognostic depth of the study. Assuming the data were missing at random (MAR)-a standard assumption in critical care registries where test frequency fluctuates based on clinical acuity-we employed the Multiple Imputation by Chained Equations (MICE) algorithm to robustly handle the missing values. This sophisticated predictive mean matching approach generated five parallel imputed datasets, minimizing potential bias while securely preserving the physiological architecture of our predictive framework. The robustness of including these high-missingness variables was subsequently confirmed through complete-case sensitivity analysis (see Section 3.6). Model robustness was further evaluated by complete-case sensitivity analysis restricted to patients with complete data for all nine variables (\u003cem\u003en\u003c/em\u003e = 5,941; \u003cstrong\u003eSupplementary Table S1)\u003c/strong\u003e.\u003c/p\u003e\n\u003ch2\u003e2.4\u0026nbsp; \u0026nbsp; \u0026nbsp; Statistical Analysis and Model Evaluation\u003c/h2\u003e\n\u003cp\u003eBaseline continuous variables, exhibiting predominantly non-normal distributions, were reported as medians accompanied by interquartile ranges (IQR) and compared utilizing the Mann-Whitney U test. Categorical variables were expressed as absolute frequencies with percentages and assessed via the Chi-square test.\u003c/p\u003e\n\u003cp\u003eThe parsimonious Ventilation-CRS was formulated using multivariable logistic regression within the MIMIC-IV derivation cohort, with variable inclusion strictly guided by a priori clinical reasoning and baseline statistical significance. To rigorously construct a high-dimensional comparative baseline, an advanced machine learning pipeline was simultaneously developed. The derivation data was partitioned (70:30 ratio) to train an Extreme Gradient Boosting (XGBoost) classifier. To systematically mitigate overfitting, 40 candidate features were initially regularized using the Least Absolute Shrinkage and Selection Operator (LASSO) with an L1 penalty, successfully isolating a subset of 35 optimal predictors for XGBoost training. To strictly circumvent potential collinearity between arterial oxygen saturation (SaO\u003csub\u003e2\u003c/sub\u003e) and peripheral capillary oxygen saturation (SpO₂), which represent identical physiological constructs, SpO₂ was pre-emptively excluded prior to LASSO regularization.\u003c/p\u003e\n\u003cp\u003eThe risk-stratification thresholds (\u0026lt;5% for low-risk, 5\u0026ndash;20% for medium-risk, and \u0026gt;20% for high-risk) were pre-defined based on clinical utility: the low-risk threshold targets candidates for early liberation from MV, while the high-risk threshold acts as a clinical trigger for escalated multidisciplinary review and proactive Goals of Care (GOC) discussions.\u003c/p\u003e\n\u003cp\u003eThe eICU-CRD dataset served as the independent external validation cohort. Discriminative capacity across all models (Ventilation-CRS, XGBoost, SOFA, and APS III) was quantified using the Area Under the Receiver Operating Characteristic curve (AUC), with formal pairwise statistical comparisons executed via the DeLong method. Calibration fidelity was evaluated visually through calibration curves and quantitatively using the Brier score. To evaluate tangible clinical utility, Decision Curve Analysis (DCA) was deployed to estimate the net clinical benefit across a continuum of threshold probabilities. Furthermore, the Integrated Discrimination Improvement (IDI) index was computed to assess the exact reclassification superiority of the Ventilation-CRS over legacy severity scores.\u003c/p\u003e\n\u003cp\u003eTo decrypt the \u0026quot;black-box\u0026quot; nature of predictive modeling and provide bedside transparency, SHapley Additive exPlanations (SHAP) values were extracted to visualize the individualized pathophysiological drivers behind specific mortality predictions. Finally, rigorous subgroup sensitivity analyses (presented via forest plots) were conducted across distinct chronic comorbidity profiles and advanced age strata to confirm spatial and demographic generalizability.\u003c/p\u003e\n\u003cp\u003eTo robustly validate the internal stability of the logistic regression-based Ventilation-CRS and account for potential model overfitting, internal validation was performed using 1,000 bootstrap resamples on the derivation cohort to compute the optimism-corrected AUC and its 95% confidence intervals (CI).\u003c/p\u003e\n\u003cp\u003eAll analytical procedures and visualizations were executed utilizing Python (version 3.9) and R software (version 4.2.2). A two-tailed \u003cem\u003ep\u003c/em\u003e-value of less than 0.05 was considered to denote statistical significance.\u003c/p\u003e"},{"header":"3 Results","content":"\u003ch2\u003e3.1\u0026nbsp; \u0026nbsp; \u0026nbsp; Participant Selection and Flow\u003c/h2\u003e\n\u003cp\u003eThe detailed flowing screening process for study participants from both distinct multicenter databases is horizontally delineated in \u003cstrong\u003eFigure 1\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eIn the derivation cohort from the MIMIC-IV database (v3.1), we initiated the selection with a comprehensive pool of 94,458 distinct patients. The first round of pragmatic screening simultaneously applied the predefined inclusion and exclusion criteria, focusing strictly on geriatric patients (Age\u0026nbsp;\u0026ge;\u0026nbsp;75 years), their first documented ICU admission, and a required ICU length of stay of at least 24 hours. This initial wave of screening refined the potential population to 16,329 individuals. Subsequently, a critical second-round precise refinement was performed to include only those who underwent invasive mechanical ventilation during their hospitalization. This robust two-step selection process culminated in a final derivation cohort of 5,941 patients.\u003c/p\u003e\n\u003cp\u003eTo maintain perfect methodological symmetry for external validation, an identical hierarchical screening pathway was implemented within the eICU-CRD database (v2.0), which initially comprised 200,859 patients. Following the same first-round parameters for age, first ICU admission, and stay duration, a stratified population of 28,810 patients was obtained. The identical second-round requirement for invasive mechanical ventilation therapy ultimately defined a highly validated and stable external cohort of 1,293 patients.\u003c/p\u003e\n\u003ch2\u003e3.2\u0026nbsp; \u0026nbsp; \u0026nbsp; Baseline Characteristics of the Study Populations\u003c/h2\u003e\n\u003cp\u003eA total of 7,234 mechanically ventilated patients were included: 5,941 in the MIMIC-IV derivation cohort (overall mortality, 22.3%) and 1,293 in the eICU external validation cohort (overall mortality, 32.3%). Baseline demographic and clinical characteristics are summarized in \u003cstrong\u003eTable 1\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eAcross both cohorts, non-survivors were significantly older and presented with worse admission physiology, characterized by profound hypoxia (lower PaO\u003csub\u003e2\u003c/sub\u003e), exacerbated acidemia (lower pH), and pronounced multi-organ dysfunction (all \u003cem\u003ep\u003c/em\u003e \u0026lt;0.05). Notably, biomarkers reflecting the intersection of acute hypoxic stress and baseline nutritional depletion were highly discriminative: non-survivors exhibited significantly higher lactate and lower albumin levels, yielding a markedly elevated \u0026nbsp;LAR in both the derivation (0.7 vs 0.4, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001) and validation (1.0 vs 0.6, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001) cohorts.\u003c/p\u003e\n\u003cp\u003eSeverity of illness, as measured by the APS III score, was consistently higher among non-survivors. However, the atypically low median SOFA score observed in the MIMIC-IV cohort (median, 2.0) highlights the inherent limitations of sub-score documentation (e.g., Glasgow Coma Scale) in retrospective electronic health records, reinforcing the necessity of laboratory-driven prognostic models. The overlapping median SOFA scores (median 2.0 for both groups) further illustrate its limited early discriminative power in this specific population.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1. Baseline Demographic and Clinical Characteristics of Mechanically Ventilated Patients Stratified by Hospital Mortality.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" rowspan=\"2\" style=\"width: 14px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVariable\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 36px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMIMIC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" rowspan=\"2\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003ep\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"3\" style=\"width: 34px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eeICU\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" rowspan=\"2\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003ep\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSurvived (\u003cem\u003en\u003c/em\u003e = 4615)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNon-survived (\u003cem\u003en\u003c/em\u003e = 1326)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 15px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSurvived (\u003cem\u003en\u003c/em\u003e = 875)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNon-survived (\u003cem\u003en\u003c/em\u003e = 418)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 30px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDemographics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eAge, years\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e81.1 [77.8-85.1]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e83.0 [79.0-87.