Use of Dynamic Soluble Programmed Death-Ligand 1 Trajectories to Identify Early Organ Dysfunction and Predict Mortality in Critical Coronavirus Disease 2019

Use of Dynamic Soluble Programmed Death-Ligand 1 Trajectories to Identify Early Organ Dysfunction and Predict Mortality in Critical Coronavirus Disease 2019 | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Use of Dynamic Soluble Programmed Death-Ligand 1 Trajectories to Identify Early Organ Dysfunction and Predict Mortality in Critical Coronavirus Disease 2019 Shungo Takeuchi, Eiji Kawamoto, Takashi Matsusaki, Daisuke Ono, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7521856/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 12 You are reading this latest preprint version Abstract Background Persistent immune checkpoint activation is a recognized feature of critical Coronavirus Disease 2019 (COVID-19). However, the temporal behavior and clinical utility of soluble Programmed Death-Ligand 1 (sPD-L1) remain unclear. We aimed to investigate the longitudinal changes in sPD-L1, their relationship with organ dysfunction markers, and their prognostic value when combined with machine learning (ML) models. Methods This single-center observational study included 40 adults with severe COVID-19 pneumonia admitted to the intensive care units (ICU) (April 2021–December 2022) and 23 healthy volunteers. We measured plasma sPD-L1 on ICU days 1, 5, 7, 14, and 21. Routine biochemistry, full blood counts, and arterial blood gas analyses were conducted in parallel. Cox regression analysis was used to identify independent predictors of hospital mortality, which was the primary outcome. Eight ML classifiers were trained on admission variables, as well as day 1, 5, and 7 sPD-L1 levels. Discrimination was assessed using stratified five-fold cross-validation and Shapley Additive Explanations (SHAP) attribution. Results Ten of the forty patients died during hospitalization. Overall, sPD-L1 levels declined during the ICU stay but remained persistently high in non-survivors. Values on days 5 and 7 differed significantly between survivors and non-survivors (p = 0.023 and 0.001, respectively). In multivariable Cox analysis, day-7 sPD-L1 and arterial lactate levels on admission independently predicted mortality. Day 7 sPD-L1 level correlated positively with creatinine, C-reactive protein, and fibrinogen (all p < 0.05), linking immune checkpoint activation to renal injury, inflammation, and coagulopathy. A support vector machine model achieved the highest discriminative accuracy (mean area under the curve = 0.917). Day 5 sPD-L1 was designated as the primary predictor of mortality based on SHAP attribution, with lactate contributing minimally. Conclusion Sustained sPD-L1 elevation in the initial ICU week is strongly associated with early organ dysfunction and independently predicts death in critical COVID-19. Incorporating serial sPD-L1 levels into bedside ML models significantly enhances risk discrimination. These findings support sPD-L1 as an integrative biomarker of immune–renal–coagulation interplay, thus necessitating validation in larger multicenter cohorts and exploration as a potential companion marker for immune-modulatory interventions. COVID-19 soluble PD-L1 immune checkpoints organ dysfunction ICU mortality machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Background Severe Coronavirus Disease 2019 (COVID-19) requiring intensive care is associated with high mortality, with intensive care unit (ICU) mortality rates for patients with COVID-19 who are critically ill averaging 30–40% across various studies and settings [ 1 ]. An early “cytokine storm” hyperinflammatory phase of COVID-19 has been well recognized [ 2 , 3 ]; however, there is growing evidence that many patients with critical illness also develop a state of immune exhaustion or secondary immunosuppression [ 4 ]. Marked lymphopenia and T-cell dysfunction in severe COVID-19 correlate with worse outcomes, mirroring the “immunoparalysis” observed in septic shock [ 5 , 6 ]. Programmed death-ligand 1 (PD-L1), a pivotal immune checkpoint molecule, downregulates immune responses by engaging PD-1 on T cells, leading to their functional exhaustion [ 7 , 8 ]. In sepsis, increased PD-L1 expression is implicated in monocyte dysfunction, impaired cytokine production, and poor clinical outcomes [ 5 ]. Circulating or soluble PD-L1 (sPD-L1) has emerged as a potential biomarker of immune suppression in patients with critical illness. Preclinical studies show that blocking this pathway can improve survival [ 9 ]. These findings position PD-L1 as a mechanistic marker of immune dysfunction as well as a potential prognostic indicator for critical illness. Recent evidence suggests that the programmed death 1 (PD-1)/PD-L1 pathway is also dysregulated in severe COVID-19 cases [ 10 , 11 ]. Beserra et al. demonstrated significantly higher sPD-L1 concentrations in hospitalized patients with COVID-19 than in healthy controls, suggesting its role in the pathophysiology of severe infection [ 11 ]. Other studies have identified a “storm” of soluble immune checkpoints, including PD-L1, which correlates with disease severity [ 12 ]. These data show that PD-L1 may contribute to COVID-19-associated immune dysregulation, consistent with observations in bacterial sepsis. Currently, the lack of longitudinal data on PD-L1 dynamics during critical illness is a significant gap in the literature. In most studies, PD-L1 (or sPD-L1) has been assessed at a single time point, such as upon hospital admission, providing a limited snapshot of the disease. The evolution of PD-L1 levels during intensive care remains unclear, and it is unknown whether these changes correlate with clinical outcomes. Dynamic monitoring of immune checkpoints, such as PD-1/PD-L1, has been proposed for prognosis and immunotherapy in sepsis [ 5 ]; however, longitudinal studies remain rare. Temporal data on PD-L1 are particularly scarce in COVID-19. One study reported no significant change in sPD-L1 levels over time after Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) infection, but only a few time points were analyzed [ 11 ]. Overall, it remains uncertain whether PD-L1 levels remain elevated, fluctuate, or decline during ICU stay. Understanding these patterns could help identify transitions between hyperinflammatory and immunosuppressive phases, guiding the optimal timing of immunomodulatory interventions. To address this gap, our study was designed to characterize the longitudinal trajectory of PD-L1 levels in patients with critical illness owing to COVID-19 and evaluate the prognostic significance of these time-dependent changes. We hypothesized that patients who died would show persistently elevated or increasing PD-L1 levels, whereas survivors would show declining levels. By analyzing the temporal dynamics of PD-L1 levels in this study, we aimed to provide novel insights into the host immune response in severe COVID-19 and determine whether PD-L1 could serve as a dynamic prognostic biomarker in critical care. Methods 1. Study Design and Participants This single-center, retrospective observational study was conducted at Mie University Hospital between April 2021 and December 2022. We included 40 consecutive adult patients who were hospitalized with laboratory-confirmed COVID-19—defined by a positive reverse-transcription polymerase chain reaction assay for SARS-CoV-2—in the analysis. We also recruited 23 age- and sex-matched healthy volunteers to serve as the control group. Patients were excluded if key variables were missing or if informed consent could not be obtained. The study protocol was approved by the Institutional Review Board of Mie University Hospital (approval numbers: IRB No. 3026 and H2021-191). We obtained written informed consent from the patients or their next of kin upon admission to the hospital. 2. Data Collection and Measurements We collected demographic data, comorbidities, and laboratory values from the electronic medical records. Plasma samples were collected within 24 h of hospital admission (day 1) and subsequently on days 5, 7, 14, and 21 of hospitalization. Serum samples were obtained via EDTA-containing tubes, which were stored at − 80°C until use. PD-L1 levels were measured using the Human PD-L1 SimpleStep ELISA Kit (28 − 8 clone; Abcam, ab277712, Cambridge, UK) according to the manufacturer’s instructions. We also routinely performed blood tests—including complete blood counts, coagulation studies, and biochemical assays—to monitor the clinical status and severity of the disease. The primary outcome was in-hospital mortality. 3. Statistical Analysis Continuous variables were expressed as mean ± standard deviation (SD) or median interquartile range (IQR) and compared using Student’s t-test or Mann–Whitney U test. Categorical variables were analyzed with Fisher’s exact test. Kaplan–Meier survival analysis, along with a log-rank test, was used to compare mortality between groups and assess statistical significance. The Cox proportional hazards regression model was used to identify predictors of mortality. Furthermore, it was implemented using the lifelines Python package to incorporate longitudinal data. All 46 candidate predictors were first analyzed using separate univariable Cox models. We subsequently adjusted p-values for multiplicity using the Benjamini–Hochberg procedure, controlling the false discovery rate (FDR) at 5%. Predictors with an FDR-adjusted q-value < 0.05, along with prespecified clinically essential covariates, were carried forward into the multivariable analysis. To prevent overfitting in this small data set (10 events), we: (i) restricted the final model to three clinically supported variables (event-per-variable ratio ≈ 3.3); (ii) applied L1-penalized (LASSO) Cox regression with λ chosen by stratified 3-fold cross-validation to shrink the coefficients; and (iii) estimated model optimism using the same cross-validation scheme. Simulation studies suggest that under strong penalization, acceptable bias and calibration can be maintained with event-to-variable ratios as low as 5 [ 13 , 14 ]. Therefore, the resulting model should be considered hypothesis-generating, and its stability will require confirmation in larger, independent cohorts. 4. Machine Learning Models Machine learning (ML) modeling was performed in Python 3.10 using scikit-learn v1.4 and Shapley Additive Explanations (SHAP) v0.45.0. The analysis included all admission-time laboratory and clinical variables, along with serial sPD-L1 concentrations (days 1, 5, 7, 14, and 21). We retained variables with ≤ 25% missing data, which were replaced with the column mean. Non-informative identifiers and the binary outcome label were excluded from the predictor matrix. Continuous predictors were z-transformed using StandardScaler, with scaling parameters estimated on each training fold and subsequently applied to the corresponding test fold to prevent data leakage. We evaluated eight supervised classifiers: support vector machine (SVM), neural network, decision tree (DT), AdaBoost, gradient boosting machine (GBM), linear discriminant analysis (LDA), logistic regression, and random forest (RF). Hyperparameters for SVM, neural network (NN), decision tree (DT), and RF were tuned using a grid search (Additional file 1). Model development followed a stratified 5-fold cross-validation scheme. Within each training fold, an inner grid search identified the optimal hyperparameter set (scoring = area under the receiver operating characteristic [ROC] curve [AUC]). The best estimator was then refitted on the full training fold and evaluated on the held-out test fold. We used the predicted class probabilities to construct ROC curves and to calculate the AUC for each fold. True positive rates were interpolated at 100 equally spaced false positive rate points, and mean ROC curves were generated across folds for every algorithm. Model interpretability was assessed using SHAP. KernelExplainer was fitted on the training data, and global feature importance was ranked using the mean absolute SHAP values. All computations were executed with a fixed random seed (random_state = 42). 