Construction and validation of a prognostic model for in-hospital multiple organ dysfunction syndrome in ICU patients with respiratory failure based on ultrasound and laboratory parameters Authors: Qian Wang, Qun Wu, Fengying Jie, Wen Sun, Fengtao Wang, Shuping Deng, Jiao Han, Fang Dong Published: 2026 ABSTRACT BackgroundSevere respiratory failure (SRF) is a major cause of intensive care unit (ICU) admission, while multiple organ dysfunction syndrome (MODS) serves as a critical contributor to poor prognosis. This research examined the risk factors for in-hospital MODS in individuals with SRF based on ultrasound and laboratory parameters. A predictive model was constructed via the least absolute shrinkage and selection operator (LASSO)-Cox regression and subsequently validated.MethodsData were collected from individuals with SRF admitted to the ICU of Wuhan Third Hospital between January 1, 2024, and May 31, 2025. LASSO regression was utilized to identify the risk factors for MODS. A Cox proportional hazards model was then established based on the selected variables by LASSO regression. The predictive performance of the models was appraised via the concordance index (C-index). Risk stratification was conducted via X-tile software, and the performance of the stratification system was assessed with the Kaplan-Meier method.ResultsIn total, 246 individuals with SRF were enrolled and randomly stratified into a training cohort (n=173) and a validation cohort (n=73) in a 7:3 ratio. Variables selected by LASSO regression, including pH, HCO3−, respiratory rate, activated partial thromboplastin time, procalcitonin, and inferior vena cava, were included in the Cox model. The model yielded a C-index of 0.793 in the training cohort and 0.748 in the validation cohort. In the training cohort, the area under the curve (AUC) of the predictive model was 0.824 [95% confidence interval (CI): 0.726–0.923] for the 15-day outcome and 0.809 (95% CI: 0.676–0.942) for the 28-day outcome. In the validation cohort, the corresponding AUCs were 0.721 (95% CI: 0.501–0.942) and 0.737 (95% CI: 0.503–0.971). Based on the constructed model, risk stratification was carried out via X-tile software. According to the optimal cutoff value of 118.4, individuals were categorized into high- and low-risk groups. Statistical analysis demonstrated that the risk of MODS was significantly elevated in the high-risk group in comparison to the low-risk group (P<0.05).ConclusionsThis research constructed and validated a nomogram based on LASSO-Cox regression to predict the MODS risk among individuals with SRF. This nomogram may assist clinicians in identifying individuals at high risk of MODS and tailoring individualized follow-up and treatment strategies based on risk prediction, thereby improving patients’ long-term outcomes. Highlight box Key findings • This research constructed and validated a nomogram based on least absolute shrinkage and selection operator (LASSO)-Cox regression to predict the multiple organ dysfunction syndrome (MODS) risk among individuals with severe respiratory failure (SRF). What is known and what is new? • Traditional critical care scoring systems including APACHE II and SOFA mainly predict all-cause mortality, and fail to timely warn of MODS in patients with SRF. Existing MODS prediction models mostly adopt single laboratory indicators and lack combined application of ultrasound parameters. Related models also show limited predictive performance compared with emerging analytical methods. • We are the first to combine ultrasound and laboratory indicators. Based on ultrasound and laboratory indicators, the present study constructed and validated a prognostic model for in-hospital MODS in intensive care unit patients with SRF. The developed model allows for risk stratification of SRF patients at 15 and 28 days after disease onset based on a nomogram-based scoring system. What is the implication, and what should change now? • This nomogram may assist clinicians in identifying individuals at high risk of MODS and tailoring individualized follow-up and treatment strategies based on risk prediction, thereby improving patients’ long-term outcomes. Introduction Severe respiratory failure (SRF) remains a major clinical