This investigation enrolled participants admitted to the ICUs at the Central Theater General Hospital of the People’s Liberation Army of China and Hefei BOE Hospital from June 2019 to June 2022. We included individuals meeting the following criteria: 1) diagnosed with moderate to severe ARDS according to the 2012 Berlin criteria, displaying a partial pressure of arterial oxygen to fractional inspired oxygen (PaO₂/FiO₂) ratio ≤ 200 mmHg under a standardized ventilation regime (Force et al., 2012); 2) ARDS originating from indirect lung injury causes, including but not limited to non-pulmonary sepsis, trauma not involving the lungs, adverse reactions to blood transfusions, pancreatitis, among others; 3) diagnosis of ARDS confirmed within the initial 24 h following ICU admission. The exclusion criteria encompassed: 1) age below 18 or above 85 years; 2) pregnancy; 3) history of pulmonary tuberculosis, lung cancer, severe chronic obstructive pulmonary disease (COPD), chronic interstitial lung disease, or prior lung surgery; 4) presence of massive pleural effusion or pneumothorax; 5) cardiogenic pulmonary edema or hypoxemia primarily due to cardiac conditions; 6) inadequate acoustic window for ultrasound imaging; 7) inability to ascertain clinical outcomes within the study’s observation period.

This study was conducted in strict accordance with ethical principles and received approval from the review boards of the two participating hospital institutions (Approval No: 201910216; 20190010). Prior to enrollment, written informed consent was obtained from the legal guardians of the patients. All patient information was anonymized before analysis to ensure confidentiality.

ARDS was identified following the Berlin Definition criteria, which necessitate: an acute inception, evident bilateral infiltrates on chest radiographs not fully explained by effusions, lobar/lung collapse, or nodules; a non-cardiogenic origin as indicated by the absence of left atrial hypertension; and a PaO₂/FiO₂ ratio less than 300 mmHg (Force et al., 2012). Distinctively, extrapulmonary ARDS results from systemic factors that precipitate vascular endothelial damage, enhance pulmonary vascular permeability, lead to interstitial and alveolar exudation, and culminate in alveolar collapse, edema, and respiratory failure (Bernard et al., 1994).

Upon enrollment, we systematically collected comprehensive baseline clinical data for each participant. This encompassed demographic and physiological characteristics, the etiological factors contributing to ARDS, and any pre-existing health conditions. Within the first 24 h post-enrollment, we meticulously recorded vital signs, laboratory findings, assessed the severity of the disease using the Acute Physiology and Chronic Health Evaluation (APACHE) II scoring system, and determined the degree of oxygenation impairment via the PaO₂/FiO₂ ratio. These initial assessments provided insight into the patients’ baseline health status and the immediate therapeutic interventions initiated upon study entry.

Demographic data included age, sex, body mass index (BMI), and smoking history. The causative factors of ARDS were categorized into distinct groups: non-pulmonary sepsis, non-pulmonary trauma, complications arising from multiple blood transfusions, pancreatitis, and other causes. We also compiled an extensive list of pre-existing conditions, such as hypertension, coronary artery disease, diabetes mellitus, and cerebrovascular diseases.

Vital signs included mean arterial pressure, heart rate, respiratory rate, and body temperature. The laboratory parameters assessed encompassed hemoglobin concentration, white blood cell count, platelet count, serum creatinine, total bilirubin, albumin levels, troponin T, NT-proBNP, D-dimer, lactate, procalcitonin, C-reactive protein, and electrolytes (sodium and potassium). The APACHE II score was calculated to assess the initial severity of the illness (Knaus et al., 1985), while the PaO₂/FiO₂ ratio (Force et al., 2012) provided an index of the baseline level of pulmonary oxygenation dysfunction.

Therapeutic interventions initiated within the first 24 h of patient enrollment were meticulously recorded. These included the application of mechanical ventilation, the adoption of PP, the administration of vasoactive drugs, sedation, neuromuscular blocking agents, systemic corticosteroids, antibiotics, antifungal treatments, continuous renal replacement therapy, and venovenous ECMO (VV ECMO). These management strategies were individualized based on each patient’s clinical status and underlying conditions, in alignment with current best practice guidelines and clinical protocols.