5. Software and Tools All statistical and ML analyses were performed using Python version 3.11. The following libraries were used: lifelines (for survival modeling) and scikit-learn (for ML model training). Data preprocessing and visualization were conducted using pandas, NumPy, Matplotlib, and Seaborn. Results Patient Characteristics A total of 40 patients with severe COVID-19 who were admitted to Mie University Hospital were included in the study. Ten of these patients (25%) died during hospitalization. Table 1 shows a summary of the baseline characteristics and laboratory data at admission, stratified by survival status. The mean age was 59.8 ± 14.3 years in survivors and 64.3 ± 14.4 years in non-survivors (p = 0.109). Non-survivors had significantly higher Sequential Organ Failure Assessment (SOFA) scores (7.40 ± 3.60 vs. 3.57 ± 1.77; p < 0.001) and lower platelet counts (182.7 ± 102.7 vs. 249.1 ± 107.6 ×10³/µL; p = 0.036) than their counterparts. Creatinine and blood urea nitrogen (BUN) levels were also significantly elevated in the non-survivor group (Creatinine: 2.40 ± 1.80 vs. 0.90 ± 0.52 mg/dL, p = 0.009; BUN: 45.5 ± 29.8 vs. 23.9 ± 17.6 mg/dL, p = 0.046), suggesting impaired renal function. Other variables, including Glasgow Coma Scale score, PaO 2 /FiO 2 (P/F) ratio, inflammatory markers, coagulation parameters, and arterial blood gas values, showed no significant differences between groups. Table 1 Baseline characteristics of patients with COVID-19 Survivors Non-survivors n = 30 n = 10 p-value Age, years 59.78 ± 14.32 64.29 ± 14.44 0.109 Sex (Female/Male) 8/22 1/9 0.404 Body Mass Index, kg/m 2 30.00 ± 7.75 27.67 ± 5.44 0.739 SOFA score 3.57 ± 1.77 7.40 ± 3.60 < 0.001 Glasgow Coma Scale 13.40 ± 3.77 11.50 ± 5.02 0.396 P/F ratio 202.36 ± 112.84 175.01 ± 58.76 0.612 Comorbidities Hypertension 14 5 1.000 Diabetes mellites 11 5 0.482 Kidney disorder 1 3 0.042 Respiratory disorder 8 1 0.404 Cardiovascular disorder 4 3 0.338 Liver disorder 0 2 0.058 Immune disease 1 0 1.000 CBC Hemoglobin(g/dL) 13.41 ± 1.40 12.84 ± 2.93 0.656 WBC (×10³/µL) 9.32 ± 5.01 7.75 ± 4.09 0.301 Lymphocyte (%) 9.42 ± 6.22 13.26 ± 10.76 0.48 Neutrophil (%) 84.98 ± 7.88 78.58 ± 13.54 0.18 Platelet (×10³/µL) 249.10 ± 107.60 182.67 ± 102.74 0.036 Coag D-dimer (µg/mL) 5.31 ± 8.56 28.26 ± 74.65 0.301 PT-INR 1.17 ± 0.26 1.18 ± 0.25 0.818 APTT (s) 34.50 ± 5.73 35.94 ± 5.30 0.548 Fibrinogen (mg/dL) 522.97 ± 148.45 441.90 ± 170.84 0.155 Chem Total Protein (g/dL) 6.51 ± 0.56 6.13 ± 0.56 0.109 Albumin (g/dL) 2.75 ± 0.50 2.60 ± 0.59 0.89 Creatinine (mg/dL) 0.90 ± 0.52 2.40 ± 1.80 0.009 BUN (mg/dL) 23.85 ± 17.62 45.47 ± 29.80 0.046 LDH (U/L) 536.23 ± 181.92 548.80 ± 435.44 0.379 CRP (mg/dL) 11.83 ± 6.51 8.33 ± 7.12 0.123 ABG pH 7.35 ± 0.10 7.28 ± 0.16 0.379 PaO 2 (mmHg) 116.20 ± 47.61 110.39 ± 57.87 0.382 PaCO 2 (mmHg) 44.32 ± 13.83 49.83 ± 19.40 0.59 Lactate 1.55 ± 0.62 3.02 ± 3.29 0.023 Bicarbonate (mmol/L) 23.29 ± 3.52 22.06 ± 4.03 0.469 Values are presented as mean ± SD, median [IQR], or n (%), as appropriate. For between-group comparisons of survivors (n=30) and non-survivors (n=10), a two-tailed Student’s t-test was used for approximately normally distributed continuous variables or the Mann–Whitney U test otherwise; categorical variables were compared using Fisher’s exact test. The significance level was α=0.05, with no adjustment for multiple comparisons. All measurements (including arterial blood gases and lactate) were obtained at ICU admission. Available cases were used for the analysis; variable-specific sample sizes may differ where data were missing. Abbreviations: SOFA: Sequential Organ Failure Assessment; P/F ratio: PaO₂/FiO₂ ratio; CBC: complete blood count; WBC: white blood cell count; Coag: coagulation tests; PT-INR: prothrombin time - International Normalized Ratio; aPTT: activated partial thromboplastin time; Chem: blood chemistry; BUN: blood urea nitrogen; LDH: lactate dehydrogenase; CRP: C-reactive protein; ABG: arterial blood gas analysis; PaO₂: arterial oxygen pressure; PaCO₂: arterial carbon dioxide pressure. Plasma PD-L1 Levels Plasma PD-L1 levels were significantly higher in patients with COVID-19 compared to healthy controls at admission (median [IQR]: 294.15 vs. 55.88 pg/mL, p < 0.001). However, no significant difference was observed between survivors and non-survivors at admission ( p = 0.390) (Figure 1A–B). Longitudinal analysis revealed a decreasing trend in PD-L1 levels over time in both groups. Notably, PD-L1 concentrations remained higher in non-survivors throughout the observation period. Significant differences were detected on days 5 and 7 ( p = 0.023 and p = 0.001, respectively), but not on days 1, 14, and 21 (Figure 1C). These findings suggest that elevated PD-L1 levels during the early phase of infection, particularly on days 5 and 7, may be associated with poor prognosis. ＜Figure１＞ Correlation Between PD-L1 Levels and Clinical Parameters We performed a Spearman correlation analysis to explore the relationship between PD-L1 levels and various clinical and laboratory parameters at five time points (days 1, 5, 7, 14, and 21). The results are summarized as a heatmap in Figure 2, with correlation coefficients annotated in each cell. Statistically significant correlations ( p <0.05) are marked with an asterisk (*). A statistically significant inverse correlation was observed between platelet count and sPD-L1 level at admission (day 1), indicating that higher PD-L1 concentration levels were associated with lower platelet counts from the outset of critical illness. Conversely, creatinine showed a positive correlation with sPD-L1 level from day 1 through day 14, becoming statistically significant on day 7. This suggests that higher PD-L1 level parallels early renal dysfunction. Similarly, inflammatory and coagulation indices, such as C-reactive protein (CRP) and fibrinogen, both showed a positive correlation with sPD-L1 level on multiple days. This suggests that heightened sPD-L1 level reflects the combined burden of renal dysfunction, systemic inflammation, and coagulation activation throughout critical illness. ＜Figure２＞ Cox Proportional Hazards Analysis Forty-six longitudinal candidate variables were entered into separate univariable Cox models. After the Benjamini–Hochberg adjustment (FDR = 5%), six predictors remained below the prespecified threshold (p < 0.05) and were, therefore, eligible for multivariable modeling. Complete univariable screening statistics are provided in Additional file 2. We then used L1-penalized Cox regression, limited to three clinically plausible covariates: retained lactate, sPD-L1 (day 7), and creatinine (λ selected by 3-fold stratified cross-validation). The final coefficients and hazard ratios are summarized in Table 2. This penalized model achieved a concordance (Harrell’s C-index) of 0.90 and a partial Akaike information criterion of 45.4, indicating good discrimination with limited overfitting. Bootstrap optimism was < 0.02. After shrinkage, higher lactate and elevated sPD-L1 (day 7) were independently associated with an increased instantaneous risk of death (≈ 4.4-fold and 4.0-fold, respectively). Serum creatinine levels were not statistically significant after penalization. Table 2. Penalized multivariable Cox model for in-hospital mortality β (log-HR) HR 95% CI p-value (q) Lactate 1.49 4.42 1.79–10.92 < 0.005 (0.009) sPD-L1 day 7 1.38 3.98 1.47–10.76 0.010 (0.028) Creatinine −0.28 0.76 0.38–1.51 0.43 (0.61) Values include multivariable penalized Cox proportional hazards regression analysis of arterial lactate, day-7 plasma sPD-L1, and serum creatinine levels. Data shown are the regression coefficients β (log-hazard ratio), corresponding hazard ratios (HR), 95% confidence intervals, and two-sided p-values (Wald tests); q-values are p-values adjusted for multiple testing (false discovery rate). For continuous predictors, HRs represent the relative change in hazard per 1-unit increase in the predictor. Abbreviations: CI: Confidence Interval; HR: Hazard Ratio; sPD-L1, soluble programmed death-ligand 1 Risk Stratification Based on sPD-L1 and Lactate Patients were dichotomized based on the median values of soluble PD-L1 on day 7 (171 pg mL⁻¹) and arterial lactate (1.4 mmol L⁻¹). Individuals who exceeded both thresholds were classified as high-risk (n = 10), whereas all others formed the low - risk group (n = 30). Kaplan–Meier analysis showed a clear separation between the survival curves (Figure 3), and the log-rank test confirmed a statistically significant difference (χ² = 5.04, df = 1, p = 0.025). These findings reveal that higher levels of circulating sPD-L1 and lactate upon ICU admission—which reflect immune dysregulation and tissue hypoxia—can be used to identify patients with COVID-19 at a substantially higher risk of in-hospital mortality. ＜Figure３＞ ML-Based Mortality Prediction To further explore the predictive potential of clinical variables, including PD-L1 level, we trained and evaluated eight classification algorithms to predict in-hospital mortality. Model performance was assessed using stratified 5-fold cross-validation, and mean AUC values were reported. SVM showed the highest performance (mean AUC = 0.917), followed by the NN (AUC = 0.883) and LDA (AUC = 0.733). In contrast, the DT and GBM showed lower predictive accuracy (Figure 4A). SHAP analysis, conducted using the SVM model, revealed that PD-L1 levels on day 5 were the most significant predictor of mortality, followed by lactate dehydrogenase level, platelet count, and fibrinogen level (Figure 4B). The beeswarm plot (Figure 4C) confirmed that higher PD-L1 values were associated with higher predicted mortality, whereas higher platelet and fibrinogen levels were associated with decreased risk. ＜Figure４＞ Discussion In our cohort of patients with COVID-19 who were critically ill, circulating sPD-L1 generally declined during the first 3 weeks of hospitalization, yet remained persistently elevated in non-survivors. sPD-L1 measurements taken on days 5 and 7 distinguished survivors from non-survivors, indicating that sustained elevation of sPD-L1 is associated with adverse outcomes. Multivariable Cox analysis further identified day-7 sPD-L1 and lactate as independent predictors of in-hospital mortality; the prognostic value of lactate aligns with previous reports [15– 17]. Day 7 sPD-L1 correlated significantly with creatinine, fibrinogen, and C-reactive protein, suggesting that renal dysfunction, coagulopathy, and systemic inflammation are part of the same pathophysiological axis. These findings suggest that sPD-L1 could be a valuable addition to established prognostic indices. Complementary evidence was obtained from the SVM model, which incorporated day 1 and 7 sPD-L1 levels, along with other routine variables, including age, SOFA score, and CRP level. This model achieved excellent discrimination (AUC > 0.91), and based on SHAP attribution, the day 7 sPD-L1 level was ranked as the dominant predictor of mortality. The difference in importance ranking between the Cox and SVM models likely reflects variance inflation because of multicollinearity, as well as the capacity of non-linear kernels to capture complex interactions that linear proportional hazards models cannot [18, 19]. Overall, our findings support prior studies that have described elevated sPD-L1 in severe COVID-19 cases [11, 20]. Critically, we expand upon these observations by demonstrating that the temporal trajectory of sPD-L1 is directly tied to the prognosis of a patient when evaluated by both classical regression and modern ML methods. The PD-1/PD-L1 axis acts as an immune checkpoint that helps control T-cell activation during inflammatory stress [7]. In COVID-19, sPD-L1 facilitates viral immune evasion during the early phase of infection, ultimately dampening an overexuberant host response [21]. Therefore, the sustained elevation of sPD-L1 levels observed on ICU day 7 likely signals a shift toward the late, immunosuppressive phase of the disease, characterized by T-cell exhaustion. This persistent PD-L1 expression could be driven by a late surge of interferon-γ, interleukin-6, and hypoxia-inducible factors, which upregulate PD-L1 through signal transducer and activator of transcription 1/ interferon regulatory factor 1 or hypoxia-inducible factor-1α pathways [22, 23]. Such cytokine- and hypoxia-dependent regulation is consistent with the positive correlation we observed between Day 7 sPD-L1, lactate, and CRP