challenge in critical care medicine and continues to impose a growing epidemiological burden worldwide. Globally, SRF accounts for approximately 15–30% of intensive care unit (ICU) admissions, making it one of the major causes of ICU hospitalization. As a prototypical form of SRF, acute respiratory distress syndrome (ARDS) has an annual incidence of about 10–86 cases per 100,000 population, with considerable variability across regions and populations (1). Notably, the coronavirus disease 2019 (COVID-19) pandemic has further exacerbated the burden of SRF and placed unprecedented strain on ICU resources globally (2). Underlying conditions, including pancreatitis, major trauma, sepsis, chronic obstructive pulmonary disease (COPD), and pneumonia, serve as the primary triggers of SRF. Smoking history, advanced age, comorbid chronic diseases [including renal or cardiovascular disease (CVD)], and immunosuppression are major factors that notably elevate the risk of SRF (3). Multiple organ dysfunction syndrome (MODS) occurs with high frequency in individuals with SRF and is a major determinant of poor outcomes (4). Respiratory failure can result in systemic inflammatory response syndrome, severe hypercapnia, acidosis, and hypoxemia, all of which may directly or indirectly impair the function of other organs, for example, through ventilator-induced lung injury. Organ failure often follows a sequential pattern, typically beginning with the lungs and subsequently involving the circulatory system (shock), kidneys [acute kidney injury (AKI)], liver, coagulation system, and central nervous system (5,6). The presence of MODS markedly elevates the risk of death among SRF patients. Existing research reveals a positive relationship between the number of failed organs and mortality. The death rate is approximately 30% in cases of single-organ failure, rises to 50–60% with failure of two organs, and reaches 70–95% when three or more organs are involved (7). Early identification of individuals at risk of developing MODS is therefore critical. Timely interventions to control the primary disease, optimization of ventilatory strategies to minimize secondary injury, and implementation of organ-specific protective approaches are crucial for improving both survival and quality of life in this population. To date, a few risk scoring systems incorporating demographic and clinical variables have been proposed (8,9). However, no studies have combined ultrasound parameters with laboratory indicators to predict the risk of MODS. Most existing studies rely mainly on static analysis of laboratory parameters for screening (10,11), and the least absolute shrinkage and selection operator (LASSO) regression has been rarely applied in SRF studies. Moreover, few studies have incorporated critical care ultrasound parameters into risk prediction models, which could enhance model precision. Furthermore, the LASSO-Cox method has rarely been applied for timely risk prediction. LASSO regression was combined with Cox regression in the study. The former effectively selected variables, and the latter facilitated modeling and visualization, allowing for direct interpretation. Based on ultrasound and laboratory indicators, the present study constructed and validated a prognostic model for in-hospital MODS in ICU patients with SRF. The developed model allowed for risk stratification of SRF patients at 15 and 28 days after disease onset based on a nomogram-based scoring system. This tool can improve individualized prediction of MODS risk, facilitate the identification of individuals at high risk, and offer insights into developing personalized diagnostic and therapeutic strategies, ultimately improving patient prognosis. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1-0291/rc). Methods Study population A retrospective analysis was carried out on individuals diagnosed with SRF between January 2024 and May 2025 at Wuhan Third Hospital. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethics committee of Wuhan Third Hospital (No. KY2025-062). Due to the retrospective design of the study, written informed consent was waived for all patients. The inclusion criteria for individuals with respiratory failure encompassed: (I) admission to the ICU; (II) diagnosis of SRF (12): (i) acute onset defined as acute hypoxemic respiratory failure occurring within two weeks of an inciting event; (ii) oxygenation criteria: a ratio of arterial oxygen partial pressure to fractional inspired oxygen (PaO2/FiO2) ≤300 mmHg, or a ratio of peripheral capillary oxygen saturation to FiO2 (SpO2/FiO2) ≤315 mmHg (SpO2 ≤97%); (iii) requirement of respiratory support: positive end-expiratory pressure or continuous positive airway pressure ≥5 cmH2O, or high-flow nasal cannula oxygen therapy ≥30 L/min; (iv) imaging findings: bilateral pulmonary infiltrates on chest X-ray or CT scan that cannot be fully explained by pleural effusion, lobar atelectasis, or pulmonary nodules. Lung ultrasound (LUS) was also considered valid for the imaging assessment; (v) origin of pulmonary edema: respiratory failure not fully explained by heart failure or fluid overload. In the absence of obvious risk factors, an objective assessment (such as cardiac ultrasound) was required to rule out hydrostatic pulmonary edema. Exclusion criteria included: (I) age ≤18 years; (II) absence of ultrasound examination; (III) presence of cerebrovascular disease, hematologic disorders, pregnancy, or use of immunosuppressive agents. Clinical data collection and definitions This study retrospectively collected potential predictor variables for model development, including demographic characteristics, date of onset, treatment timing, length of hospital stay, laboratory test results, ultrasound parameters, complications, and other relevant clinical indicators. Demographic data encompassed sex, body mass index, age, and history of comorbidities. Laboratory parameters included: (I) arterial blood gas analysis: pH, PO2, PCO2, bicarbonate (HCO3–), base excess, lactate, and PaO2/FiO2 ratio; (II) complete blood count: hemoglobin, platelet count, white blood cell count, and lymphocyte count; (III) liver function: lactate dehydrogenase, aspartate aminotransferase, total bilirubin, and alanine aminotransferase; (IV) renal function: creatinine and blood urea nitrogen; (V) electrolytes: sodium (Na+), potassium (K+), phosphate (P); (VI) inflammatory markers: C-reactive protein and procalcitonin (PCT); (VII) cardiac function marker: brain natriuretic peptide; (VIII) ultrasound indicators: left ventricular ejection fraction, inferior vena cava (IVC), LUS score, and left ventricular outflow tract velocity-time integral; (IX) complications included ventilator-associated pneumonia, AKI, MODS, and gastrointestinal bleeding. To ensure the integrity of the original data, variables with missing values exceeding 20% were ruled out. Demographic and clinical data were primarily obtained through standardized questionnaires completed by physicians on the first day of treatment. Laboratory data were based on test results from the day of SRF diagnosis. In this study, MODS was defined as the reversible dysfunction of two or more organs, occurring simultaneously or sequentially, on the basis of primary disease caused by various acute pathogenic insults. If uncontrolled, MODS may progress to multiple organ failure (MOF) (13). The time from disease onset to MODS, as well as the time to clinical improvement or discharge due to withdrawal of treatment, was recorded. The primary outcome was the risk of developing MODS within 15 and 28 days after disease onset. Determination of sample size The sample size was determined using the event per variable (EPV) principle developed using the Cox regression model (14). The model was planned to include 6 predictive factors and required a minimum EPV of 10 to ensure model stability. The minimum number of required outcome events (hospitalized MODS cases) = final number of predictive factors * minimum EPV = 6*10=60 cases. Based on clinical practice, the expected incidence rate of MODS in ICU SRF patients is 30% to 40% (4). The total sample size was calculated using the following formula: total sample size = minimum number of events/expected outcome event occurrence rate. Therefore, the final sample size was estimated to be 150–200 cases, which can ensure the reliability and stability of the prognostic model, as well as the statistical power of subsequent validation and risk stratification analysis. Statistical