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e80.0 [77.0-84.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e82.0 [78.0-86.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eMale, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e2579 (55.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e685 (51.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.007\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e435 (49.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e224 (53.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.228\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eWeight, kg\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e81.0 [68.9-93.9]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e76.2 [64.0-90.7]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e74.8 [63.4-88.4]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e74.6 [62.0-87.6]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.499\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" colspan=\"3\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eComorbidities, n (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eCHF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1663 (36.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e592 (44.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e138 (15.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e75 (17.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.366\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eCOPD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1273 (27.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e389 (29.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.223\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e106 (12.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e49 (11.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.911\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eDiabetes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1491 (32.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e432 (32.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.878\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e136 (15.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e71 (17.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.561\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eLiver disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e44 (1.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e42 (3.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e8 (0.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e10 (2.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.062\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eRenal disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1242 (26.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e443 (33.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e114 (13.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e54 (12.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" colspan=\"3\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVital Signs\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eSBP, mmHg\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e119.0 [104.0-136.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e121.0 [103.0-140.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e128.0 [108.0-149.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e120.0 [101.0-141.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eDBP, mmHg\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e59.0 [51.0-70.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e62.5 [52.0-76.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e67.0 [56.0-79.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e64.0 [52.0-78.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.092\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eHR, bpm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e80.0 [72.0-90.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e88.0 [74.0-102.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e85.0 [73.0-99.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e86.5 [73.0-108.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.031\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eRR, bpm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e16.0 [14.0-20.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e20.0 [16.0-24.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e18.0 [15.0-21.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e20.0 [16.0-24.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eSpO\u003csub\u003e2\u003c/sub\u003e, %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e100.0 [97.0-100.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e98.0 [95.0-100.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e99.0 [96.0-100.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e98.0 [95.0-100.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.003\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" colspan=\"3\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRespiratory \u0026amp; Gas Balance\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003ePaO\u003csub\u003e2\u003c/sub\u003e, mmHg\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e243.0 [100.0-382.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e99.5 [61.0-183.8]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e118.7 [78.0-218.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e107.0 [70.0-180.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eFiO\u003csub\u003e2\u003c/sub\u003e, %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e54.0 [45.6-61.1]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e51.4 [43.8-62.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e60.0 [40.0-100.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e60.0 [40.0-100.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.026\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003ePaCO\u003csub\u003e2\u003c/sub\u003e, mmHg\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e40.0 [36.0-45.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e41.0 [35.0-49.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.355\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e39.0 [33.4-46.1]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e40.2 [33.0-48.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.477\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003epH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e7.39 [7.34-7.44]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e7.36 [7.27-7.42]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e7.39 [7.32-7.44]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e7.36 [7.27-7.42]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eBE, mmol/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.0 [-2.0-2.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e-2.0 [-6.0-1.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e0.1 [-3.7-3.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e-1.8 [-6.0-2.6]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eSaO2, %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e97.0 [95.0\u0026ndash;98.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e95.0 [83.0\u0026ndash;98.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e98.0 [95.3\u0026ndash;99.9]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e98.0 [94.0\u0026ndash;99.7]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.048\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" colspan=\"3\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLaboratory Findings\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eLactate, mmol/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1.4 [1.1-2.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e1.9 [1.3-3.2]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e1.7 [1.1-2.7]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e2.3 [1.4-4.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eAlbumin, g/dL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e3.4 [2.8-3.8]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e2.9 [2.5-3.3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e3.0 [2.5-3.4]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e2.6 [2.2-3.1]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eLAR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.4 [0.3-0.6]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e0.7 [0.4-1.2]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e0.6 [0.4-1.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1.0 [0.5-1.7]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eLY, \u0026times;10⁹/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1.3 [0.8-1.9]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e0.9 [0.5-1.4]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e1.0 [0.6-1.