levels, which are markers of systemic inflammation and tissue hypoxia. A similar phenomenon is well documented in bacterial sepsis, where persistent PD-1/PD-L1 signaling drives prolonged immune paralysis, and PD-L1-deficient mice exhibit improved survival [24]. PD-1/PD-L1 blockade has yielded favorable outcomes in preclinical sepsis models; however, its therapeutic benefit in clinical trials has yet to be demonstrated [24]. Our observational study provides pathophysiological evidence that sustained PD-L1 upregulation accompanies critical illness because of viral infection. These findings suggest that if checkpoint inhibition is carried out therapeutically, the timing of administration may be crucial to clinical efficacy. sPD-L1 shows promise as a clinically useful prognostic marker. Rapid assessment on ICU admission and again on day 7 could enable early identification of patients with COVID-19 who are critically ill and at heightened risk of death, thereby allowing for timely escalation or modulation of immunotherapeutic strategies. In our study, adding sPD-L1 level to an ML model that is based on routine blood tests significantly improved its predictive accuracy, highlighting its added value in multivariable risk scoring. Expanding this approach to include other soluble immune checkpoint molecules such as sCD40, soluble T-cell immunoglobulin and mucin domain (TIM) 1, and galectin-9, which have also been linked to disease severity [20, 25], could facilitate more granular immune profiling and guide personalized treatment algorithms in future studies. This study has some limitations. First, our study was a small, single-center, observational cohort. Consequently, the generalizability of our findings and any inference regarding therapeutic impact are restricted. Second, our the focus of our analyses was solely on sPD-L1. A more comprehensive evaluation of host immune dysfunction would require the simultaneous quantification of additional immune checkpoint mediators, such as PD-1, TIM 3, and Lymphocyte Activation Gene 3, coupled with functional T-cell assays. Third, since randomized trials have not yet shown unequivocal benefit from PD-1/PD-L1 blockade in sepsis [24], the clinical utility of targeting this axis in severe COVID-19 remains speculative and requires rigorous testing. Validation in larger, multicenter, longitudinal cohorts, together with broad-spectrum immune profiling, is required to confirm our observations and determine their translational relevance. Conclusion In this study, we found that persistently elevated plasma PD-L1 levels are associated with poor prognosis in patients with COVID-19 who are critically ill. This suggests that PD-L1–mediated immune suppression could contribute to disease progression and mortality in severe cases. PD-L1 and lactate levels were used for risk stratification to effectively predict in-hospital mortality. ML models further demonstrated the clinical utility of these markers. PD-L1 may serve as both a prognostic biomarker and a potential therapeutic target, and further studies are warranted to validate its role in the immunopathogenesis of COVID-19. Abbreviations ABG: Arterial Blood Gas aPTT: Activated Partial Thromboplastin Time AUC: Area Under the Curve BUN: Blood Urea Nitrogen- CBC: Complete Blood Count CI: Confidence Interval Coag: Coagulation tests COVID-19: Coronavirus Disease 2019 CRP: C-reactive protein DT: Decision tree eGFR: Estimated Glomerular Filtration Rate FiO₂: Fraction of inspired oxygen GBM: Gradient Boosting Machine HR: Hazard Ratio ICU: Intensive Care Unit IQR: Interquartile Range LDA: Linear Discriminant Analysis LR: Logistic Regression ML: Machine Learning NN: Neural Network PaCO₂: Arterial partial pressure of carbon dioxide PaO₂: Arterial partial pressure of oxygen P/F ratio: PaO₂ / FiO₂ ratio PD-1: Programmed Death 1 PD-L1: Programmed Death-Ligand 1 PT-INR: Prothrombin Time–International Normalized Ratio RF: Random Forest ROC: Receiver Operating Characteristic SARS-CoV-2: Severe Acute Respiratory Syndrome Coronavirus 2 SHAP: Shapley Additive Explanations SOFA: Sequential Organ Failure Assessment sPD-L1: soluble Programmed Death-Ligand 1 SVM: Support Vector Machine WBC: White Blood Cell Declarations Ethics approval and consent to participate The study was approved by the Institutional Review Board of Mie University Hospital (approval numbers: IRB No. 3026 and H2021-191). Written informed consent was obtained from the patients or their next of kin upon hospital admission. Consent for publication Not applicable. Availability of data and materials Data will be shared upon reasonable request and institutional approval. Competing Interest The authors declare that they have no conflicts of interest. Author Contributions Study conception and design, Data acquisition, Data analysis: ST, EK; Data interpretation: ST, EK, DO, YS, TM, RK; Manuscript drafting and revising: ST, EK, TM, DO, YS, AG, EP, MS, and RK. All authors read and approved the final manuscript. Funding This work was supported by JSPS KAKENHI Grant Number JP25K12167, JP24K02546, JP22K06910, and JP21K09015. Acknowledgements Not applicable. References Armstrong RA, Kane AD, Kursumovic E, Oglesby FC, Cook TM. Mortality in patients admitted to intensive care with COVID‐19: An updated systematic review and meta‐analysis of observational studies. Anaesthesia. 2021;76:537. Hu B, Huang S, Yin L. The cytokine storm and COVID-19. J Med Virol. 2021;93:250-6. Del Valle DM, Kim-Schulze S, Huang HH, Beckmann ND, Nirenberg S, Wang B, et al. An inflammatory cytokine signature predicts COVID-19 severity and survival. Nat Med. 2020;26:1636-43. Alahdal M, Elkord E. Exhaustion and over-activation of immune cells in COVID-19: Challenges and therapeutic opportunities. Clin Immunol. 2022;245:109177. Gao X, Cai S, Li X, Wu G. Sepsis-induced immunosuppression: Mechanisms, biomarkers and immunotherapy. Front Immunol. 2025;16:1577105. Zhou F, Yu T, Du R, Fan G, Liu Y, Liu Z, et al. Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study. Lancet. 2020;395:1054-62. Freeman GJ, Long AJ, Iwai Y, Bourque K, Chernova T, Nishimura H, et al. Engagement of the Pd-1 immunoinhibitory receptor by a novel B7 family member leads to negative regulation of lymphocyte activation. J Exp Med. 2000;192:1027-34. Barber DL, Wherry EJ, Masopust D, Zhu B, Allison JP, Sharpe AH, et al. Restoring function in exhausted CD8 T cells during chronic viral infection. Nature. 2006;439:682-7. Yang L, Gao Q, Li Q, Guo S. PD-L1 blockade improves survival in sepsis by reversing monocyte dysfunction and immune disorder. Inflammation. 2024;47:114-28. Sabbatino F, Conti V, Franci G, Sellitto C, Manzo V, Pagliano P, et al. PD-L1 dysregulation in COVID-19 patients. Front Immunol. 2021;12:695242. Beserra DR, Alberca RW, Branco ACCC, De Mendonça Oliveira L, De Souza Andrade MM, Gozzi-Silva SC, et al. Upregulation of PD-1 expression and high sPD-L1 levels associated with COVID-19 severity. J Immunol Res. 2022;2022:9764002. Schniederova M, Bobcakova A, Grendar M, Markocsy A, Ceres A, Cibulka M, et al. Lymphocyte inhibition mechanisms and immune checkpoints in COVID-19: Insights into prognostic markers and disease severity. Medicina (Kaunas). 2025;61:189. Ogundimu EO, Altman DG, Collins GS. Adequate sample size for developing prediction models is not simply related to events per variable. J Clin Epidemiol. 2016;76:175-82. van Smeden M, Moons KGM, de Groot JAH, Collins GS, Altman DG, Eijkemans MJC, et al. Sample size for binary logistic prediction models: Beyond events per variable criteria. Stat Methods Med Res. 2019;28:2455-74. Carpenè G, Onorato D, Nocini R, Fortunato G, Rizk JG, Henry BM, et al. Blood lactate concentration in COVID-19: A systematic literature review. Clin Chem Lab Med. 2022;60:332-7. Birben B, Birben OD, Akın T, Akkurt G, Surel AA, Yakısık E, et al. Efficacy of the delta neutrophil index in predicting 30-day mortality in COVID-19 patients requiring intensive care. Int J Clin Pract. 2021;75:e13970. Alharthy A, Aletreby W, Faqihi F, Balhamar A, Alaklobi F, Alanezi K, et al. Clinical characteristics and predictors of 28-day mortality in 352 critically ill patients with COVID-19: A retrospective study. J Epidemiol Glob Health. 2021;11:98-104. Saarela M, Jauhiainen S. Comparison of feature importance measures as explanations for classification models. SN Appl Sci. 2021;3:1-12. Noble WS. What is a support vector machine? Nat Biotechnol. 2006;24:1565-7. Avendaño‐Ortiz J, Lozano‐Rodríguez R, Martín‐Quirós A, Terrón V, Maroun‐Eid C, Montalbán‐Hernández K, et al. The immune checkpoints storm in COVID‐19: Role as severity markers at emergency department admission. Clin Transl Med. 2021;11:e573. Sabbatino F, Pagliano P, Sellitto C, Stefanelli B, Corbi G, Manzo V, et al. Different prognostic role of soluble PD-L1 in the course of severe and non-severe COVID-19. J Clin Med. 2023;12:6812. Garcia-Diaz A, Shin DS, Moreno BH, Saco J, Escuin-Ordinas H, Rodriguez GA, et al. Interferon receptor signaling pathways regulating PD-L1 and PD-L2 expression. Cell Rep. 2017;19:1189-201. Noman MZ, Desantis G, Janji B, Hasmim M, Karray S, Dessen P, et al. PD-L1 is a novel direct target of HIF-1α, and its blockade under hypoxia enhanced: MDSC-mediated T cell activation. J Exp Med. 2014;211:781-90. Zhang T, Yu-jing L, Ma T. Role of regulation of PD-1 and PD-L1 expression in sepsis. Front Immunol. 2023;14:1029438. Paranga TG, Pavel-Tanasa M, Constantinescu D, Iftimi E, Plesca CE, Miftode IL, et al. Distinct soluble immune checkpoint profiles characterize COVID-19 severity, mortality and SARS-CoV-2 variant infections. Front Immunol. 2024;15:1464480. Additional Declarations No competing interests reported. Supplementary Files SupplementaryTableS1.xlsx SupplementarytableS2.xlsx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 30 Sep, 2025 Reviews received at journal 30 Sep, 2025 Reviews received at journal 30 Sep, 2025 Reviews received at journal 24 Sep, 2025 Reviewers agreed at journal 21 Sep, 2025 Reviewers agreed at journal 19 Sep, 2025 Reviewers agreed at journal 19 Sep, 2025 Reviewers agreed at journal 19 Sep, 2025 Reviewers invited by journal 19 Sep, 2025 Editor assigned by journal 19 Sep, 2025 Submission checks completed at journal 18 Sep, 2025 First submitted to journal 18 Sep, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7521856","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":519987725,"identity":"8ca8d387-157d-4bc0-a1c2-762daf8bf439","order_by":0,"name":"Shungo Takeuchi","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABCUlEQVRIiWNgGAWjYHACZhCRwMbewMDAA0ZIwlgBG0wLzwFStTBIJCCpxwfk5zc/NvjYZpfHJ/k68cEbhm0y/Aw8Bgw/ahjYzXFoMTjGZpw4sy25mE06d7PhHIbbPJINPAaMPccYmC0bcGhhYzA+zNt2ILFNOnebNA9Qi8H9NwYMvA0MzAYHcDisjf3z4b8gLZJnt/8GabE/ALTlLx4tDMd4jJMZQVokeLcxg20B+oUZny0Gx3KKDXvOJSe28eRulpxjcJtH4gBbwWGZYxI4/SLffHyzxI8yu8T57Wc3fnhTcduev4F548M3NTbJuEIM3VIIBXSSRLIBcVqQgB3pWkbBKBgFo2CYAgAmzFDPepD4ogAAAABJRU5ErkJggg==","orcid":"","institution":"Mie University Graduate School of Medicine","correspondingAuthor":true,"prefix":"","firstName":"Shungo","middleName":"","lastName":"Takeuchi","suffix":""},{"id":519987726,"identity":"66a7c9e4-8542-4aa0-83b5-a684cfdfc06a","order_by":1,"name":"Eiji Kawamoto","email":"","orcid":"","institution":"Mie University Graduate School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Eiji","middleName":"","lastName":"Kawamoto","suffix":""},{"id":519987727,"identity":"a8fa7347-980d-446d-8a8e-ad686a139d86","order_by":2,"name":"Takashi Matsusaki","email":"","orcid":"","institution":"Mie University Graduate School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Takashi","middleName":"","lastName":"Matsusaki","suffix":""},{"id":519987728,"identity":"eccd4098-feda-43b5-b0f7-36a4a3dc4ed7","order_by":3,"name":"Daisuke Ono","email":"","orcid":"","institution":"Mie University Graduate School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Daisuke","middleName":"","lastName":"Ono","suffix":""},{"id":519987729,"identity":"d77eeef1-1c99-4e98-8c69-58680b58f2e0","order_by":4,"name":"Yosuke Sakakura","email":"","orcid":"","institution":"Mie University Graduate School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yosuke","middleName":"","lastName":"Sakakura","suffix":""},{"id":519987730,"identity":"9ec53baf-cec0-4042-99e3-cb43f62bfc0d","order_by":5,"name":"Arong