analysis R software was utilized to carry out statistical analyses. Mean ± standard deviation was used to represent normally distributed continuous variables, and between-group comparisons were made through t-tests. The median (P25, P75) was used to represent non-normally distributed variables, and between-group comparisons were conducted using the Mann-Whitney U test. Frequencies and percentages were used to represent categorical variables, and Fisher’s exact test or Pearson’s chi-square test was employed to compare differences between groups. The included individuals were randomly allocated to a training cohort and a validation cohort in a 7:3 ratio. Variables in the training cohort were selected via the LASSO regression, with ten-fold cross-validation applied to determine the predictor variables. LASSO regression used L1 regularization to automatically select the most predictive variables by penalizing weak predictors. This method avoided overfitting and produced a parsimonious model. The optimal λ was chosen via cross-validation to balance prediction accuracy and model simplicity. The variables with non-zero coefficients at the optimal λ were selected as the final predictors used to construct the prognostic score (15). The variables selected via LASSO regression were enrolled in Cox regression analysis to identify independent risk factors for MODS at 15 and 28 days after disease onset in SRF patients. Statistically significant variables were used to construct a nomogram. For each variable, corresponding point values were calculated, and a total risk score was assigned to each patient. The total risk score was used to generate Kaplan-Meier (K-M) curves to estimate the MODS risk within 15 and 28 days. The discriminative ability of the model in the training and validation cohorts was appraised via the receiver operating characteristic (ROC) curve and concordance index (C-index). Calibration curves were used to examine the agreement between predicted probabilities and reported outcomes. Subsequently, the training cohort was used to validate the nomogram. X-tile software (version 3.6.1, Yale University) was employed to determine the optimal cutoff value for nomogram-based risk scores in predicting in-hospital MODS among SRF patients. Individuals were stratified into high- and low-risk groups accordingly. In both the training and validation cohorts, the K-M method and log-rank test were used to compare the cumulative incidence of MODS and intergroup differences between the risk groups. All tests were two-sided, and statistical significance was defined as P<0.05. Results General information According to the predefined eligibility criteria, 246 individuals with SRF were enrolled. They were randomly categorized into a training cohort (n=173) and a validation cohort (n=73). The median age of the overall cohort was 74.5 years [interquartile range (IQR), 64.0–84.75 years], suggesting that the study population predominantly consisted of older individuals. Males accounted for 68.7% (169/246) of the cohort. Hypertension was the most prevalent comorbidity (61.4%), followed by diabetes mellitus (36.2%). The prevalence of CVD (17.1%), renal disease (13.8%), COPD (8.5%), and malignancy (4.9%) was relatively low. All patients were critically ill, presenting with severe hypoxemia [PaO2/FiO2 (IQR, 142.25–240.00)], impaired consciousness [Glasgow Coma Scale score (IQR, 5.00–10.00)], and a high Sequential Organ Failure Assessment (SOFA) score (median =9). LUS and respiratory support assessments demonstrated a median LUS score of 18 (IQR, 11.25–24), suggesting significant pulmonary exudation or consolidation. Invasive mechanical ventilation was required in 71.1% of patients (175/246). No significant differences were found in baseline characteristics between the training and validation cohorts (P>0.05) (Table 1). Table 1Comparison of clinical data between the training cohort and validation cohortVariableOverall patients (N=246)Training set (N=173)Validation set (N=73)P valueAge (years)74.50 [64.00, 84.75]75.00 [66.00, 85.00]70.00 [63.00, 83.00]0.16Male169 (68.7)120 (69.4)49 (67.1)0.85BMI (kg/m2)22.35 [20.76, 24.22]22.46 [20.76, 24.22]22.31 [20.05, 24.22]0.75Previous history Smoking history78 (31.7)51 (29.5)27 (37.0)0.31 Drinking history39 (15.9)26 (15.0)13 (17.8)0.72 Hypertension151 (61.4)108 (62.4)43 (58.9)0.71 Diabetes89 (36.2)62 (35.8)27 (37.0)0.98 Cardiovascular diseases42 (17.1)34 (19.7)8 (11.0)0.14 COPD21 (8.5)12 (6.9)9 (12.3)0.26 Kidney diseases34 (13.8)24 (13.9)10 (13.7)>0.99 Tumor12 (4.9)10 (5.8)2 (2.7)0.49Blood gas analysis PH7.37 [7.26, 7.43]7.36 [7.26, 7.43]7.38 [7.28, 7.43]0.82 PO2 (mmHg)100.00 [77.25, 162.00]98.00 [77.00, 154.00]102.00 [78.00, 172.00]0.35 PCO2 (mmHg)36.90 [28.73, 48.68]36.90 [28.60, 49.30]36.90 [29.10, 46.60]0.64 HCO3− (mmol/L)20.90 [16.75, 25.08]21.30 [16.90, 25.30]20.70 [16.70, 24.70]0.74 BE−3.75 [−8.38, 0.58]−3.40 [−8.30, 0.90]−4.00 [−8.40, 0.20]0.65 Lac (mmol/L)1.75 [1.00, 3.90]1.60 [1.00, 3.70]2.00 [1.10, 4.90]0.23 PaO2/FiO2200.00 [142.25, 240.00]205.00 [150.00, 244.00]186.00 [139.00, 235.70]0.26Vital signs HR (bpm)107.93±29.05108.37±30.23106.90±26.240.72 RR (bpm)26.00 [20.00, 32.75]26.00 [20.00, 32.00]27.00 [20.00, 34.00]0.70 SPO2 (%)88.00 [79.25, 94.00]90.00 [80.00, 95.00]88.00 [79.00, 93.00]0.67 MAP (mmHg)83.50 [64.33, 99.83]82.00 [63.00, 98.33]85.00 [68.00, 102.00]0.35 GCS (points)6.00 [5.00, 10.00]6.00 [5.00, 9.00]6.00 [5.00, 10.00]0.42 Vasoactive drugs (%)98 (39.8)68 (39.3)30 (41.1)0.91 SOFA (points)9.00 [7.00, 11.00]9.00 [7.00, 11.00]9.00 [7.00, 11.00]0.80Blood routine WBC (×109/L)11.36 [8.15, 15.69]11.26 [8.04, 15.70]11.48 [8.63, 15.44]0.75 HB (g/L)105.13±29.01104.45±29.62106.74±27.650.57 HCT (%)32.59±8.9532.33±9.0633.20±8.710.49 PLT (×109/L)194.00 [135.75, 260.00]193.00 [130.00, 264.00]195.00 [148.00, 258.00]0.71 LYMPH (×109/L)0.86 [0.48, 1.42]0.88 [0.47, 1.46]0.81 [0.58, 1.26]0.82Coagulation index PT (s)12.00 [11.00, 13.40]12.08 [11.10, 13.40]11.80 [10.80, 13.40]0.52 APTT (s)29.15 [25.70, 35.58]29.80 [25.90, 35.70]28.10 [25.60, 34.90]0.38 FIB (g/L)4.01 [2.89, 5.04]4.21 [3.06, 5.12]3.67 [2.57, 4.79]0.06 D-D (mg/L)3.30 [1.47, 7.22]3.30 [1.47, 7.88]3.28 [1.48, 6.39]0.69Liver function ALT (IU/L)19.50 [12.00, 39.00]19.00 [12.00, 36.00]21.00 [13.00, 41.00]0.32 AST (IU/L)30.00 [19.00, 52.00]29.00 [19.00, 52.00]34.00 [18.00, 52.00]0.66 TBIL(μmol/L)11.90 [7.23, 20.10]11.40 [7.20, 18.80]12.40 [7.50, 22.90]0.18 LDH (U/L)223.50 [172.00, 321.00]220.00 [170.00, 301.00]231.00 [177.00, 337.00]0.33 ALB (g/L)33.16±6.3433.12±6.1833.25±6.750.88Renal function BUN (mmol/L)11.20 [7.50, 17.23]11.60 [7.90, 17.40]10.80 [7.20, 16.00]0.24 Cr (μmol/L)105.15 [66.90, 185.88]106.50 [69.30, 188.80]91.50 [62.90, 174.50]0.42Electrolyte Na+ (mmol/L)138.00 [134.00, 141.60]138.00 [134.10, 142.40]137.20 [134.00, 140.90]0.25 K+ (mmol/L)3.98 [3.56, 4.53]3.96 [3.55, 4.58]4.08 [3.62, 4.48]0.44 P (mmol/L)1.08 [0.78, 1.45]1.10 [0.78, 1.47]1.06 [0.79, 1.41]0.86Inflammatory indicators PCT (ng/mL)0.34 [0.06, 2.79]0.34 [0.07, 2.77]0.24 [0.05, 3.06]0.58 CRP (mg/L)49.26 [11.62, 105.70]48.00 [12.30, 108.55]49.98 [10.23, 104.80]0.91Cardiac indicators TNI (ng/mL)0.04 [0.01, 0.15]0.05 [0.01, 0.19]0.03 [0.01, 0.09]0.05 BNP (pg/mL)213.56 [79.24, 655.33]227.57 [88.80, 744.74]157.63 [69.71, 422.31]0.12 CT161 (65.4)115 (66.5)46 (63.0)0.71 Pneumonia124 (50.4)90 (52.0)34 (46.6)0.52 Pleural effusion53 (21.5)39 (22.5)14 (19.2)0.68Ultrasonic indicators LVEF (%)56.00 [52.00, 58.00]56.00 [52.00, 58.00]56.00 [53.00, 58.00]0.21 LVOT VTI (cm)18.00 [17.00, 19.00]18.00 [17.00, 19.00]18.00 [17.06, 19.00]0.10 IVC (cm)1.60 [1.30, 1.80]1.60 [1.30, 1.80]1.60 [1.30, 1.80]0.34 LUS (points)18.00 [11.25, 24.00]18.00 [11.00, 24.00]20.00 [13.00, 23.00]0.99Respiratory support OT38 (15.4)27 (15.6)11 (15.1)>0.99 IMV175 (71.1)122 (70.5)53 (72.6)0.86 NIV35 (14.2)25 (14.5)10 (13.7)>0.99 ETI148 (60.2)105 (60.7)43 (58.9)0.91 Tracheostomy75 (30.5)55 (31.8)20 (27.4)0.59Data are presented as median [P25, P75], mean ± standard deviation or n (%). ALB, albumin; ALT, alanine aminotransferase; APTT, activated partial thromboplastin time; AST, aspartate aminotransferase; BMI, body mass index; BNP, brain natriuretic peptide; BUN, blood urea nitrogen; COPD, chronic obstructive pulmonary disease; Cr, creatinine; CRP, C-reactive protein; CT, computed tomography; D-D, D-dimer; ETI, endotracheal intubation; FIB, fibrinogen; GCS, Glasgow Coma Scale; HB, hemoglobin; HCT, hematocrit; HR, heart rate; IMV, invasive mechanical ventilation; IVC, inferior vena