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1.0 [0.6-1.4]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.147\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eWBC , \u0026times;10⁹/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e9.6 [7.1-13.3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e11.7 [8.3-16.4]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e10.6 [7.8-14.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e11.9 [8.6-17.8]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003ePlt, \u0026times;10⁹/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e182.0 [137.0-237.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e188.0 [136.0-259.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.008\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e189.0 [143.0-242.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e182.5 [140.8-246.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.931\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eHb, g/dL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e10.9 [9.2-12.4]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e10.7 [9.4-12.2]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.317\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e11.2 [9.5-12.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e10.8 [9.3-12.2]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eCr, mg/dL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1.0 [0.8-1.4]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e1.2 [0.9-1.8]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e1.1 [0.8-1.6]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1.3 [0.9-1.9]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eBUN, mg/dL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e21.0 [16.0-31.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e28.0 [19.0-45.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e23.0 [16.2-35.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e28.0 [19.0-44.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eGlu, mg/dL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e126.0 [104.0-158.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e140.0 [112.0-183.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e130.0 [104.0-169.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e143.0 [108.0-195.2]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eNa mmol/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e139.0 [137.0-142.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e139.0 [136.0-142.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.073\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e139.0 [137.0-142.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e138.0 [135.0-141.8]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eK, mmol/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e4.1 [3.8-4.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e4.2 [3.8-4.7]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e4.0 [3.7-4.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e4.2 [3.7-4.6]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.012\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eCl, mmol/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e105.0 [101.0-109.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e104.0 [100.0-108.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e105.0 [101.0-109.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e105.0 [100.0-109.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.15\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eHCO\u003csub\u003e3\u003c/sub\u003e, mmol/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e24.0 [21.0-26.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e22.0 [19.0-25.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e24.0 [21.0-27.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e23.0 [19.4-27.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eAG, mmol/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e13.0 [11.0-16.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e15.0 [13.0-18.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e10.0 [7.0-12.4]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e10.0 [8.0-14.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.015\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eTBil, mg/dL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.6 [0.4-0.9]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e0.7 [0.4-1.1]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e0.6 [0.4-0.9]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.7 [0.5-1.2]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eALT, U/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e20.0 [14.0-34.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e26.0 [15.0-63.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e22.0 [14.0-38.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e26.0 [16.0-55.8]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eAST, U/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e28.0 [20.0-47.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e43.0 [25.0-101.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e26.0 [19.0-49.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e39.0 [23.2-90.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eALP, U/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e72.0 [55.0-97.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e82.0 [61.0-118.5]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e72.0 [56.0-93.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e80.0 [64.0-105.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eINR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1.2 [1.1-1.4]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e1.3 [1.1-1.6]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e1.2 [1.1-1.4]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1.3 [1.1-1.7]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 30px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eClinical Scores\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eSOFA score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e2.0 [1.0-4.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e2.0 [0.0-4.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e0.018\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e9.0 [7.0-12.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e10.0 [8.0-13.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 14px;\"\u003e\n \u003cp\u003eAPS III score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e42.0 [32.0-56.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 19px;\"\u003e\n \u003cp\u003e60.0 [45.0-79.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\" style=\"width: 17px;\"\u003e\n \u003cp\u003e58.0 [43.0-79.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 16px;\"\u003e\n \u003cp\u003e74.0 [55.8-97.0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 7px;\"\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: 94px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 115px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 138px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 109px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 110px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eNotes:\u003c/strong\u003e Data are presented as median [interquartile range] for continuous variables and No. (%) for categorical variables. Differences between survival groups were compared using the Mann-Whitney \u003cem\u003eU\u003c/em\u003e test for continuous variables and the Chi-square test for categorical variables.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003e3.3\u0026nbsp; \u0026nbsp; \u0026nbsp; Construction and Validation of the Ventilation-CRS Model\u003c/h2\u003e\n\u003cp\u003eThe Ventilation-CRS model was constructed using the initial values of nine pre-selected, objective clinical variables recorded upon ICU admission. By prioritizing the first available data points following admission, the model ensures maximum utility for early bedside triage. The selection of these nine variables was primarily guided by a priori clinical reasoning focusing on respiratory, metabolic, and systemic stability. This was further corroborated by preliminary univariable screenings in the MIMIC-IV cohort, ensuring each included marker was both biologically plausible and statistically significant. Multivariable logistic regression analysis (\u003cstrong\u003eTable 2\u003c/strong\u003e) identified the LAR \u0026nbsp;as the most potent independent predictor of in-hospital mortality (OR, 1.816; 95% CI, 1.606\u0026ndash;2.053; \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001). Other parameters, including age, glucose, neutrophil count, AST, INR, and respiratory rate, were also identified as significant risk factors. Conversely, baseline oxygen saturation (SaO\u003csub\u003e2\u003c/sub\u003e) demonstrated a significant protective effect (OR, 0.977; 95% CI, 0.971\u0026ndash;0.983; \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001).