Gaowa","email":"","orcid":"","institution":"Mie University Graduate School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Arong","middleName":"","lastName":"Gaowa","suffix":""},{"id":519987731,"identity":"26f4f251-d40b-4c8c-9b7e-88da2f754b68","order_by":6,"name":"Eun Jeong Park","email":"","orcid":"","institution":"Mie University Graduate School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Eun","middleName":"Jeong","lastName":"Park","suffix":""},{"id":519987732,"identity":"5d543f5e-ced8-4f0b-9468-cee91c8c72f1","order_by":7,"name":"Motomu Shimaoka","email":"","orcid":"","institution":"Mie University Graduate School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Motomu","middleName":"","lastName":"Shimaoka","suffix":""},{"id":519987733,"identity":"12fbe84c-421b-4d74-bdbc-acce48a8165b","order_by":8,"name":"Ryuji Kaku","email":"","orcid":"","institution":"Mie University Graduate School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Ryuji","middleName":"","lastName":"Kaku","suffix":""}],"badges":[],"createdAt":"2025-09-03 02:08:16","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7521856/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7521856/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":92680344,"identity":"2eff64de-cb87-4567-b4b6-0c24cfc3b778","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4883109,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript002.docx","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/9e32429371ff378b74a026d5.docx"},{"id":92680328,"identity":"de2f9d14-39d4-4576-8dc3-ad693adbd78f","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10499,"visible":true,"origin":"","legend":"","description":"","filename":"b86af7b3acce4ca296072f465bdd77ec.json","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/70015561221f2124bdbbf26c.json"},{"id":92682338,"identity":"ae9191ce-4502-48d1-a597-ee338bcf1908","added_by":"auto","created_at":"2025-10-03 01:12:54","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15415,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementarytableS2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/3e1a263f82003fed68b56b16.xlsx"},{"id":92680334,"identity":"c061c8ef-796b-4e53-87a9-dc8e65d93f41","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9320,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/2c3b7db72de5dd694ab9876a.xlsx"},{"id":92680337,"identity":"2bd3e223-e2d4-43f7-8725-47d1ed5209f3","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"xml","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":101361,"visible":true,"origin":"","legend":"","description":"","filename":"b86af7b3acce4ca296072f465bdd77ec1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/51688f4f64864affa3285f44.xml"},{"id":92680342,"identity":"3dbd9f29-ee94-4c63-9524-f7b6c1a31725","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"emf","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5456364,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.emf","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/17d3d87dde947e39654fcfd4.emf"},{"id":92680346,"identity":"ce8792ed-e958-4e6f-9c64-ad05e4477311","added_by":"auto","created_at":"2025-10-03 01:04:55","extension":"emf","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6487480,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.emf","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/b917a978ffb5d040ff5b429c.emf"},{"id":92680345,"identity":"3a147502-17db-4dc5-b134-ff5cfcd16e72","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"emf","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6343080,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.emf","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/5a272c5164079a06b22173ee.emf"},{"id":92680347,"identity":"6ed6f962-41bd-4000-9e89-d89d7b4c6a6e","added_by":"auto","created_at":"2025-10-03 01:04:55","extension":"emf","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":22873868,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.emf","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/c090ab7e13e30a3f32a1934b.emf"},{"id":92682683,"identity":"6ff79d47-2818-4138-8b17-0bd01eb21ade","added_by":"auto","created_at":"2025-10-03 01:20:54","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18645,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/a1a7c26537b26b62fddcfcff.png"},{"id":92680341,"identity":"c92a4482-a4ec-41f4-a4f4-fcd1c7ae0f7a","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":12157,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/b847135c080d1d141b7d4d30.png"},{"id":92680338,"identity":"bcaff6d2-423b-486d-b727-1dd857f966d8","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6781,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/55279b095907b88246a7efb9.png"},{"id":92680339,"identity":"3bd1cdd5-5973-475c-a360-74a3c1c4e6ec","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":23035,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/af9763d37208d6581f3ab825.png"},{"id":92680340,"identity":"5f8b50f2-2435-4e16-942f-b354cfc5b226","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"xml","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":101095,"visible":true,"origin":"","legend":"","description":"","filename":"b86af7b3acce4ca296072f465bdd77ec1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/df0329b888196939675428a9.xml"},{"id":92680343,"identity":"96d571dc-070b-4205-a100-9fc0576f7cf5","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"html","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":111912,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/3fce63fc71341edd341cb9db.html"},{"id":92680326,"identity":"92397e82-6feb-4589-84d4-13a614198e69","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":18645,"visible":true,"origin":"","legend":"\u003cp\u003ePlasma sPD-L1 at Admission and Longitudinally in COVID-19\u003cbr\u003e\n(A) Comparison of plasma Programmed Death-Ligand 1 (PD-L1) levels at admission between healthy volunteers and the overall patient group. (B) Plasma PD-L1 levels at admission in healthy volunteers, Coronavirus Disease 2019 survivors, and non-survivors. (C) Longitudinal soluble (s)PD-L1 in survivors vs non-survivors on days 1, 5, 7, 14, and 21 after admission; group sizes (survivors/non-survivors): day 1, 30/10; day 5, 23/8; day 7, 21/8; day 14, 9/5; and day 21, 3/2.\u003cbr\u003e\nBoxes show interquartile range (IQR); center line represents the median; whiskers indicate 1.5×IQR; points denote individuals. Two-tailed Mann–Whitney U tests were used for between-group comparisons at each time point; p\u0026lt;0.05 is considered significant (asterisk indicates p\u0026lt;0.05).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations:\u003c/em\u003e PD-L1, Programmed Death-Ligand 1; IQR, interquartile range\u003c/p\u003e","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/4e4378db254f123e99268b67.png"},{"id":92680327,"identity":"5c0371ed-34d9-4d14-8d14-1e6caf92ff32","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":12157,"visible":true,"origin":"","legend":"\u003cp\u003eSpearman Correlations Between Plasma sPD-L1 Levels and Clinical Variables in Patients with Severe COVID-19 \u003cbr\u003e\nThe heatmap shows Spearman’s rank correlation coefficients (ρ) between Soluble Programmed Death-Ligand 1 concentrations and selected clinical parameters measured on intensive care unit days 1, 5, 7, 14, and 21. Warmer colors indicate positive correlations, and cooler colors represent negative correlations. Numerical ρ values are displayed in each cell. Asterisks (*) indicate significant correlations after two-tailed testing (p \u0026lt; 0.05). Analyses used available samples at each time point.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations:\u003c/em\u003e eGFR: Estimated Glomerular Filtration Rate; AST: Aspartate Aminotransferase; ALT: Alanine Aminotransferase; ALP: Alkaline Phosphatase; CPK: Creatine Phosphokinase; RBC: Red Blood Cell Count; sPD-L1, Soluble Programmed Death-Ligand 1\u003c/p\u003e","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/1d0f1ba05be0c7ab032cc1d9.png"},{"id":92680331,"identity":"8ff6c29d-8552-4842-9d32-4e382488d3e6","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":6781,"visible":true,"origin":"","legend":"\u003cp\u003eKaplan–Meier Survival Curves Stratified by Combined Day-7 sPD-L1 and Arterial Lactate Risk Groups in Severe COVID-19 \u003cbr\u003e\nSurvival probability is shown for the high-risk (orange) and low-risk (blue) cohorts identified in the study population (N = 40). Patients were classified as high-risk when plasma Soluble Programmed Death-Ligand 1 on intensive care unit day 7 exceeded the median (171 pg mL⁻¹) and their arterial lactate exceeded the median (1.4 mmol L⁻¹). All others were categorized as low risk. Censored observations are marked by vertical ticks. The survival difference between groups was significant using the log-rank test (χ² = 5.04, df = 1, p = 0.025). Abbreviation: sPD-L1: Soluble Programmed Death-Ligand 1\u003c/p\u003e","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/b123ccaa4a7d2cb03a4377cb.png"},{"id":92682337,"identity":"ed611ba2-1921-48d6-b170-a1a05725535e","added_by":"auto","created_at":"2025-10-03 01:12:54","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":23035,"visible":true,"origin":"","legend":"\u003cp\u003eMachine Learning performance and SHAP Attribution for Mortality\u003cbr\u003e\n(A) Receiver operating characteristic (ROC) curves generated using stratified 5-fold cross-validation for support vector machine (SVM), neural network (NN), decision tree (DT), AdaBoost, gradient boosting machine (GBM), linear discriminant analysis (LDA), logistic regression (LR), and random forest (RF). The mean area under the curve (AUC)-ROC for each model is displayed in the inset; SVM achieved the highest discrimination (AUC = 0.917). The diagonal dashed line indicates no discrimination.\u003c/p\u003e\n\u003cp\u003e(B) Global feature importance ranked by the mean absolute SHAP values for the best SVM classifier. Day-5 sPD-L1 was the most significant predictor, followed by lactate dehydrogenase, platelet count, and fibrinogen.\u003c/p\u003e\n\u003cp\u003e(C) Beeswarm SHAP plot illustrating the direction and magnitude of impact of each feature on the SVM mortality prediction for individual patients. Positive SHAP values (rightward shift) increase the risk of death, whereas negative values (leftward shift) decrease it. Continuous variables are color-coded from low (blue) to high (red) values.\u003c/p\u003e\n\u003cp\u003eAbbreviations: AUC, area under the curve; CRP, C-reactive protein; DT, decision tree; FiO₂, fraction of inspired oxygen; GBM, gradient boosting machine; LDA, linear discriminant analysis; LDH, lactate dehydrogenase; LR, logistic regression; NN, neural network; P/F, PaO₂/FiO₂ ratio; RF, random forest, ROC, Receiver operating characteristic; SHAP, Shapley Additive Explanation; ). sPD-L1, Soluble Programmed Death-Ligand 1; SVM, support vector machine\u003c/p\u003e","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/78e47e3e25c3ca56e9baf9c2.png"},{"id":92683021,"identity":"a277a939-3c91-4454-bd47-0ab943e12eed","added_by":"auto","created_at":"2025-10-03 01:28:54","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":853094,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/9d78a636-d245-4523-82f4-0f98f434f5f1.pdf"},{"id":92680330,"identity":"8673786c-0be9-43fa-9780-3ea3ff8f76eb","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":9320,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/307f298fc161e4135909dfd3.xlsx"},{"id":92680335,"identity":"49b3f962-b5c8-4dde-aa39-651eb1812286","added_by":"auto","created_at":"2025-10-03 01:04:54","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":15415,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementarytableS2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7521856/v1/93338bbae695a7e2307d7a58.