cava; Lac, lactic acid; LDH, lactate dehydrogenase; LUS, lung ultrasound; LVEF, left ventricular ejection fraction; LVOT-VTI, left ventricular outflow tract velocity time integral; LYMPH, lymphocyte; MAP, mean arterial pressure; NIV, non-invasive ventilation; OT, oxygen therapy; P, phosphate; PCT, procalcitonin; PLT, platelet; PT, prothrombin time; RR, respiratory rate; SOFA, Sequential Organ Failure Assessment; TBIL, total bilirubin; TNI, cardiac troponin I; WBC, white blood cell. Comparison of clinical data between the training cohort and validation cohort Data are presented as median [P25, P75], mean ± standard deviation or n (%). ALB, albumin; ALT, alanine aminotransferase; APTT, activated partial thromboplastin time; AST, aspartate aminotransferase; BMI, body mass index; BNP, brain natriuretic peptide; BUN, blood urea nitrogen; COPD, chronic obstructive pulmonary disease; Cr, creatinine; CRP, C-reactive protein; CT, computed tomography; D-D, D-dimer; ETI, endotracheal intubation; FIB, fibrinogen; GCS, Glasgow Coma Scale; HB, hemoglobin; HCT, hematocrit; HR, heart rate; IMV, invasive mechanical ventilation; IVC, inferior vena cava; Lac, lactic acid; LDH, lactate dehydrogenase; LUS, lung ultrasound; LVEF, left ventricular ejection fraction; LVOT-VTI, left ventricular outflow tract velocity time integral; LYMPH, lymphocyte; MAP, mean arterial pressure; NIV, non-invasive ventilation; OT, oxygen therapy; P, phosphate; PCT, procalcitonin; PLT, platelet; PT, prothrombin time; RR, respiratory rate; SOFA, Sequential Organ Failure Assessment; TBIL, total bilirubin; TNI, cardiac troponin I; WBC, white blood cell. Development of prediction models The training cohort was used to develop the model, while the validation cohort was used to test its performance. As illustrated in Table 1, all variables demonstrated no significant differences between cohorts (P>0.05), indicating that the data were randomly and reasonably allocated. Parameters were selected via the LASSO regression, and the trajectory of coefficient changes for these variables is illustrated in Figure 1A. A ten-fold cross-validation was used for iterative analysis (Figure 1B). When the lambda (λ) value was 0.044, the selected variables included: pH, HCO3−, respiratory rate (RR), activated partial thromboplastin time (APTT), PCT, and IVC. Based on the variables obtained by LASSO regression (Figure 2), a Cox regression model was subsequently constructed. The nomogram was created by proportionately translating each regression coefficient in multivariable Cox analysis to a scale of 0 to 100 points. Each covariate among the 6 variables was given a score by drawing a matching vertical line straight down to the axis labeled points. The 15- and 28-day survival probabilities may be calculated by summing all scores and determining their position on the scale. Variable selection based on LASSO regression. (A) Coefficient paths of candidate variables. (B) Selection of the optimal λ value using 10-fold cross-validation. LASSO, least absolute shrinkage and selection operator. Nomogram for predicting MODS risk in individuals with SRF. APTT, activated partial thromboplastin time; IVC, inferior vena cava; MODS, multiple organ dysfunction syndrome; PCT, procalcitonin; RR, respiratory rate; SRF, severe respiratory failure. Validation of prediction models The C-index of the predictive model was 0.793 in the training cohort and 0.748 in the validation cohort. Excellent agreement between the predicted and observed risks was observed in the calibration curves for both cohorts (Figure 3A-3D). In the training cohort, the area under the ROC curve (AUC) value for assessing MODS risk was 0.824 (95% CI: 0.726–0.923) at 15 days and 0.809 (95% CI: 0.676–0.942) at 28 days (Figure 4A). The corresponding AUCs were 0.721 (95% CI: 0.501–0.942) and 0.737 (95% CI: 0.503–0.971) in the validation cohort (Figure 4B). Calibration curves of the nomogram at 15 and 28 days in the training cohort (A,B) and validation cohort (C,D). ROC curves for predicting MODS at 15 and 28 days in the training cohort (A) and validation cohort (B). AUC, area under the curve; CI, confidence interval; MODS, multiple organ dysfunction syndrome; ROC, receiver operating characteristic. Survival analysis