\u003c/p\u003e\n\u003cp\u003eThe predictive performance and clinical utility of the Ventilation-CRS model are comprehensively illustrated in \u003cstrong\u003eFigure 2\u003c/strong\u003e. The model demonstrated satisfactory discriminative ability, yielding an apparent area under the curve (AUC) of 0.753 in the MIMIC-IV derivation cohort. Crucially, to rigorously account for potential model overfitting and ensure internal stability, we performed an internal validation using 1,000 bootstrap resamples. This stringent analysis revealed a minimal optimism of 0.0019, yielding an optimism-corrected AUC of 0.752 (95% CI, 0.738\u0026ndash;0.765). This indicates highly robust internal consistency with negligible overfitting. Furthermore, the model maintained a stable AUC of 0.693 in the completely independent and highly heterogeneous eICU external validation cohort (\u003cstrong\u003eFigure 2A\u003c/strong\u003e). Calibration analysis in the external cohort revealed acceptable agreement between the predicted probabilities and observed mortality, with a calibration slope of 0.82 and an intercept of 0.22 (\u003cstrong\u003eFigure 2B\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eDecision curve analysis (DCA) confirmed that the Ventilation-CRS model provided a consistent net clinical benefit across a wide and clinically relevant range of threshold probabilities compared to the default \u0026quot;treat-all\u0026quot; or \u0026quot;treat-none\u0026quot; strategies (\u003cstrong\u003eFigure 2C\u003c/strong\u003e). Furthermore, the model effectively stratified patients in the external cohort into low (\u0026lt; 5%), medium (5%-20%), and high (\u0026gt; 20%) risk categories. The observed mortality rates exhibited a clear stepwise increase: 0.0% for the low-risk, 20.2% for the medium-risk, and 45.0% for the high-risk groups (\u003cstrong\u003eFigure 2D\u003c/strong\u003e). Notably, the absence of mortality (0.0%) in the strictly defined low-risk group accurately reflects the inherently high baseline acuity of patients requiring mechanical ventilation, further validating the model\u0026apos;s reliability in safely identifying true high-risk populations for actionable bedside interventions.\u003c/p\u003e\n\u003cp\u003eAlthough Na exhibited marginal statistical significance in the multivariable model (\u003cem\u003ep\u003c/em\u003e = 0.093), it was deliberately retained in the final Ventilation-CRS because sodium balance is a fundamental homeostatic marker in the geriatric population, where electrolyte derangements are strongly associated with underlying frailty and acute mortality. Excluding this core physiological parameter based strictly on a \u003cem\u003ep\u003c/em\u003e-value threshold would diminish the model\u0026apos;s clinical grounding and face validity. In the complete-case sensitivity analysis (\u003cem\u003en\u003c/em\u003e = 5,941), all coefficients showed minimal changes (\u0026lt;6%) compared with the primary multiple imputation model (\u003cstrong\u003eSupplementary Table S1\u003c/strong\u003e). Notably, the LAR ratio remained the most potent predictor with an identical odds ratio of 1.816.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2. Multivariable Logistic Regression Analysis of the 9-Variable Ventilation-CRS Model.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVariable\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCoef\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOR (95% CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003ep\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003eAge, y\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.0523\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 33px;\"\u003e\n \u003cp\u003e1.054 (1.040-1.067)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003eLAR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.5965\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 33px;\"\u003e\n \u003cp\u003e1.816 (1.606-2.053)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003eSaO\u003csub\u003e2\u003c/sub\u003e, %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e-0.023\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 33px;\"\u003e\n \u003cp\u003e0.977 (0.971-0.983)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003eGlu, mg/dL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.0019\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 33px;\"\u003e\n \u003cp\u003e1.002 (1.001-1.003)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003eNa, mmol/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e-0.0124\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 33px;\"\u003e\n \u003cp\u003e0.988 (0.974-1.002)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.093\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003eNE,\u0026times;10\u003csup\u003e9\u003c/sup\u003e/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.0389\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 33px;\"\u003e\n \u003cp\u003e1.040 (1.027-1.052)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003eAST, U/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.0006\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 33px;\"\u003e\n \u003cp\u003e1.001 (1.000-1.001)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003eINR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.1203\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 33px;\"\u003e\n \u003cp\u003e1.128 (1.019-1.249)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003eRR, bpm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.0628\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 33px;\"\u003e\n \u003cp\u003e1.065 (1.053-1.077)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch2\u003e3.4\u0026nbsp; \u0026nbsp; \u0026nbsp; Diminishing Returns of Machine Learning Complexity\u003c/h2\u003e\n\u003cp\u003eTo rigorously evaluate whether incorporating a larger number of variables and utilizing complex, non-linear algorithms could yield superior prognostic performance, we developed an advanced machine learning pipeline. The MIMIC-IV derivation cohort was randomly partitioned into a training set (\u003cem\u003en\u003c/em\u003e = 4158, 70%) and an internal validation set (\u003cem\u003en\u003c/em\u003e = 1783, 30%), while the eICU dataset served as the external validation cohort (\u003cem\u003eN\u003c/em\u003e = 1293).\u003c/p\u003e\n\u003cp\u003eInitially, 39 candidate clinical features (excluding peripheral oxygen saturation to strictly prevent multicollinearity with arterial oxygen saturation) were introduced into a Least Absolute Shrinkage and Selection Operator (LASSO) regression model. The L1 penalty effectively regularized the model, eliminating redundant variables and identifying a subset of 35 optimal predictors (\u003cstrong\u003eFigure 3A)\u003c/strong\u003e. Subsequently, an extreme gradient boosting (XGBoost) model was trained using these 35 features. Variable importance analysis, quantified by the Gain metric, revealed the top contributors to the XGBoost model (\u003cstrong\u003eTable 3, Figure 3B\u003c/strong\u003e). Notably, the LAR maintained a top-tier prognostic weight, ranking as the third most important predictor (Gain = 10.32) across the entire high-dimensional feature space, surpassed only by PaO\u003csub\u003e2\u003c/sub\u003e and respiratory rate, further confirming its robust biological and prognostic relevance in mechanically ventilated patients.\u003c/p\u003e\n\u003cp\u003eDespite the sophisticated algorithmic architecture and the inclusion of 35 features, the machine learning model exhibited classic signs of severe overfitting and poor external generalizability. While the XGBoost model achieved an exceptionally high AUC of 0.980 in the training set and 0.795 in the internal validation set, its discriminative performance sharply declined to an AUC of 0.665 in the eICU external validation cohort (\u003cstrong\u003eFigure 3C1\u003c/strong\u003e). Furthermore, external calibration was suboptimal, demonstrating notable deviation from the ideal line with a calibration slope of 0.55 and an intercept of -0.47 (\u003cstrong\u003eFigure 3C2\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eCompared to the 9-variable clinical experience model (Ventilation-CRS, external AUC = 0.693) developed in Section 3.2, the heavily parameterized 35-variable XGBoost model completely failed to improve predictive accuracy in unseen data. These findings strongly demonstrate a \u0026quot;diminishing return of complexity\u0026quot; in this clinical context, highlighting that a parsimonious model strictly grounded in fundamental, objective physiological parameters offers superior robustness, stability, and clinical utility across heterogeneous ICU settings.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3. XGBoost Predictor Importance Ranking Based on Gain Values.