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Use of Dynamic Soluble Programmed Death-Ligand 1 Trajectories to Identify Early Organ Dysfunction and Predict Mortality in Critical Coronavirus Disease 2019","fulltext":[{"header":"Background","content":"\u003cp\u003eSevere Coronavirus Disease 2019 (COVID-19) requiring intensive care is associated with high mortality, with intensive care unit (ICU) mortality rates for patients with COVID-19 who are critically ill averaging 30\u0026ndash;40% across various studies and settings [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. An early \u0026ldquo;cytokine storm\u0026rdquo; hyperinflammatory phase of COVID-19 has been well recognized [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]; however, there is growing evidence that many patients with critical illness also develop a state of immune exhaustion or secondary immunosuppression [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Marked lymphopenia and T-cell dysfunction in severe COVID-19 correlate with worse outcomes, mirroring the \u0026ldquo;immunoparalysis\u0026rdquo; observed in septic shock [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eProgrammed death-ligand 1 (PD-L1), a pivotal immune checkpoint molecule, downregulates immune responses by engaging PD-1 on T cells, leading to their functional exhaustion [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. In sepsis, increased PD-L1 expression is implicated in monocyte dysfunction, impaired cytokine production, and poor clinical outcomes [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Circulating or soluble PD-L1 (sPD-L1) has emerged as a potential biomarker of immune suppression in patients with critical illness. Preclinical studies show that blocking this pathway can improve survival [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. These findings position PD-L1 as a mechanistic marker of immune dysfunction as well as a potential prognostic indicator for critical illness.\u003c/p\u003e\u003cp\u003eRecent evidence suggests that the programmed death 1 (PD-1)/PD-L1 pathway is also dysregulated in severe COVID-19 cases [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Beserra et al. demonstrated significantly higher sPD-L1 concentrations in hospitalized patients with COVID-19 than in healthy controls, suggesting its role in the pathophysiology of severe infection [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Other studies have identified a \u0026ldquo;storm\u0026rdquo; of soluble immune checkpoints, including PD-L1, which correlates with disease severity [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. These data show that PD-L1 may contribute to COVID-19-associated immune dysregulation, consistent with observations in bacterial sepsis.\u003c/p\u003e\u003cp\u003eCurrently, the lack of longitudinal data on PD-L1 dynamics during critical illness is a significant gap in the literature. In most studies, PD-L1 (or sPD-L1) has been assessed at a single time point, such as upon hospital admission, providing a limited snapshot of the disease. The evolution of PD-L1 levels during intensive care remains unclear, and it is unknown whether these changes correlate with clinical outcomes. Dynamic monitoring of immune checkpoints, such as PD-1/PD-L1, has been proposed for prognosis and immunotherapy in sepsis [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]; however, longitudinal studies remain rare. Temporal data on PD-L1 are particularly scarce in COVID-19. One study reported no significant change in sPD-L1 levels over time after Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) infection, but only a few time points were analyzed [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Overall, it remains uncertain whether PD-L1 levels remain elevated, fluctuate, or decline during ICU stay. Understanding these patterns could help identify transitions between hyperinflammatory and immunosuppressive phases, guiding the optimal timing of immunomodulatory interventions.\u003c/p\u003e\u003cp\u003eTo address this gap, our study was designed to characterize the longitudinal trajectory of PD-L1 levels in patients with critical illness owing to COVID-19 and evaluate the prognostic significance of these time-dependent changes. We hypothesized that patients who died would show persistently elevated or increasing PD-L1 levels, whereas survivors would show declining levels. By analyzing the temporal dynamics of PD-L1 levels in this study, we aimed to provide novel insights into the host immune response in severe COVID-19 and determine whether PD-L1 could serve as a dynamic prognostic biomarker in critical care.\u003c/p\u003e"},{"header":"Methods","content":"\n\u003ch3\u003e1. Study Design and Participants\u003c/h3\u003e\n\u003cp\u003eThis single-center, retrospective observational study was conducted at Mie University Hospital between April 2021 and December 2022. We included 40 consecutive adult patients who were hospitalized with laboratory-confirmed COVID-19\u0026mdash;defined by a positive reverse-transcription polymerase chain reaction assay for SARS-CoV-2\u0026mdash;in the analysis. We also recruited 23 age- and sex-matched healthy volunteers to serve as the control group. Patients were excluded if key variables were missing or if informed consent could not be obtained. The study protocol was approved by the Institutional Review Board of Mie University Hospital (approval numbers: IRB No. 3026 and H2021-191). We obtained written informed consent from the patients or their next of kin upon admission to the hospital.\u003c/p\u003e\n\u003ch3\u003e2. Data Collection and Measurements\u003c/h3\u003e\n\u003cp\u003eWe collected demographic data, comorbidities, and laboratory values from the electronic medical records. Plasma samples were collected within 24 h of hospital admission (day 1) and subsequently on days 5, 7, 14, and 21 of hospitalization. Serum samples were obtained via EDTA-containing tubes, which were stored at \u0026minus;\u0026thinsp;80\u0026deg;C until use. PD-L1 levels were measured using the Human PD-L1 SimpleStep ELISA Kit (28\u0026thinsp;\u0026minus;\u0026thinsp;8 clone; Abcam, ab277712, Cambridge, UK) according to the manufacturer\u0026rsquo;s instructions. We also routinely performed blood tests\u0026mdash;including complete blood counts, coagulation studies, and biochemical assays\u0026mdash;to monitor the clinical status and severity of the disease. The primary outcome was in-hospital mortality.\u003c/p\u003e\n\u003ch3\u003e3. Statistical Analysis\u003c/h3\u003e\n\u003cp\u003eContinuous variables were expressed as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation (SD) or median interquartile range (IQR) and compared using Student\u0026rsquo;s t-test or Mann\u0026ndash;Whitney U test. Categorical variables were analyzed with Fisher\u0026rsquo;s exact test.\u003c/p\u003e\u003cp\u003eKaplan\u0026ndash;Meier survival analysis, along with a log-rank test, was used to compare mortality between groups and assess statistical significance. The Cox proportional hazards regression model was used to identify predictors of mortality. Furthermore, it was implemented using the lifelines Python package to incorporate longitudinal data. All 46 candidate predictors were first analyzed using separate univariable Cox models. We subsequently adjusted p-values for multiplicity using the Benjamini\u0026ndash;Hochberg procedure, controlling the false discovery rate (FDR) at 5%. Predictors with an FDR-adjusted q-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05, along with prespecified clinically essential covariates, were carried forward into the multivariable analysis.\u003c/p\u003e\u003cp\u003eTo prevent overfitting in this small data set (10 events), we: (i) restricted the final model to three clinically supported variables (event-per-variable ratio\u0026thinsp;\u0026asymp;\u0026thinsp;3.3); (ii) applied L1-penalized (LASSO) Cox regression with λ chosen by stratified 3-fold cross-validation to shrink the coefficients; and (iii) estimated model optimism using the same cross-validation scheme. Simulation studies suggest that under strong penalization, acceptable bias and calibration can be maintained with event-to-variable ratios as low as 5 [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Therefore, the resulting model should be considered hypothesis-generating, and its stability will require confirmation in larger, independent cohorts.\u003c/p\u003e\n\u003ch3\u003e4. Machine Learning Models\u003c/h3\u003e\n\u003cp\u003eMachine learning (ML) modeling was performed in Python 3.10 using scikit-learn v1.4 and Shapley Additive Explanations (SHAP) v0.45.0. The analysis included all admission-time laboratory and clinical variables, along with serial sPD-L1 concentrations (days 1, 5, 7, 14, and 21). We retained variables with \u0026le;\u0026thinsp;25% missing data, which were replaced with the column mean. Non-informative identifiers and the binary outcome label were excluded from the predictor matrix. Continuous predictors were z-transformed using StandardScaler, with scaling parameters estimated on each training fold and subsequently applied to the corresponding test fold to prevent data leakage.\u003c/p\u003e\u003cp\u003eWe evaluated eight supervised classifiers: support vector machine (SVM), neural network, decision tree (DT), AdaBoost, gradient boosting machine (GBM), linear discriminant analysis (LDA), logistic regression, and random forest (RF). Hyperparameters for SVM, neural network (NN), decision tree (DT), and RF were tuned using a grid search (Additional file 1).\u003c/p\u003e\u003cp\u003eModel development followed a stratified 5-fold cross-validation scheme. Within each training fold, an inner grid search identified the optimal hyperparameter set (scoring\u0026thinsp;=\u0026thinsp;area under the receiver operating characteristic [ROC] curve [AUC]). The best estimator was then refitted on the full training fold and evaluated on the held-out test fold. We used the predicted class probabilities to construct ROC curves and to calculate the AUC for each fold. True positive rates were interpolated at 100 equally spaced false positive rate points, and mean ROC curves were generated across folds for every algorithm.\u003c/p\u003e\u003cp\u003eModel interpretability was assessed using SHAP. KernelExplainer was fitted on the training data, and global feature importance was ranked using the mean absolute SHAP values. All computations were executed with a fixed random seed (random_state\u0026thinsp;=\u0026thinsp;42).\u003c/p\u003e\n\u003ch3\u003e5. Software and Tools\u003c/h3\u003e\n\u003cp\u003eAll statistical and ML analyses were performed using Python version 3.11. The following libraries were used: lifelines (for survival modeling) and scikit-learn (for ML model training). Data preprocessing and visualization were conducted using pandas, NumPy, Matplotlib, and Seaborn.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003ePatient Characteristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA total of 40 patients with severe COVID-19 who were admitted to Mie University Hospital were included in the study. Ten of these patients (25%) died during hospitalization.\u003c/p\u003e\n\u003cp\u003eTable\u0026nbsp;1 shows a summary of the baseline characteristics and laboratory data at admission, stratified by survival status. The mean age was 59.8 ± 14.3 years in survivors and 64.3 ± 14.4 years in non-survivors (p = 0.109).\u003c/p\u003e\n\u003cp\u003eNon-survivors had significantly higher Sequential Organ Failure Assessment (SOFA) scores (7.40 ± 3.60 vs. 3.57 ± 1.77; p \u0026lt; 0.001) and lower platelet counts (182.7 ± 102.7 vs. 249.1 ± 107.6 ×10³/µL; p = 0.036) than their counterparts. Creatinine and blood urea nitrogen (BUN) levels were also significantly elevated in the non-survivor group (Creatinine: 2.40 ± 1.80 vs. 0.90 ± 0.52 mg/dL, p = 0.009; BUN: 45.5 ± 29.8 vs. 23.9 ± 17.6 mg/dL, p = 0.046), suggesting impaired renal function.\u003c/p\u003e\n\u003cp\u003eOther variables, including Glasgow Coma Scale score, PaO\u003csub\u003e2\u003c/sub\u003e/FiO\u003csub\u003e2\u003c/sub\u003e(P/F) ratio, inflammatory markers, coagulation parameters, and arterial blood gas values, showed no significant differences between groups.\u0026nbsp;\u003c/p\u003e\n\u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 1\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eBaseline characteristics of patients with COVID-19\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eSurvivors\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eNon-survivors\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003en = 30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003en = 10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ep-value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eAge, years\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e59.78 ± 14.