All study participants received care in the ICU adhering to globally recognized ARDS management guidelines and consensus documents. Initial respiratory support was provided using non-invasive or invasive mechanical ventilation (NIV or IMV, respectively) upon patient enrollment. Transition from NIV to IMV was considered under specific conditions such as a high Simplified Acute Physiological Score (>34), PaO₂/FiO₂ ratio ≤ 175 mmHg (indicating severe hypoxemia) (Antonelli et al., 2007), lack of hypoxemia improvement after 1 h of NIV, or excessive spontaneous breathing leading to large tidal volumes (>9 mL/kg) (Frat et al., 2018).

For patients on IMV, a lung-protective ventilation strategy was employed aiming to optimize oxygenation and minimize ventilator-induced lung injury. Specific targets included maintaining peripheral capillary oxygen saturation between 88% and 95%, PaO₂ levels of 55–80 mmHg, partial pressure of arterial carbon dioxide (PaCO₂) ≤ 50 mmHg or pH ≥ 7.25, plateau pressure (Pplat) ≤ 30 cmH₂O, and driving pressure (∆P) ≤ 15 cmH₂O (Khemani et al., 2018). Tidal volume was initially set at 6 mL/kg of predicted body weight and adjusted based on Pplat and ∆P assessments. PEEP levels were set according to FiO₂ requirements and subsequently fine-tuned based on Pplat, ∆P, and PaCO₂ levels. Respiratory rate adjustments were made to manage PaCO₂ and pH levels effectively. PP was implemented for durations exceeding 12 h per day in cases of persistent acidosis (pH < 7.25), PaCO₂ > 50 mmHg, or when the PaO₂/FiO₂ ratio was ≤150 mmHg despite a PEEP ≥ 5 cmH₂O, unless contraindicated (Fan et al., 2017; Fichtner et al., 2018). Neuromuscular blockade was utilized in scenarios of unresolving hypoxemia, during PP, or to prevent ventilator-induced injuries, in line with recent rapid practice guidelines (Alhazzani et al., 2020).

VV ECMO was considered for patients with severe respiratory failure unresponsive to mechanical ventilation if specific criteria were met, including a treatment duration under 7 days with FiO₂ ≥ 0.80, a PaO₂/FiO₂ ratio < 50 mmHg for over 3 h or <80 mmHg for over 6 h, pH < 7.25 with PaCO₂ ≥ 60 mmHg for more than 6 h, respiratory rates ≥ 35 min, or Pplat ≤ 32 cmH₂O (Tonna et al., 2021).

Fluid management and vasoactive drug administration were tailored by the attending physician based on hemodynamic status, ARDS severity, and cardiac and renal function. In cases of concurrent sepsis, fluid and vasopressor strategies followed sepsis-specific guidelines (Singer et al., 2016). Corticosteroid use adhered to guidelines addressing critical illness-associated corticosteroid insufficiency and international sepsis guidelines (Singer et al., 2016; Pastores et al., 2018). Additionally, nutritional support and venous thromboembolism prophylaxis were provided in accordance with critical care guidelines and consensus statements (Di Nisio et al., 2016; Burgos et al., 2018).

LUS examinations were conducted within the initial 24 h post-enrollment. The pulmonary regions were systematically partitioned into twelve segments via anatomical delineations, which included the parasternal, anterior axillary, posterior axillary, posterior median lines, and the interscapular line. Utilizing a phased array convex transducer (frequency range: 3.5–10 MHz, manufactured by GE Healthcare, United States), scanning was meticulously performed across these predefined segments. Pulmonary ultrasound findings were quantified based on a four-point scale reflecting the extent of ventilatory compromise: 1) Score 0 indicating normal aeration evidenced by lung sliding and the presence of A-lines or fewer than three discrete B-lines; 2) Score 1 for moderate aeration loss, characterized by the appearance of three or more scattered B-lines; 3) Score 2 denoting severe ventilatory impairment with confluent B-lines; 4) Score 3 corresponding to pulmonary consolidation, identifiable by tissue-like echotexture and bronchogram signs. A composite score ranging from 0 to 36 was then computed, providing an aggregate measure of pulmonary aeration deficit (Mongodi et al., 2017).

Serum samples for sTM quantification were obtained within the first 24 h following patient enrollment. Concentrations of sTM were determined employing a dual-antibody sandwich enzyme-linked immunosorbent assay, specifically the kit supplied by Abcam (catalogue number ab46508, Shanghai, China). All procedures were rigorously conducted in adherence to the manufacturer-provided protocol.