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePredictor\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGain Value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eImportance Rank\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003ePaO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e18.92723\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eRR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e10.39721\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eLAR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e10.32179\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eLY\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e7.963913\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eALT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e7.109571\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eBUN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e6.049428\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eAST\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e6.035837\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eBE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e5.792546\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eFiO2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e5.764657\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eNE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e5.756567\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003epH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e5.635104\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eINR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e5.547956\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e5.301125\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eWBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e5.271259\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 29px;\"\u003e\n \u003cp\u003eHR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 34px;\"\u003e\n \u003cp\u003e5.054357\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 35px;\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eNotes:\u003c/strong\u003e The table presents the predictors selected by the LASSO regression TOP 15 (35 in total), ranked by their relative contribution (Gain value) to the XGBoost model\u0026apos;s predictive accuracy.\u003c/p\u003e\n\u003ch2\u003e3.5\u0026nbsp; \u0026nbsp; \u0026nbsp; Model Performance, Head-to-Head Comparison, and Clinical Utility\u003c/h2\u003e\n\u003cp\u003eTo thoroughly evaluate the robustness and clinical value of the Ventilation-CRS model, we conducted a head-to-head comparison against the high-dimensional XGBoost model and two universally established clinical scoring systems (SOFA and APS III) using the independent eICU validation cohort. The comprehensive performance metrics are summarized in \u003cstrong\u003eTable 4\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eIn terms of discrimination, the 9-variable Ventilation-CRS model demonstrated the highest predictive accuracy, achieving an AUC of 0.693 (95% CI, 0.659-0.721). It significantly outperformed both the traditional SOFA score (AUC = 0.437; \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001) and the APS III score (AUC = 0.651; \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001) (\u003cstrong\u003eFigure 4A\u003c/strong\u003e). Notably, the SOFA score exhibited an AUC below 0.50, indicating a complete failure to discriminate mortality risk within this specific mechanically ventilated geriatric cohort. Furthermore, the parsimonious Ventilation-CRS model surpassed the complex 35-variable XGBoost model (AUC = 0.665) in the unseen external data, strongly reinforcing our previous observation regarding the diminishing returns of machine learning complexity and highlighting the superior generalizability of our clinical model.\u003c/p\u003e\n\u003cp\u003eRegarding calibration, the Ventilation-CRS model achieved the lowest Brier score (0.2049) among all evaluated models. The calibration curve (\u003cstrong\u003eFigure 4D\u003c/strong\u003e) visually confirmed that the Ventilation-CRS probabilities aligned most closely with the ideal reference line, whereas the XGBoost model and traditional scores exhibited varying degrees of overestimation or underestimation. The Integrated Discrimination Improvement (IDI) analysis corroborated these findings, demonstrating that the Ventilation-CRS model provided a massive reclassification improvement over the SOFA score (IDI = -0.1035 in favor of CRS). While XGBoost showed nominal fluctuations in reclassification (IDI = 0.0234), its overall clinical reliability was compromised by inferior calibration and discrimination.\u003c/p\u003e\n\u003cp\u003eClinical utility was further evaluated via Decision Curve Analysis (DCA) (\u003cstrong\u003eFigure 4B\u003c/strong\u003e). The Ventilation-CRS model consistently provided the highest net clinical benefit across the entire clinically relevant threshold probability range (0.05\u0026ndash;0.50). Conversely, the SOFA score offered no clinical benefit, plunging below the \u0026quot;treat-none\u0026quot; reference line almost immediately.\u003c/p\u003e\n\u003cp\u003eFinally, risk stratification accuracy was critically assessed against the overall baseline mortality (32.3%) of the external cohort (\u003cstrong\u003eFigure 4C\u003c/strong\u003e). The Ventilation-CRS model effectively separated patients into logical trajectories, consistently distinguishing those above and below the baseline average. Notably, in the high-acuity external eICU cohort (median SOFA = 9.0), no patients (n = 0) met the ultra-low-risk threshold (\u0026lt;5%), reflecting the model\u0026apos;s conservative and safe orientation when faced with extremely ill geriatric populations. In contrast, the Ventilation-CRS demonstrated a clear and reliable prognostic gradient in the medium-risk (5%\u0026ndash;20%, observed mortality 20.2%) and high-risk (\u0026gt;20%, observed mortality 45.0%) strata.\u003c/p\u003e\n\u003cp\u003eIn stark contrast, traditional scoring systems demonstrated dangerous clinical misclassification. For instance, patients classified as \u0026quot;low risk\u0026quot; (\u0026lt; 5%) by APS III astonishingly experienced a 50.0% actual mortality rate, significantly exceeding the cohort average. Similarly, those deemed \u0026quot;high risk\u0026quot; (\u0026gt; 20%) by SOFA exhibited a mortality rate of only 29.0%, which was even lower than the baseline mortality of the entire cohort. These results collectively highlight the Ventilation-CRS model as a highly accurate, reliable, and clinically safe decision-support tool that avoids the perilous underestimation of mortality risk in vulnerable older adults.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4. Head-to-Head Comparison of Model Performance in the External Validation Cohort (eICU).\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 22px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAUC (95% CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 15px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBrier Score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 15px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eIDI vs CRS\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003ep\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e-value (DeLong)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 22px;\"\u003e\n \u003cp\u003eVentilation-CRS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003e0.693 (0.659-0.721)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.2049\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 15px;\"\u003e\n \u003cp\u003eRef\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003eRef\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 22px;\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003e0.665 (0.634-0.693)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.2144\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.0234\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003ens\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 22px;\"\u003e\n \u003cp\u003eSOFA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003e0.437 (0.403-0.471)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.2336\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 15px;\"\u003e\n \u003cp\u003e-0.1035\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" style=\"width: 22px;\"\u003e\n \u003cp\u003eAPS III\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 28px;\"\u003e\n \u003cp\u003e0.651 (0.616-0.690)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.2085\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.0048\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eNotes:\u003c/strong\u003e Ref denotes the reference model for head-to-head comparisons; ns (not significant) indicates a \u003cem\u003ep\u003c/em\u003e-value \u0026gt; 0.05 for the DeLong test compared to the Ventilation-CRS model. IDI values were calculated relative to the Ventilation-CRS model; a negative value for SOFA indicates significantly inferior reclassification performance compared to our parsimonious model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNote regarding risk stratification:\u003c/strong\u003e The 0.0% observed mortality in the low-risk group (\u0026lt;5%) was recorded in the MIMIC-IV derivation cohort; in the higher-acuity external eICU cohort, no patients met this threshold (\u003cem\u003en\u003c/em\u003e = 0).\u003c/p\u003e\n\u003ch2\u003e3.6\u0026nbsp; \u0026nbsp; \u0026nbsp; Clinical Significance, Interpretability, and Sensitivity Analysis\u003c/h2\u003e\n\u003cp\u003eTo bridge the gap between population-level statistical metrics and individualized bedside decision-making, we employed SHapley Additive exPlanations (SHAP) to interpret the Ventilation-CRS model. This method visualizes the positive or negative contribution of each physiological parameter to the final mortality risk for individual patients. \u003cstrong\u003eFigure 5\u003c/strong\u003e displays three representative SHAP waterfall plots for patients stratified into high (~50%), medium (~15%), and the lowest predicted risk categories in the external validation cohort. The base value represents the cohort\u0026apos;s average predicted risk, with red bars pushing the individual risk higher and blue bars pulling it lower. For instance, in the high-risk patient (\u003cstrong\u003eFigure 5A\u003c/strong\u003e), an elevated LAR \u0026nbsp;and advanced age served as the primary drivers of increased mortality risk. Conversely, in the lowest-risk patient (\u003cstrong\u003eFigure 5C\u003c/strong\u003e), robust physiological reserves-characterized by a healthy respiratory rate, stable SaO\u003csub\u003e2\u003c/sub\u003e, and normal sodium levels-acted as strong protective factors, effectively reducing the predicted mortality to 5.2%. This individualized transparency allows clinicians to identify the precise physiological derangements driving a specific patient\u0026apos;s deterioration.