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e64.29 ± 14.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.109\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003eSex (Female/Male)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8/22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1/9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.404\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003eBody Mass Index, kg/m\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e30.00 ± 7.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e27.67 ± 5.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.739\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003eSOFA score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e3.57 ± 1.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e7.40 ± 3.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003eGlasgow Coma Scale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e13.40 ± 3.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e11.50 ± 5.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.396\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eP/F ratio\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e202.36 ± 112.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e175.01 ± 58.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.612\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003eComorbidities\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eHypertension\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eDiabetes mellites\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.482\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eKidney disorder\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.042\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eRespiratory disorder\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.404\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eCardiovascular disorder\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.338\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eLiver disorder\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.058\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eImmune disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eHemoglobin(g/dL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e13.41 ± 1.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e12.84 ± 2.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.656\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eWBC (×10³/µL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e9.32 ± 5.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e7.75 ± 4.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.301\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eLymphocyte (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e9.42 ± 6.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e13.26 ± 10.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.48\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eNeutrophil (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e84.98 ± 7.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e78.58 ± 13.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003ePlatelet (×10³/µL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e249.10 ± 107.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e182.67 ± 102.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.036\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCoag\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eD-dimer (µg/mL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e5.31 ± 8.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e28.26 ± 74.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.301\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003ePT-INR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e1.17 ± 0.26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e1.18 ± 0.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.818\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eAPTT (s)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e34.50 ± 5.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e35.94 ± 5.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.548\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eFibrinogen (mg/dL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e522.97 ± 148.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e441.90 ± 170.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.155\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChem\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eTotal Protein (g/dL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e6.51 ± 0.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e6.13 ± 0.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.109\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eAlbumin (g/dL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e2.75 ± 0.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e2.60 ± 0.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eCreatinine (mg/dL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e0.90 ± 0.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e2.40 ± 1.80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.009\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eBUN (mg/dL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e23.85 ± 17.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e45.47 ± 29.80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.046\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eLDH (U/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e536.23 ± 181.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e548.80 ± 435.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.379\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eCRP (mg/dL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e11.83 ± 6.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e8.33 ± 7.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.123\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eABG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003epH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e7.35 ± 0.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e7.28 ± 0.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.379\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003ePaO\u003csub\u003e2\u003c/sub\u003e (mmHg)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e116.20 ± 47.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e110.39 ± 57.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.382\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003ePaCO\u003csub\u003e2\u003c/sub\u003e (mmHg)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e44.32 ± 13.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e49.83 ± 19.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.59\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eLactate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e1.55 ± 0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e3.02 ± 3.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.023\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eBicarbonate (mmol/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e23.29 ± 3.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e22.06 ± 4.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.469\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eValues are presented as mean ± SD, median [IQR], or n (%), as appropriate. For between-group comparisons of survivors (n=30) and non-survivors (n=10), a two-tailed Student’s t-test was used for approximately normally distributed continuous variables or the Mann–Whitney U test otherwise; categorical variables were compared using Fisher’s exact test. The significance level was α=0.05, with no adjustment for multiple comparisons. All measurements (including arterial blood gases and lactate) were obtained at ICU admission. Available cases were used for the analysis; variable-specific sample sizes may differ where data were missing.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations:\u003c/em\u003e SOFA: Sequential Organ Failure Assessment; P/F ratio: PaO₂/FiO₂ ratio; CBC: complete blood count; WBC: white blood cell count; Coag: coagulation tests; PT-INR: prothrombin time - International Normalized Ratio; aPTT: activated partial thromboplastin time; Chem: blood chemistry; BUN: blood urea nitrogen; LDH: lactate dehydrogenase; CRP: C-reactive protein; ABG: arterial blood gas analysis; PaO₂: arterial oxygen pressure; PaCO₂: arterial carbon dioxide pressure.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePlasma PD-L1 Levels\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePlasma PD-L1 levels were significantly higher in patients with COVID-19 compared to healthy controls at admission (median [IQR]: 294.15 vs. 55.88 pg/mL, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001). However, no significant difference was observed between survivors and non-survivors at admission (\u003cem\u003ep\u003c/em\u003e = 0.390) (Figure 1A–B).\u003c/p\u003e\n\u003cp\u003eLongitudinal analysis revealed a decreasing trend in PD-L1 levels over time in both groups. Notably, PD-L1 concentrations remained higher in non-survivors throughout the observation period. Significant differences were detected on days 5 and 7 (\u003cem\u003ep\u003c/em\u003e = 0.023 and \u003cem\u003ep\u003c/em\u003e = 0.001, respectively), but not on days 1, 14, and 21 (Figure 1C).\u003c/p\u003e\n\u003cp\u003eThese findings suggest that elevated PD-L1 levels during the early phase of infection, particularly on days 5 and 7, may be associated with poor prognosis.\u003c/p\u003e\n\u003cp\u003e＜Figure１＞\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCorrelation Between PD-L1 Levels and Clinical Parameters\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe performed a Spearman correlation analysis to explore the relationship between PD-L1 levels and various clinical and laboratory parameters at five time points (days 1, 5, 7, 14, and 21). The results are summarized as a heatmap in Figure 2, with correlation coefficients annotated in each cell. Statistically significant correlations (\u003cem\u003ep\u003c/em\u003e \u0026lt;0.05) are marked with an asterisk (*).\u003c/p\u003e\n\u003cp\u003eA statistically significant inverse correlation was observed between platelet count and sPD-L1 level at admission (day 1), indicating that higher PD-L1 concentration levels were associated with lower platelet counts from the outset of critical illness. Conversely, creatinine showed a positive correlation with sPD-L1 level from day 1 through day 14, becoming statistically significant on day 7. This suggests that higher PD-L1 level parallels early renal dysfunction. Similarly, inflammatory and coagulation indices, such as C-reactive protein (CRP) and fibrinogen, both showed a positive correlation with sPD-L1 level on multiple days. This suggests that heightened sPD-L1 level reflects the combined burden of renal dysfunction, systemic inflammation, and coagulation activation throughout critical illness.\u003c/p\u003e\n\u003cp\u003e＜Figure２＞\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCox Proportional Hazards Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eForty-six longitudinal candidate variables were entered into separate univariable Cox models. After the Benjamini–Hochberg adjustment (FDR = 5%), six predictors remained below the prespecified threshold (p \u0026lt; 0.05) and were, therefore, eligible for multivariable modeling. Complete univariable screening statistics are provided in Additional file 2.\u003c/p\u003e\n\u003cp\u003eWe then used L1-penalized Cox regression, limited to three clinically plausible covariates: retained lactate, sPD-L1 (day 7), and creatinine (λ selected by 3-fold stratified cross-validation). The final coefficients and hazard ratios are summarized in Table 2.\u003c/p\u003e\n\u003cp\u003eThis penalized model achieved a concordance (Harrell’s C-index) of 0.90 and a partial Akaike information criterion of 45.4, indicating good discrimination with limited overfitting. Bootstrap optimism was \u0026lt; 0.02.