The principal outcome was assessed as all-cause mortality within a 28-day interval post-admission. Patients were subsequently categorized into two distinct cohorts predicated on their 28-day survival outlook: those who survived (survival group) and those who did not (non-survival group).

Data analyses were executed using standardized statistical methodologies. Continuous variables underwent normality testing through the Shapiro-Wilk test, alongside Bartlett’s test for assessing variance homogeneity. Variables adhering to normal distribution were articulated as mean ± standard deviation, with inter-group discrepancies evaluated via the independent samples t-test. Conversely, variables deviating from normal distribution were presented as median (interquartile range, IQR), with the Mann-Whitney U test facilitating inter-group comparisons. Categorical variables were represented as percentages (%) and analyzed using the chi-square test. Univariate Cox regression analysis was employed for variables deemed clinically pertinent and demonstrating significant variances between survival outcomes. Lasso regression was utilized for risk factor identification, with the Schoenfeld residual test confirming the proportional hazard (PH) assumption post Lasso-Cox regression. A prognostic nomogram was formulated based on Lasso-Cox regression outcomes, its predictive accuracy gauged through the concordance index (C-index), calibration plots, internal validation, decision curve analysis (DCA), interventions avoided analysis (IAA), and receiver operating characteristic (ROC) analysis. The area under the ROC curve (AUC) was used to determine the predictive accuracy of each factor, and the DeLong test was employed to compare the AUCs. Patient survival probabilities were delineated via the Kaplan-Meier method. Data processing was accomplished using SPSS version 22.0 and STATA version 17.0, with statistical significance set at a p-value < 0.05 (two-tailed).

During the recruitment phase, 113 individuals were identified as eligible according to the inclusion criteria, as depicted in Figure 1. However, 18 participants were subsequently excluded due to reasons including age constraints (below 18 or above 85 years), the presence of severe pre-existing pulmonary conditions or recent thoracic surgeries, significant pleural effusion or pneumothorax, cardiogenic pulmonary edema, inadequate echogenicity for ultrasound examination, and the inability to ascertain clinical outcomes within the study’s observational timeframe. Consequently, the study proceeded with 95 participants, comprising 32 individuals in the non-survival cohort (33.68%) and 63 in the survival cohort (66.32%).

The baseline levels of sTM and LUS were compared between patients who survived and those who did not survive in order to investigate potential differences related to patient outcomes. Table 3 illustrates that survivors exhibited significantly lower serum sTM levels and LUS compared to non-survivors (all p < 0.05).

Univariate Cox regression analysis was conducted to identify potential predictors for 28-day mortality. The analysis revealed that sTM level [odds ratio (OR) = 1.030, 95% confidence interval (CI) 1.018–1.042, p < 0.001], LUS (OR = 1.255, 95% CI 1.122–1.404, p < 0.001), APACHE II score (OR = 1.232, 95% CI 1.073–1.415, p = 0.003), PaO₂/FiO₂ ratio (OR = 0.959, 95% CI 0.934–0.985, p = 0.002), platelet count (OR = 0.994, 95% CI 0.988–0.999, p = 0.028), creatinine (OR = 1.005, 95% CI 1.001–1.009, p = 0.026), and lactate (OR = 1.212, 95% CI 1.073–1.369, p = 0.002) were significantly associated with 28-day mortality (Table 4).

Subsequently, Lasso regression analysis was employed to further refine the predictive variables. The parameter path was estimated over a range of values for λ, as depicted in Figure 2A. Through iterative analysis and 10-fold crossvalidation, the optimal value of λ was determined to be 0.095, resulting in a model with exceptional performance and minimal variables (Figure 2B). The final screened variables included sTM, LUS, and lactate.

To ensure the validity of the model, the baseline levels of sTM, LUS, and lactate were subjected to the proportional risk hypothesis test based on Schoenfeld residuals (Figure 2C). The results showed that all variables met the PH assumptions (sTM, p = 0.407; LUS, p = 0.672; lactate, p = 0.066). Therefore, all three variables were utilized in the predictive modeling for 28-day mortality.

The constructed nomogram visually represents the impact of sTM, LUS, and lactate on patient survival outcomes. The “total points” projection provides an estimation of the probability of 28-day mortality based on the combination of these variables (Figure 3A), facilitating informed clinical decision-making. The median C-index for the nomogram was 0.807 (0.740–0.873), indicating excellent predictive accuracy. Calibration curve analysis demonstrated strong agreement between predicted and observed outcomes (Figure 3B), reinforcing the reliability of the nomogram. Internal validation of the nomogram yielded a C-index of 0.875, confirming its robustness (Figure 3C).