\u003c/p\u003e\n\u003cp\u003eIn the complete-case sensitivity analysis (\u003cem\u003en\u003c/em\u003e = 5,941), all coefficients showed minimal changes (\u0026lt;6%) compared with the primary multiple imputation model (\u003cstrong\u003eSupplementary Table S1\u003c/strong\u003e). Notably, the LAR ratio remained the most potent predictor with an identical odds ratio of 1.816. This consistency across both imputed and non-imputed datasets confirms the robust structural integrity of the Ventilation-CRS.\u003c/p\u003e\n\u003cp\u003eTo rigorously evaluate the generalizability of the Ventilation-CRS model across highly heterogeneous ICU populations, we conducted a comprehensive sensitivity analysis assessing the model\u0026apos;s discriminative performance (AUC) across major chronic comorbidities and advanced age subgroups. As illustrated in the forest plot (\u003cstrong\u003eFigure 6\u003c/strong\u003e), the predictive accuracy of the Ventilation-CRS model remained remarkably robust and consistent. Regardless of whether patients presented with concomitant congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD), diabetes, or renal disease, the subgroup AUCs tightly clustered around the overall cohort AUC line (0.693, red dashed line). Importantly, the model maintained its robust performance across different geriatric strata, achieving an AUC of 0.694 in patients aged 75-84 years and remaining stable at 0.677 in the extremely elderly group (\u0026ge;85 years). These findings confirm that the Ventilation-CRS model accurately captures fundamental acute physiological stress signals, demonstrating reliable prognostic utility independent of baseline chronic disease phenotypes or extreme age.\u003c/p\u003e\n\u003cp\u003eFurthermore, given the high missing rates of baseline lactate and albumin, we performed a stringent complete-case sensitivity analysis (\u003cem\u003en\u003c/em\u003e = 5,941) to verify that multiple imputation did not introduce significant bias. All coefficients showed minimal changes (\u0026lt;6%), and the LAR \u0026nbsp;remained the strongest predictor with an identical odds ratio of 1.816 (\u003cstrong\u003eSupplementary Table S1\u003c/strong\u003e). The overall model performance remained robust, confirming that the missing-at-random (MAR) assumption was valid and that the high prognostic weight of LAR is biologically authentic rather than a statistical artifact of imputation.\u003c/p\u003e\n\u003ch2\u003e3.7\u0026nbsp; \u0026nbsp; \u0026nbsp; Clinical Translation: Bedside Nomogram and Risk-Based Decision Framework\u003c/h2\u003e\n\u003cp\u003eTo facilitate the seamless integration of the Ventilation-CRS model into daily clinical practice, we developed two highly actionable bedside tools: a prognostic nomogram and a risk-stratified decision-making framework. The 9-variable nomogram (\u003cstrong\u003eFigure 7\u003c/strong\u003e) provides an intuitive visual interface for individualized risk assessment, enabling clinicians to derive a precise probability of in-hospital death based on initial baseline values recorded at the onset of mechanical ventilation. This bedside-ready tool eliminates the need for complex computational devices, providing immediate prognostic clarity at the earliest clinical juncture. By locating a patient\u0026rsquo;s specific values for the nine objective predictors-including Age, SaO\u003csub\u003e2\u003c/sub\u003e, respiratory rate, Sodium, Glucose, AST, INR, Neutrophil count, and LAR-and projecting them onto the standardized points scale, a cumulative \u0026quot;Total Points\u0026quot; score can be calculated and mapped directly to the final mortality risk axis.\u003c/p\u003e\n\u003cp\u003eBeyond mere risk quantification, we further proposed a structured clinical decision-making framework to translate these statistical predictions into specific management strategies (\u003cstrong\u003eFigure 8\u003c/strong\u003e). Based on the model\u0026rsquo;s validated thresholds, patients are triaged into three distinct risk trajectories. For the Low-Risk group (predicted mortality\u0026nbsp;\u0026le;\u0026nbsp;5%, observed 0.0%), clinical triggers such as stable SaO\u003csub\u003e2\u003c/sub\u003e and lower respiratory rates indicate a high potential for successful liberation from the ventilator; therefore, management should prioritize early spontaneous breathing trials and extubation. Conversely, the High-Risk group (predicted mortality \u0026gt; 20%, observed 45.0%) represents the most critical cohort, where triggers such as significantly elevated LAR, advanced age (\u0026ge; 85 years), or severe coagulopathy necessitate aggressive care escalation. In this high-acuity setting, the framework serves as a vital clinical trigger for multidisciplinary review and early proactive family communication regarding the goals of care. By linking objective physiological markers to targeted management pathways, this integrated system ensures that the Ventilation-CRS model functions as a comprehensive decision-support tool for optimizing outcomes in older critically ill patients.\u003c/p\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eA notable finding in this study was the clear diminishing returns of increasing model complexity. Despite achieving an exceptionally high AUC of 0.980 in the training set, the 35-variable XGBoost model exhibited substantial performance degradation in the independent external eICU cohort (AUC 0.665). In contrast, our parsimonious 9-variable Ventilation-CRS maintained superior discriminative ability (AUC 0.693) with markedly better calibration and clinical net benefit. These results illustrate that, in the heterogeneous population of very elderly patients receiving invasive mechanical ventilation, overly complex machine-learning algorithms often capture database-specific noise rather than generalizable physiology. This supports a return to the principle of parsimony in clinical prediction modeling\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e,\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eHead-to-head comparison further revealed the limited discriminative performance of the SOFA score in this specific cohort (AUC 0.437). This finding is consistent with known challenges in applying traditional organ failure-based scores to very elderly patients, where factors such as sedation, incomplete Glasgow Coma Scale documentation, and baseline frailty may substantially underestimate risk. In contrast, the Ventilation-CRS provided a clearer prognostic gradient across risk strata and avoided dangerous underestimation of mortality, particularly in high-acuity populations.\u003c/p\u003e \u003cp\u003eTo facilitate rapid translation into clinical practice, we developed a bedside nomogram and a structured risk-stratified decision-making framework. The nomogram enables clinicians to calculate individualized in-hospital mortality probability using only nine objective admission variables, without requiring complex computation. The accompanying framework links predefined risk categories (low, medium, and high) to specific management recommendations, such as prioritizing early spontaneous breathing trials and extubation in low-risk patients, and triggering multidisciplinary review with proactive goals-of-care discussions in high-risk patients. By connecting readily available physiologic markers to actionable pathways, these tools offer immediate bedside utility for early triage, resource allocation, and ethical decision-making in older critically ill patients.