\u003c/p\u003e\n\u003cp\u003eAfter shrinkage, higher lactate and elevated sPD-L1 (day 7) were independently associated with an increased instantaneous risk of death (≈ 4.4-fold and 4.0-fold, respectively). Serum creatinine levels were not statistically significant after penalization.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003eTable 2.\u003c/strong\u003e Penalized multivariable Cox model for in-hospital mortality\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"614\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e \u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eβ (log-HR)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eHR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e95% CI\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003ep-value (q)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eLactate\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.79–10.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026lt; 0.005 (0.009)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003esPD-L1 day 7\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.47–10.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.010 (0.028)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eCreatinine\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e−0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.38–1.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.43 (0.61)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eValues include multivariable penalized Cox proportional hazards regression analysis of arterial lactate, day-7 plasma sPD-L1, and serum creatinine levels. Data shown are the regression coefficients β (log-hazard ratio), corresponding hazard ratios (HR), 95% confidence intervals, and two-sided p-values (Wald tests); q-values are p-values adjusted for multiple testing (false discovery rate). For continuous predictors, HRs represent the relative change in hazard per 1-unit increase in the predictor.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations:\u0026nbsp;\u003c/em\u003eCI: Confidence Interval; HR: Hazard Ratio; sPD-L1, soluble programmed death-ligand 1\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRisk Stratification Based on sPD-L1 and Lactate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePatients were dichotomized based on the median values of soluble PD-L1 on day 7 (171 pg mL⁻¹) and arterial lactate (1.4 mmol L⁻¹). Individuals who exceeded both thresholds were classified as \u003cem\u003ehigh-risk\u003c/em\u003e (n = 10), whereas all others formed the low\u003cem\u003e-\u003c/em\u003erisk group (n = 30).\u003c/p\u003e\n\u003cp\u003eKaplan–Meier analysis showed a clear separation between the survival curves (Figure 3), and the log-rank test confirmed a statistically significant difference (χ² = 5.04, df = 1, \u003cem\u003ep\u003c/em\u003e = 0.025).\u003c/p\u003e\n\u003cp\u003eThese findings reveal that higher levels of circulating sPD-L1 and lactate upon ICU admission—which reflect immune dysregulation and tissue hypoxia—can be used to identify patients with COVID-19 at a substantially higher risk of in-hospital mortality.\u003c/p\u003e\n\u003cp\u003e＜Figure３＞\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eML-Based Mortality Prediction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo further explore the predictive potential of clinical variables, including PD-L1 level, we trained and evaluated eight classification algorithms to predict in-hospital mortality. Model performance was assessed using stratified 5-fold cross-validation, and mean AUC values were reported.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSVM showed the highest performance (mean AUC = 0.917), followed by the NN (AUC = 0.883) and LDA (AUC = 0.733). In contrast, the DT and GBM showed lower predictive accuracy (Figure 4A).\u003c/p\u003e\n\u003cp\u003eSHAP analysis, conducted using the SVM model, revealed that PD-L1 levels on day 5 were the most significant predictor of mortality, followed by lactate dehydrogenase level, platelet count, and fibrinogen level (Figure 4B). The beeswarm plot (Figure 4C) confirmed that higher PD-L1 values were associated with higher predicted mortality, whereas higher platelet and fibrinogen levels were associated with decreased risk.\u003c/p\u003e\n\u003cp\u003e＜Figure４＞\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn our cohort of patients with COVID-19 who were critically ill, circulating sPD-L1 generally declined during the first 3 weeks of hospitalization, yet remained persistently elevated in non-survivors. sPD-L1 measurements taken on days 5 and 7 distinguished survivors from non-survivors, indicating that sustained elevation of sPD-L1 is associated with adverse outcomes. Multivariable Cox analysis further identified day-7 sPD-L1 and lactate as independent predictors of in-hospital mortality; the prognostic value of lactate aligns with previous reports [15\u0026ndash; 17]. Day 7 sPD-L1 correlated significantly with creatinine, fibrinogen, and C-reactive protein, suggesting that renal dysfunction, coagulopathy, and systemic inflammation are part of the same pathophysiological axis. These findings suggest that sPD-L1 could be a valuable addition to established prognostic indices.\u003c/p\u003e\n\u003cp\u003eComplementary evidence was obtained from the SVM model, which incorporated day 1 and 7 sPD-L1 levels, along with other routine variables, including age, SOFA score, and CRP level. This model achieved excellent discrimination (AUC \u0026gt; 0.91), and based on SHAP attribution, the day 7 sPD-L1 level was ranked as the dominant predictor of mortality. The difference in importance ranking between the Cox and SVM models likely reflects variance inflation because of multicollinearity, as well as the capacity of non-linear kernels to capture complex interactions that linear proportional hazards models cannot [18, 19]. Overall, our findings support prior studies that have described elevated sPD-L1 in severe COVID-19 cases [11, 20]. Critically, we expand upon these observations by demonstrating that the temporal trajectory of sPD-L1 is directly tied to the prognosis of a patient when evaluated by both classical regression and modern ML methods.\u003c/p\u003e\n\u003cp\u003eThe PD-1/PD-L1 axis acts as an immune checkpoint that helps control T-cell activation during inflammatory stress [7]. In COVID-19, sPD-L1 facilitates viral immune evasion during the early phase of infection, ultimately dampening an overexuberant host response [21]. Therefore, the sustained elevation of sPD-L1 levels observed on ICU day 7 likely signals a shift toward the late, immunosuppressive phase of the disease, characterized by T-cell exhaustion. This persistent PD-L1 expression could be driven by a late surge of interferon-\u0026gamma;, interleukin-6, and hypoxia-inducible factors, which upregulate PD-L1 through signal transducer and activator of transcription 1/\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003einterferon regulatory factor 1 or hypoxia-inducible factor-1\u0026alpha; pathways [22, 23]. Such cytokine- and hypoxia-dependent regulation is consistent with the positive correlation we observed between Day 7 sPD-L1, lactate, and CRP levels, which are markers of systemic inflammation and tissue hypoxia.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;A similar phenomenon is well documented in bacterial sepsis, where persistent PD-1/PD-L1 signaling drives prolonged immune paralysis, and PD-L1-deficient mice exhibit improved survival [24]. PD-1/PD-L1 blockade has yielded favorable outcomes in preclinical sepsis models; however, its therapeutic benefit in clinical trials has yet to be demonstrated [24]. Our observational study provides pathophysiological evidence that sustained PD-L1 upregulation accompanies critical illness because of viral infection. These findings suggest that if checkpoint inhibition is carried out therapeutically, the timing of administration may be crucial to clinical efficacy.\u003c/p\u003e\n\u003cp\u003esPD-L1 shows promise as a clinically useful prognostic marker. Rapid assessment on ICU admission and again on day 7 could enable early identification of patients with COVID-19 who are critically ill and at heightened risk of death, thereby allowing for timely escalation or modulation of immunotherapeutic strategies. In our study, adding sPD-L1 level to an ML model that is based on routine blood tests significantly improved its predictive accuracy, highlighting its added value in multivariable risk scoring. Expanding this approach to include other soluble immune checkpoint molecules such as sCD40, soluble T-cell immunoglobulin and mucin domain (TIM) 1, and galectin-9, which have also been linked to disease severity [20, 25], could facilitate more granular immune profiling and guide personalized treatment algorithms in future studies.\u003c/p\u003e\n\u003cp\u003eThis study has some limitations. First, our study was a small, single-center, observational cohort. Consequently, the generalizability of our findings and any inference regarding therapeutic impact are restricted. Second, our the focus of our analyses was solely on sPD-L1. A more comprehensive evaluation of host immune dysfunction would require the simultaneous quantification of additional immune checkpoint mediators, such as PD-1, TIM 3, and Lymphocyte Activation Gene 3, coupled with functional T-cell assays. Third, since randomized trials have not yet shown unequivocal benefit from PD-1/PD-L1 blockade in sepsis [24], the clinical utility of targeting this axis in severe COVID-19 remains speculative and requires rigorous testing. Validation in larger, multicenter, longitudinal cohorts, together with broad-spectrum immune profiling, is required to confirm our observations and determine their translational relevance.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn this study, we found that persistently elevated plasma PD-L1 levels are associated with poor prognosis in patients with COVID-19 who are critically ill. This suggests that PD-L1\u0026ndash;mediated immune suppression could contribute to disease progression and mortality in severe cases.\u003c/p\u003e\n\u003cp\u003ePD-L1 and lactate levels were used for risk stratification to effectively predict in-hospital mortality. ML models further demonstrated the clinical utility of these markers. PD-L1 may serve as both a prognostic biomarker and a potential therapeutic target, and further studies are warranted to validate its role in the immunopathogenesis of COVID-19.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eABG: Arterial Blood Gas\u003cbr\u003e\u0026nbsp;aPTT: Activated Partial Thromboplastin Time\u003cbr\u003e\u0026nbsp;AUC: Area Under the Curve\u003cbr\u003e\u0026nbsp;BUN: Blood Urea Nitrogen-\u003cbr\u003e\u0026nbsp;CBC: Complete Blood Count\u003cbr\u003e\u0026nbsp;CI: Confidence Interval\u003cbr\u003e\u0026nbsp;Coag: Coagulation tests\u003cbr\u003e\u0026nbsp;COVID-19: Coronavirus Disease 2019\u003cbr\u003e\u0026nbsp;CRP: C-reactive protein\u003cbr\u003e\u0026nbsp;DT: Decision tree\u003cbr\u003e\u0026nbsp;eGFR: Estimated Glomerular Filtration Rate\u003cbr\u003e\u0026nbsp;FiO₂: Fraction of inspired oxygen\u003cbr\u003e\u0026nbsp;GBM: Gradient Boosting Machine\u003cbr\u003e\u0026nbsp;HR: Hazard Ratio\u003cbr\u003e\u0026nbsp;ICU: Intensive Care Unit\u003cbr\u003e\u0026nbsp;IQR: Interquartile Range\u003cbr\u003e\u0026nbsp;LDA: Linear Discriminant Analysis\u003cbr\u003e\u0026nbsp;LR: Logistic Regression\u003cbr\u003e\u0026nbsp;ML: Machine Learning\u003cbr\u003e\u0026nbsp;NN: Neural Network\u003cbr\u003e\u0026nbsp;PaCO₂: Arterial partial pressure of carbon dioxide\u003cbr\u003e\u0026nbsp;PaO₂: Arterial partial pressure of oxygen\u003cbr\u003e\u0026nbsp;P/F ratio: PaO₂ / FiO₂ ratio\u003cbr\u003e\u0026nbsp;PD-1: Programmed Death 1\u003cbr\u003e\u0026nbsp;PD-L1: Programmed Death-Ligand 1\u003cbr\u003e\u0026nbsp;PT-INR: Prothrombin Time\u0026ndash;International Normalized Ratio\u003cbr\u003e\u0026nbsp;RF: Random Forest\u003cbr\u003e\u0026nbsp;ROC: Receiver Operating Characteristic\u003cbr\u003e\u0026nbsp;SARS-CoV-2: Severe Acute Respiratory Syndrome Coronavirus 2\u003cbr\u003e\u0026nbsp;SHAP: Shapley Additive Explanations\u003cbr\u003e\u0026nbsp;SOFA: Sequential Organ Failure Assessment\u003cbr\u003e\u0026nbsp;sPD-L1: soluble Programmed Death-Ligand 1\u003cbr\u003e\u0026nbsp;SVM: Support Vector Machine\u003cbr\u003e WBC: White Blood Cell\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was approved by the Institutional Review Board of Mie University Hospital (approval numbers: IRB No. 3026 and H2021-191). Written informed consent was obtained from the patients or their next of kin upon hospital admission.