To evaluate the clinical utility of the predictive model, DCA and IAA were performed. Figure 4A illustrates the clinical effectiveness of the nomogram predictive model as assessed by DCA, demonstrating a higher net benefit for clinical decision-making within a threshold probability range of 0–0.71. Figure 4B shows that the IAA analysis indicates a net reduction in unnecessary interventions within a threshold probability range of 0–0.73. Table 5 presents the ROC analysis of the model’s predictive performance. The nomogram demonstrated significantly superior predictive ability for 28-day mortality compared to individual factors such as sTM, LUS, and lactate, with an AUC of 0.873 (95% CI 0.789–0.932). The DeLong test confirmed that the AUC differences between the nomogram and each individual factor were statistically significant (all p < 0.05). Additionally, the nomogram exhibited higher sensitivity (87.50%) and specificity (77.78%) in predicting 28-day mortality compared to these individual factors. Figure 4C visually illustrates the AUC values obtained from the ROC analysis. Furthermore, the Kaplan-Meier survival analysis, depicted in Table 6, indicated a significantly lower overall survival rate in the high-risk group compared to the low-risk group (66.67% vs. 7.55%, p < 0.001). The mean survival time was also substantially shorter for the high-risk group (17.69 days vs. 27.13 days, 95% CI 15.19–20.19 days vs. 26.29–27.97 days, p < 0.001). Figure 4D illustrates the Kaplan-Meier survival curves for the high-risk and low-risk groups.

In clinical practice, the APACHE II score is a widely accepted prognostic tool for assessing the severity of critical illness, while the PaO₂/FiO₂ ratio is a reliable indicator for evaluating the severity of ARDS. This study conducted a comparative analysis to assess the predictive accuracy of the combined nomogram versus traditional models, particularly the APACHE II score and PaO₂/FiO₂ ratio.

ROC analysis was employed to evaluate and compare the predictive accuracy of these models. As shown in Table 7, the nomogram achieved a higher AUC than both the APACHE II score (AUC 0.676, 95% CI 0.572–0.768) and the PaO₂/FiO₂ ratio (AUC 0.704, 95% CI 0.602–0.794). The DeLong test confirmed that the differences in AUC between the nomogram and each individual factor were statistically significant (all p < 0.05). Additionally, the nomogram showed superior sensitivity and specificity for predicting 28-day mortality compared to the APACHE II score (sensitivity 50.00%, specificity 76.19%) and the PaO₂/FiO₂ ratio (sensitivity 78.12%, specificity 60.32%). Figure 5 provides a visual representation of the AUC values for these models.

The efficacy of PP in enhancing the clinical outcomes of patients afflicted with ARDS has been well-documented in the literature (Guerin et al., 2013). Nonetheless, the introduction of medical interventions during the observation period can potentially introduce confounding factors, thereby affecting the predictive accuracy of our combined nomogram. To mitigate this issue and elucidate the specific impact of PP on clinical outcomes, we conducted a subgroup analysis, categorizing patients based on their exposure to PP during the study period. Patients were thus classified into two groups: those who did not undergo PP treatment (non-PP group) and those who received PP treatment (PP group).

At the commencement of the study, comparative analysis of baseline characteristics, specifically sTM, LUS, and lactate levels, between the PP and non-PP groups revealed no statistically significant differences, as delineated in Table 8. Subsequent stratification based on PP exposure facilitated a more nuanced analysis of the relationship between these biomarkers and patient outcomes. Table 9 delineates the differential baseline levels of sTM, LUS, and lactate between patients who survived and those who did not, following stratification by PP status.

Moreover, Table 10 articulates the findings derived from both univariate and multivariate Cox regression analyses, examining the association between sTM, LUS, lactate levels, and 28-day mortality subsequent to stratification by PP. These analyses were meticulously adjusted for potential confounders, including the APACHE II score, PaO₂/FiO₂ ratio, platelet count, and creatinine levels. Intriguingly, our analysis revealed that, upon adjusting for these confounding factors, the associations of sTM, LUS, and lactate levels with 28-day mortality were no longer statistically significant among patients stratified by PP.