Risk scores were calculated for all patients based on the constructed prognostic model. Using X-tile software, individuals were categorized into low- and high-risk groups based on the optimal cutoff value (118.4). As illustrated in Figure 5A,5B, the high-risk group had a significantly higher risk of MODS compared with the low-risk group (P<0.05). Kaplan-Meier curves comparing MODS incidence between high-risk and low-risk groups. (A) Kaplan-Meier curve in the training cohort. (B) Kaplan-Meier curve in the validation cohort. MODS, multiple organ dysfunction syndrome. Discussion This research successfully constructed and validated a prognostic model for assessing in-hospital MODS in ICU patients with SRF based on ultrasound and laboratory parameters. Six independent predictors, namely pH, HCO3−, RR, APTT, PCT, and IVC, were identified via LASSO regression (all P<0.05) and integrated into a Cox proportional hazards model. The model exhibited robust discriminative performance, with a C-index value of 0.793 in the training cohort and 0.748 in the validation cohort. Calibration curves in the validation cohort demonstrated excellent agreement between predicted and observed risks. A nomogram based on this model, along with an X-tile-based risk stratification system validated by K-M survival analysis, further supported its clinical utility in effectively identifying high-risk patients. Numerous critical care prediction models (e.g., APACHE II, SOFA) are designed to predict the risk of all-cause mortality, a retrospective terminal outcome that reflects overall patient severity but fails to identify reversible intermediate pathological processes driving poor prognosis. Peng et al. (16) have directly compared traditional scoring systems with prediction models for MODS and demonstrated that models using mortality as the endpoint, such as APACHE II, lag behind the occurrence of MODS and are unable to guide early clinical intervention. Ren et al. (17) have demonstrated in patients with severe heat stroke that prediction models for MODS are more efficacious in early warning and provide incremental clinical benefit compared with models for predicting all-cause mortality. Our model holds unique incremental clinical benefits in predicting in-hospital MODS in SRF patients compared to models for predicting mortality. MODS is the core intermediate and reversible determinant of adverse outcomes in SRF patients, rather than a final endpoint (18). Predicting MODS identifies the critical prognostic turning point earlier than the prediction of mortality, enabling clinicians to intervene before organ dysfunction progresses to irreversible failure. Notably, it was the first model to combine ultrasound indicators (such as IVC) with laboratory data, assisting in the real-time assessment of the impact of intravascular volume status on organ perfusion. A decreased IVC diameter suggests intravascular volume depletion (such as dehydration and hemorrhage), resulting in reduced venous return and cardiac output, which in turn leads to impaired organ perfusion. This may cause microcirculatory dysfunction in the kidneys, liver, and intestines, ultimately contributing to the development of MODS (19). Sustained hypoperfusion activates the renin-angiotensin system, leading to microvascular constriction in organs such as the intestines and kidneys, and ultimately resulting in ischemic renal injury or hepatic dysfunction (20). In comparison to conventional assessment tools, including SOFA and the Acute Physiology and Chronic Health Evaluation II scores, this model demonstrated superior predictive accuracy. In the validation cohort, the C-index (0.721–0.737) outperformed that of the SOFA score in similar populations (typically <0.65) (21), and was comparable to recently developed machine learning models such as XGBoost (C-index ≈0.70) (22). PCT and RR represent systemic inflammation and respiratory compensatory capacity, respectively. Among septic patients, those with PCT >5 ng/mL have a 7.1-fold elevated risk of MOF within 72 hours in comparison to those with levels <0.5 ng/mL (23). An elevated RR reflects an imbalance in oxygen supply and demand (DO2/VO2). An RR >27 breaths per minute demonstrates a sensitivity of 92% for predicting MODS