\u003c/p\u003e\n\u003cp\u003eConsent for publication\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003eAvailability of data and materials\u003c/p\u003e\n\u003cp\u003eData will be shared upon reasonable request and institutional approval.\u003c/p\u003e\n\u003cp\u003eCompeting Interest\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no conflicts of interest.\u003c/p\u003e\n\u003cp\u003eAuthor Contributions\u003c/p\u003e\n\u003cp\u003eStudy conception and design, Data acquisition, Data analysis: ST, EK; Data interpretation: ST, EK, DO, YS, TM, RK; Manuscript drafting and revising: ST, EK, TM, DO, YS, AG, EP, MS, and RK. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003eFunding\u003c/p\u003e\n\u003cp\u003eThis work was supported by JSPS KAKENHI Grant Number JP25K12167, JP24K02546, JP22K06910, and JP21K09015.\u003c/p\u003e\n\u003cp\u003eAcknowledgements\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eArmstrong RA, Kane AD, Kursumovic E, Oglesby FC, Cook TM. Mortality in patients admitted to intensive care with COVID‐19: An updated systematic review and meta‐analysis of observational studies. Anaesthesia. 2021;76:537. \u003c/li\u003e\n\u003cli\u003eHu B, Huang S, Yin L. The cytokine storm and COVID-19. J Med Virol. 2021;93:250-6. \u003c/li\u003e\n\u003cli\u003eDel Valle DM, Kim-Schulze S, Huang HH, Beckmann ND, Nirenberg S, Wang B, et al. An inflammatory cytokine signature predicts COVID-19 severity and survival. Nat Med. 2020;26:1636-43. \u003c/li\u003e\n\u003cli\u003eAlahdal M, Elkord E. Exhaustion and over-activation of immune cells in COVID-19: Challenges and therapeutic opportunities. Clin Immunol. 2022;245:109177. \u003c/li\u003e\n\u003cli\u003eGao X, Cai S, Li X, Wu G. Sepsis-induced immunosuppression: Mechanisms, biomarkers and immunotherapy. Front Immunol. 2025;16:1577105. \u003c/li\u003e\n\u003cli\u003eZhou F, Yu T, Du R, Fan G, Liu Y, Liu Z, et al. Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study. Lancet. 2020;395:1054-62. \u003c/li\u003e\n\u003cli\u003eFreeman GJ, Long AJ, Iwai Y, Bourque K, Chernova T, Nishimura H, et al. Engagement of the Pd-1 immunoinhibitory receptor by a novel B7 family member leads to negative regulation of lymphocyte activation. J Exp Med. 2000;192:1027-34. \u003c/li\u003e\n\u003cli\u003eBarber DL, Wherry EJ, Masopust D, Zhu B, Allison JP, Sharpe AH, et al. Restoring function in exhausted CD8 T cells during chronic viral infection. Nature. 2006;439:682-7. \u003c/li\u003e\n\u003cli\u003eYang L, Gao Q, Li Q, Guo S. PD-L1 blockade improves survival in sepsis by reversing monocyte dysfunction and immune disorder. Inflammation. 2024;47:114-28.\u003c/li\u003e\n\u003cli\u003eSabbatino F, Conti V, Franci G, Sellitto C, Manzo V, Pagliano P, et al. PD-L1 dysregulation in COVID-19 patients. Front Immunol. 2021;12:695242. \u003c/li\u003e\n\u003cli\u003eBeserra DR, Alberca RW, Branco ACCC, De Mendon\u0026ccedil;a Oliveira L, De Souza Andrade MM, Gozzi-Silva SC, et al. Upregulation of PD-1 expression and high sPD-L1 levels associated with COVID-19 severity. J Immunol Res. 2022;2022:9764002. \u003c/li\u003e\n\u003cli\u003eSchniederova M, Bobcakova A, Grendar M, Markocsy A, Ceres A, Cibulka M, et al. Lymphocyte inhibition mechanisms and immune checkpoints in COVID-19: Insights into prognostic markers and disease severity. Medicina (Kaunas). 2025;61:189.\u003c/li\u003e\n\u003cli\u003eOgundimu EO, Altman DG, Collins GS. Adequate sample size for developing prediction models is not simply related to events per variable. J Clin Epidemiol. 2016;76:175-82. \u003c/li\u003e\n\u003cli\u003evan Smeden M, Moons KGM, de Groot JAH, Collins GS, Altman DG, Eijkemans MJC, et al. Sample size for binary logistic prediction models: Beyond events per variable criteria. Stat Methods Med Res. 2019;28:2455-74. \u003c/li\u003e\n\u003cli\u003eCarpen\u0026egrave; G, Onorato D, Nocini R, Fortunato G, Rizk JG, Henry BM, et al. Blood lactate concentration in COVID-19: A systematic literature review. Clin Chem Lab Med. 2022;60:332-7. \u003c/li\u003e\n\u003cli\u003eBirben B, Birben OD, Akın T, Akkurt G, Surel AA, Yakısık E, et al. Efficacy of the delta neutrophil index in predicting 30-day mortality in COVID-19 patients requiring intensive care. Int J Clin Pract. 2021;75:e13970. \u003c/li\u003e\n\u003cli\u003eAlharthy A, Aletreby W, Faqihi F, Balhamar A, Alaklobi F, Alanezi K, et al. Clinical characteristics and predictors of 28-day mortality in 352 critically ill patients with COVID-19: A retrospective study. J Epidemiol Glob Health. 2021;11:98-104. \u003c/li\u003e\n\u003cli\u003eSaarela M, Jauhiainen S. Comparison of feature importance measures as explanations for classification models. SN Appl Sci. 2021;3:1-12. \u003c/li\u003e\n\u003cli\u003eNoble WS. What is a support vector machine? Nat Biotechnol. 2006;24:1565-7. \u003c/li\u003e\n\u003cli\u003eAvenda\u0026ntilde;o‐Ortiz J, Lozano‐Rodr\u0026iacute;guez R, Mart\u0026iacute;n‐Quir\u0026oacute;s A, Terr\u0026oacute;n V, Maroun‐Eid C, Montalb\u0026aacute;n‐Hern\u0026aacute;ndez K, et al. The immune checkpoints storm in COVID‐19: Role as severity markers at emergency department admission. Clin Transl Med. 2021;11:e573.\u003c/li\u003e\n\u003cli\u003eSabbatino F, Pagliano P, Sellitto C, Stefanelli B, Corbi G, Manzo V, et al. Different prognostic role of soluble PD-L1 in the course of severe and non-severe COVID-19. J Clin Med. 2023;12:6812.\u003c/li\u003e\n\u003cli\u003eGarcia-Diaz A, Shin DS, Moreno BH, Saco J, Escuin-Ordinas H, Rodriguez GA, et al. Interferon receptor signaling pathways regulating PD-L1 and PD-L2 expression. Cell Rep. 2017;19:1189-201. \u003c/li\u003e\n\u003cli\u003eNoman MZ, Desantis G, Janji B, Hasmim M, Karray S, Dessen P, et al. PD-L1 is a novel direct target of HIF-1\u0026alpha;, and its blockade under hypoxia enhanced: MDSC-mediated T cell activation. J Exp Med. 2014;211:781-90.\u003c/li\u003e\n\u003cli\u003eZhang T, Yu-jing L, Ma T. Role of regulation of PD-1 and PD-L1 expression in sepsis. Front Immunol. 2023;14:1029438. \u003c/li\u003e\n\u003cli\u003eParanga TG, Pavel-Tanasa M, Constantinescu D, Iftimi E, Plesca CE, Miftode IL, et al. Distinct soluble immune checkpoint profiles characterize COVID-19 severity, mortality and SARS-CoV-2 variant infections. Front Immunol. 2024;15:1464480. \u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":false,"email":"","identity":"journal-of-intensive-care","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Journal of Intensive Care","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"VoR Journals","inReviewEnabled":false,"inReviewRevisionsEnabled":false},"keywords":"COVID-19, soluble PD-L1, immune checkpoints, organ dysfunction, ICU mortality, machine learning","lastPublishedDoi":"10.21203/rs.3.rs-7521856/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7521856/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cb\u003eBackground\u003c/b\u003e\u003c/p\u003e\u003cp\u003ePersistent immune checkpoint activation is a recognized feature of critical Coronavirus Disease 2019 (COVID-19). However, the temporal behavior and clinical utility of soluble Programmed Death-Ligand 1 (sPD-L1) remain unclear. We aimed to investigate the longitudinal changes in sPD-L1, their relationship with organ dysfunction markers, and their prognostic value when combined with machine learning (ML) models.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMethods\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThis single-center observational study included 40 adults with severe COVID-19 pneumonia admitted to the intensive care units (ICU) (April 2021\u0026ndash;December 2022) and 23 healthy volunteers. We measured plasma sPD-L1 on ICU days 1, 5, 7, 14, and 21. Routine biochemistry, full blood counts, and arterial blood gas analyses were conducted in parallel. Cox regression analysis was used to identify independent predictors of hospital mortality, which was the primary outcome. Eight ML classifiers were trained on admission variables, as well as day 1, 5, and 7 sPD-L1 levels. Discrimination was assessed using stratified five-fold cross-validation and Shapley Additive Explanations (SHAP) attribution.\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults\u003c/b\u003e\u003c/p\u003e\u003cp\u003eTen of the forty patients died during hospitalization. Overall, sPD-L1 levels declined during the ICU stay but remained persistently high in non-survivors. Values on days 5 and 7 differed significantly between survivors and non-survivors (p\u0026thinsp;=\u0026thinsp;0.023 and 0.001, respectively). In multivariable Cox analysis, day-7 sPD-L1 and arterial lactate levels on admission independently predicted mortality. Day 7 sPD-L1 level correlated positively with creatinine, C-reactive protein, and fibrinogen (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), linking immune checkpoint activation to renal injury, inflammation, and coagulopathy. A support vector machine model achieved the highest discriminative accuracy (mean area under the curve\u0026thinsp;=\u0026thinsp;0.917). Day 5 sPD-L1 was designated as the primary predictor of mortality based on SHAP attribution, with lactate contributing minimally.\u003c/p\u003e\u003cp\u003e\u003cb\u003eConclusion\u003c/b\u003e\u003c/p\u003e\u003cp\u003eSustained sPD-L1 elevation in the initial ICU week is strongly associated with early organ dysfunction and independently predicts death in critical COVID-19. Incorporating serial sPD-L1 levels into bedside ML models significantly enhances risk discrimination. These findings support sPD-L1 as an integrative biomarker of immune\u0026ndash;renal\u0026ndash;coagulation interplay, thus necessitating validation in larger multicenter cohorts and exploration as a potential companion marker for immune-modulatory interventions.\u003c/p\u003e","manuscriptTitle":"Use of Dynamic Soluble Programmed Death-Ligand 1 Trajectories to Identify Early Organ Dysfunction and Predict Mortality in Critical Coronavirus Disease 2019","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-03 01:04:49","doi":"10.21203/rs.3.rs-7521856/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-09-30T12:45:59+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-09-30T11:36:08+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-09-30T06:04:29+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-09-24T12:39:07+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"83535148212351296693743827378503146833","date":"2025-09-21T10:24:11+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"147812860999418739867810205887828663691","date":"2025-09-20T03:17:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"134629164069420901173941083108252501327","date":"2025-09-20T01:08:22+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"182819950078425978125532010193087126980","date":"2025-09-19T23:01:59+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-09-19T10:03:25+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-09-19T08:41:00+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-09-19T01:49:30+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of Intensive Care","date":"2025-09-18T23:43:44+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":false,"email":"","identity":"journal-of-intensive-care","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Journal of Intensive Care","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"VoR Journals","inReviewEnabled":false,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"435dde4c-ec51-4cd7-bcd1-7d6627be0686","owner":[],"postedDate":"October 3rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-12-28T10:08:42+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-03 01:04:49","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7521856","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7521856","identity":"rs-7521856","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}