\u003c/p\u003e \u003cp\u003eSeveral limitations should be acknowledged. First, as a retrospective study using large multicenter databases, missing data were handled with multiple imputation; however, complete-case sensitivity analyses confirmed the robustness of the key findings, including the dominant role of the lactate-to-albumin ratio (LAR). Second, both derivation and validation cohorts were derived from Western healthcare systems, which may limit generalizability to other regions and ethnic groups; external validation in Asian or other diverse populations would be valuable\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. Third, the model relies solely on baseline admission data and does not incorporate dynamic changes during the ICU stay. Future prospective, multicenter studies are warranted to evaluate the real-world impact of implementing the Ventilation-CRS on clinical decision-making, resource utilization, and patient-centered outcomes\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn conclusion, the Ventilation-CRS represents a practical step toward simplicity in geriatric critical care \u003csup\u003e\u003cspan additionalcitationids=\"CR19\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e.By integrating only nine objective, routinely collected markers at ICU admission\u0026mdash;with the lactate-to-albumin ratio as the strongest driver\u0026mdash;this parsimonious and transparent model safely outperforms both high-dimensional machine-learning approaches and conventional scoring systems such as SOFA\u003csup\u003e\u003cspan additionalcitationids=\"CR22\" citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. Our findings demonstrate that clinically grounded, parsimonious tools leveraging initial baseline data can provide superior bedside safety and decision support for the most vulnerable elderly mechanically ventilated population. The accompanying nomogram and risk-stratified framework further empower clinicians to optimize early triage, resource allocation, and goals-of-care discussions without the need for complex computation\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e,\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e"},{"header":"5 Conclusion","content":"\u003cp\u003eThe Ventilation-CRS represents a paradigm shift toward simplicity in geriatric critical care. By utilizing only nine objective markers routinely collected at the time of ICU admission, this parsimonious model offers a high-performance, transparent, and stable prognostic tool that safely outperforms both high-dimensional machine-learning algorithms and traditional scoring systems. Our findings demonstrate that clinically-grounded models leveraging initial baseline data can provide superior bedside safety and decision support for the most vulnerable elderly MV population. By linking objective physiological markers to targeted management pathways, this framework empowers clinicians to optimize early triage, resource allocation, and ethical goals-of-care discussions without the need for complex computation.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eData Availability Statement\u003c/p\u003e\n\u003cp\u003eThe data that support the findings of this study are available from the PhysioNet repository (MIMIC‑IV and eICU‑CRD databases). These data are available under a license and are not publicly available. Data can be obtained from the corresponding author upon reasonable request, subject to approval from PhysioNet, completion of the required CITI training, and signing of the data use agreement.\u003c/p\u003e\n\u003cp\u003eConflict of Interest\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e\n\u003cp\u003eEthics statement\u003c/p\u003e\n\u003cp\u003eThe creation and public release of this database were approved by the Institutional Review Boards (IRBs) of the Massachusetts Institute of Technology (Cambridge, MA) and Beth Israel Deaconess Medical Center (Boston, MA). As the MIMIC-IV v3.1 database consists entirely of de-identified, retrospective, publicly available clinical data, this study was determined to be exempt from both full ethical review and the requirement for informed consent.\u003c/p\u003e\n\u003cp\u003eAuthor Contributions\u003c/p\u003e\n\u003cp\u003eC.W. (Chenxi Wang): Conceptualization, Methodology, Software, Data curation, Formal analysis, Visualization, Writing-original draft. X.H. (Xiujuan Hu) and J.L. (Jiayu Liu): Investigation, Resources. G.L. (Guanhua Li): Project administration, Supervision. L.Z. (Li Zhang): Conceptualization, Supervision, Validation, Writing-review \u0026amp; editing. All authors critically reviewed the manuscript and approved the final submitted version.\u003c/p\u003e\n\u003cp\u003eFunding\u003c/p\u003e\n\u003cp\u003eTianjin Key Medical Discipline Construction Project, No. TJYXZDXK-3-032C.\u003c/p\u003e\n\u003cp\u003eAcknowledgments\u003c/p\u003e\n\u003cp\u003eThe authors gratefully acknowledge the Computational Physiology Laboratory (LCP) at the Massachusetts Institute of Technology (MIT) for its sustained stewardship and maintenance of the MIMIC-IV v3.1 database.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eCao, M. et al. Population-wise incidence and outcomes of patients requiring invasive and non-invasive mechanical ventilation in China: a nationwide retrospective analysis by age, sex, and comorbidity. \u003cem\u003eAnn. Intensive Care\u003c/em\u003e. \u003cb\u003e15\u003c/b\u003e (1), 120. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13613-025-01537-w\u003c/span\u003e\u003cspan address=\"10.1186/s13613-025-01537-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMahran, G. S. K., El-aty, N. S. A., Abd-Elhamed, A. G., Ali, M. M. A. \u0026amp; Kasasbeh, M. A. M. Comparison of baseline characteristics and outcomes in ICU patients with endotracheal tubes versus tracheostomies: a prospective observational study. \u003cem\u003eNurs. Crit. Care\u003c/em\u003e. \u003cb\u003e31\u003c/b\u003e (2), e70352. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/nicc.70352\u003c/span\u003e\u003cspan address=\"10.1111/nicc.70352\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNguyen, B. H. et al. Long-term mortality in patients with chronic obstructive pulmonary disease requiring acute non-invasive ventilation with and without obstructive sleep apnoea. \u003cem\u003eBMJ Open. Respir Res.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e (1), e002496. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/bmjresp-2024-002496\u003c/span\u003e\u003cspan address=\"10.1136/bmjresp-2024-002496\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTekin, B. et al. Comparison of scoring systems: SOFA, APACHE-II, LODS, MODS, and SAPS-II in critically ill elderly sepsis patients. \u003cem\u003eJ. Infect. Dev. Ctries.\u003c/em\u003e \u003cb\u003e18\u003c/b\u003e (1), 122\u0026ndash;130. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3855/jidc.18526\u003c/span\u003e\u003cspan address=\"10.3855/jidc.18526\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou, J., Luo, X. Y., Li, H. L., Shi, G. Z. \u0026amp; Chen, G. Q. Comparison of the predictive value of APACHE II, SOFA, SAPS II, GCS and GCS-P scores for in-hospital mortality in critically ill patients after craniotomy: a retrospective cohort study in a Chinese tertiary hospital. \u003cem\u003eBMJ Open.\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e (2), e101867. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/bmjopen-2025-101867\u003c/span\u003e\u003cspan address=\"10.1136/bmjopen-2025-101867\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, H. et al. A machine learning-based prediction of hospital mortality in mechanically ventilated ICU patients. \u003cem\u003ePLoS One\u003c/em\u003e. \u003cb\u003e19\u003c/b\u003e (9), e0309383. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0309383\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0309383\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCao, Z. et al. Association between the platelet-albumin-bilirubin score and all-cause mortality in ICU-admitted heart failure patients: a retrospective cohort analysis and machine learning-based prognostic modeling. \u003cem\u003eFront. Cardiovasc. Med.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 1622554. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fcvm.2025.1622554\u003c/span\u003e\u003cspan address=\"10.3389/fcvm.2025.1622554\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWei, J. et al. Nomogram predictive model for in-hospital mortality risk in elderly ICU patients with urosepsis. \u003cem\u003eBMC Infect. Dis.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e (1), 442. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12879-024-09319-8\u003c/span\u003e\u003cspan address=\"10.1186/s12879-024-09319-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZeng, Y., Tan, N., He, X., Peng, S. \u0026amp; Qiu, E. Development of a nomogram to predict in-hospital mortality of trauma patients in the ICU: an analysis of the MIMIC-IV database. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e (1), 6802. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-026-38251-x\u003c/span\u003e\u003cspan address=\"10.1038/s41598-026-38251-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu, Y., Zhong, Y., Fan, F., Shen, Y. \u0026amp; Jiang, C. Development and validation of a bedside nomogram for predicting 30-day mortality in patients with type 1 hepatorenal syndrome: a retrospective cohort study. \u003cem\u003eMed. (Baltim).\u003c/em\u003e \u003cb\u003e105\u003c/b\u003e (8), e47640. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/MD.0000000000047640\u003c/span\u003e\u003cspan address=\"10.1097/MD.0000000000047640\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShao, D., He, L., Zhong, D., Li, K. \u0026amp; Zhang, X. Association between the geriatric nutritional risk index and 28-day mortality in critically ill sepsis-associated pneumonia patients: retrospective study based on two cohorts. \u003cem\u003eFront. Nutr.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 1698973. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fnut.2025.1698973\u003c/span\u003e\u003cspan address=\"10.3389/fnut.2025.1698973\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMu, Y., Lou, F., Feng, G. \u0026amp; Huang, M. Lactate to albumin ratio predicts one year mortality in ICU patients receiving invasive mechanical ventilation. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e (1), 5590. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-026-36028-w\u003c/span\u003e\u003cspan address=\"10.1038/s41598-026-36028-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOh, T. K. \u0026amp; Song, I. A. Trained intensivist coverage and survival outcomes in critically ill patients: a nationwide cohort study in South Korea. \u003cem\u003eAnn. Intensive Care\u003c/em\u003e. \u003cb\u003e13\u003c/b\u003e (1), 4. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13613-023-01100-5\u003c/span\u003e\u003cspan address=\"10.1186/s13613-023-01100-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, X. et al. Association of lactate to albumin ratio with the severity and prognosis of patients with sepsis admitted to the emergency intensive care unit: a prospective cohort study. \u003cem\u003eMed. (Baltim).\u003c/em\u003e \u003cb\u003e105\u003c/b\u003e (2), e47064. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/MD.0000000000047064\u003c/span\u003e\u003cspan address=\"10.1097/MD.0000000000047064\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKokkoris, S. et al. Lactate to albumin ratio and mortality in patients with severe coronavirus disease-2019 admitted to an intensive care unit. \u003cem\u003eJ. Clin. Med.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e (23), 7106. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/jcm13237106\u003c/span\u003e\u003cspan address=\"10.3390/jcm13237106\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMesinovic, M., Watkinson, P. \u0026amp; Zhu, T. Explainable machine learning for predicting ICU mortality in myocardial infarction patients using pseudo-dynamic data. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e15\u003c/b\u003e (1), 27887. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-025-13299-3\u003c/span\u003e\u003cspan address=\"10.1038/s41598-025-13299-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDu, J. et al. Stress hyperglycemia ratio improves machine learning-based mortality risk prediction in critically ill COVID-19 patients: a multicenter retrospective study. \u003cem\u003eBMC Infect. Dis.\u003c/em\u003e \u003cb\u003e26\u003c/b\u003e (1), 314. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12879-025-12376-2\u003c/span\u003e\u003cspan address=\"10.1186/s12879-025-12376-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHsu, C. H. et al. Prognostic nomogram for elderly patients with acute respiratory failure receiving invasive mechanical ventilation: a nationwide population-based cohort study in Taiwan. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e (1), 13161. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-020-70130-x\u003c/span\u003e\u003cspan address=\"10.1038/s41598-020-70130-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu, X. et al. Development of a nomogram to predict in-ICU mortality of elderly patients with sepsis-associated liver injury: an analysis of the MIMIC-IV database. \u003cem\u003eFront. Med. (Lausanne)\u003c/em\u003e. \u003cb\u003e12\u003c/b\u003e, 1516853. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fmed.2025.1516853\u003c/span\u003e\u003cspan address=\"10.3389/fmed.2025.1516853\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, X. et al. Analysis of prognostic risk factors and risk management measures for patients with ischemic stroke and bloodstream infection based on machine learning. \u003cem\u003eFront. Cell. Infect. Microbiol.\u003c/em\u003e \u003cb\u003e15\u003c/b\u003e, 1715309. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fcimb.2025.1715309\u003c/span\u003e\u003cspan address=\"10.3389/fcimb.2025.1715309\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGan, W., Chen, Z., Tao, Z. \u0026amp; Li, W. Constructing a nomogram model to estimate the risk of ventilator-associated pneumonia for elderly patients in the intensive care unit. \u003cem\u003eAdv. Respir Med.\u003c/em\u003e \u003cb\u003e92\u003c/b\u003e (1), 77\u0026ndash;88. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/arm92010010\u003c/span\u003e\u003cspan address=\"10.3390/arm92010010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, X. et al. GRACE-ICU: a multimodal nomogram-based approach for illness severity assessment of older adults in the ICU. \u003cem\u003eNPJ Digit. Med.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e (1), 519. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41746-025-01875-w\u003c/span\u003e\u003cspan address=\"10.1038/s41746-025-01875-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, J., Yuan, Y. \u0026amp; Liang, Y. Development and validation of a nomogram for predicting the need for mechanical ventilation in patients undergoing high-flow nasal cannula therapy: a retrospective study based on the MIMIC-IV database. \u003cem\u003eBr. J. Hosp. Med. (Lond)\u003c/em\u003e. \u003cb\u003e87\u003c/b\u003e (2), 50719. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.31083/BJHM50719\u003c/span\u003e\u003cspan address=\"10.31083/BJHM50719\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2026).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKong, G., Lin, K. \u0026amp; Hu, Y. Using machine learning methods to predict in-hospital mortality of sepsis patients in the ICU. \u003cem\u003eBMC Med. Inf. Decis. Mak.\u003c/em\u003e \u003cb\u003e20\u003c/b\u003e (1), 251. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12911-020-01271-2\u003c/span\u003e\u003cspan address=\"10.1186/s12911-020-01271-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBosch, N. A., Law, A. C., Rucci, J. M., Peterson, D. \u0026amp; Walkey, A. J. Predictive validity of the sequential organ failure assessment score versus claims-based scores among critically ill patients. \u003cem\u003eAnn. Am. Thorac. Soc.\u003c/em\u003e \u003cb\u003e19\u003c/b\u003e (6), 1072\u0026ndash;1076. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1513/AnnalsATS.202111-1251RL\u003c/span\u003e\u003cspan address=\"10.1513/AnnalsATS.202111-1251RL\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Mechanical Ventilation, Very Elderly, Mortality Prediction, Lactate-to-Albumin Ratio, Clinical Risk Score","lastPublishedDoi":"10.21203/rs.3.rs-9412320/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9412320/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eVery elderly patients (\u0026ge;\u0026thinsp;75 years) receiving invasive mechanical ventilation in the ICU face exceptionally high mortality, yet traditional severity scores such as SOFA frequently demonstrate limited discriminative performance in this frail population. We aimed to develop and externally validate a parsimonious, clinically actionable 9-variable Clinical Risk Score (Ventilation-CRS) driven by the lactate-to-albumin ratio (LAR), using only objective baseline data available at ICU admission.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThis retrospective multicenter study included 5,941 patients from the MIMIC-IV database (derivation) and 1,293 patients from the eICU-CRD database (external validation). A 9-variable logistic regression model was constructed and directly compared with a 35-variable XGBoost model and conventional scores (SOFA and APS III) through discrimination, calibration, reclassification (IDI), and decision curve analysis.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eIn the external validation cohort, the Ventilation-CRS achieved an AUC of 0.693 (95% CI 0.659\u0026ndash;0.721), outperforming the 35-variable XGBoost model (AUC 0.665) and markedly surpassing the SOFA score (AUC 0.437). The model demonstrated acceptable calibration, the highest net clinical benefit on decision curve analysis, and effective risk stratification (observed mortality: 0.0% low-risk, 20.2% medium-risk, 45.0% high-risk in the derivation cohort). A bedside nomogram and risk-stratified decision framework were developed to facilitate early triage and goals-of-care discussions.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThis parsimonious Ventilation-CRS, anchored by admission LAR and readily available physiologic variables, provides a transparent, generalizable, and clinically practical tool that outperforms both complex machine learning approaches and traditional scoring systems. It offers immediate bedside utility for mortality risk assessment and resource allocation in very elderly mechanically ventilated patients.\u003c/p\u003e","manuscriptTitle":"A 9-Variable Lactate-to-Albumin Ratio-Driven Clinical Risk Score for Early Mortality Prediction in Very Elderly Mechanically Ventilated ICU Patients: Multicenter Derivation and External Validation","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-09 00:18:39","doi":"10.21203/rs.3.rs-9412320/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2026-04-24T12:25:00+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-04-23T02:40:44+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-04-20T13:24:46+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-18T19:37:05+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-04-18T19:32:20+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"044ec77e-e425-4847-ab32-9959dc71a076","owner":[],"postedDate":"May 9th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":67485102,"name":"Health sciences/Diseases"},{"id":67485103,"name":"Health sciences/Health care"},{"id":67485104,"name":"Health sciences/Medical research"},{"id":67485105,"name":"Health sciences/Risk factors"}],"tags":[],"updatedAt":"2026-05-09T00:18:39+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-09 00:18:39","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9412320","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9412320","identity":"rs-9412320","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}