within 24 hours in ICU patients (AUC =0.87), outperforming the respiratory component of SOFA (AUC =0.68) (24). Coagulopathy (as reflected by prolonged APTT) and metabolic acidosis (marked by decreased pH and HCO3−) are hallmarks of microcirculatory failure and directly contribute to ischemia injury of multiple organs (25). Prolonged APTT, decreased pH, and reduced HCO3− levels incorporated into the present model collectively reflect a core pathophysiological mechanism of microcirculatory failure—namely, the interplay between coagulopathy and metabolic acidosis. Tissue hypoxia induces endothelial TF release, thereby activating the extrinsic coagulation pathway. Concurrently, exposure of subendothelial collagen activates factor XII and further initiates the intrinsic pathway, which is primarily reflected by APTT (26). In septic patients, an APTT >50 seconds increases the risk of glomerular microthrombosis by 4.2-fold, directly contributing to acute tubular necrosis (27). HCO3– <18 mmol/L indicates decompensated acidosis (typically with pH <7.25), during which microvascular autoregulation is lost. For every 0.1 decrease in pH, microcirculatory blood flow velocity declines by 15–20% (28). Patients classified as high-risk (total points ≥118.4) by the established nomogram require intensified, targeted clinical interventions to mitigate the progression of MODS and improve survival outcomes. First, enhanced hemodynamic and respiratory monitoring is warranted, including continuous arterial blood gas analysis to correct acid-base disturbances (e.g., mechanical ventilation optimization for refractory hypoxemia, targeted fluid resuscitation guided by IVC ultrasound to restore tissue perfusion) (29). Second, coagulation and inflammation control should be prioritized. Serial monitoring of APTT and PCT levels is necessary, and antimicrobial therapy should be initiated once infections are confirmed. Furthermore, anticoagulants should be used to prevent microcirculatory dysfunction (30). Third, multidisciplinary organ support is essential, including renal replacement therapy for relieving AKI and nutritional support for attenuating catabolic stress (31). By identifying patients at high risk of progressive multiple organ dysfunction, this nomogram enables clinicians to specifically discuss the escalation of organ support, intensity of monitoring, and realistic therapeutic expectations. This risk stratification based on mechanism helps clinicians translate prognostic information into actionable treatment plans and communicate with patients and families about more personalized and meaningful goals of care. By integrating multidimensional parameters, this model addresses the limitations of single-organ scoring systems and provides precise targets for early interventions, including fluid optimization, antimicrobial therapy, and anticoagulation. Despite these advantages, some limitations should be noted. Firstly, this was a single-center retrospective study. Future prospective multicenter validation is necessary to ensure the generalizability of the model. Secondly, emerging biomarkers (such as cytokines, endothelial injury markers) were not included, potentially omitting relevant inflammatory pathways. Thirdly, the model relies on ultrasound measurements, and IVC assessment is operator-dependent. Therefore, standardized training is required to reduce variability. Fourthly, the relatively small sample size, particularly in the validation cohort (n<200), may affect the stability of K-M analysis in high-risk subgroups. Conclusions The LASSO-Cox model developed in this study, which integrated ultrasound (such as IVC) and laboratory indicators (pH, HCO3−, RR, APTT, and PCT), significantly improved the predictive accuracy of MODS risk in ICU patients with SRF. The nomogram and corresponding risk stratification system derived from this model may assist clinicians in the early identification of individuals at high risk and support individualized intervention strategies. Further multicenter studies are warranted to optimize this model and investigate the incorporation of real-time dynamic monitoring indicators. Supplementary The article’s supplementary files as Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1-0291/prf