The C-Reactive Protein-to-Albumin Ratio (CAR) and All-Cause Mortality in Critically Ill Ischemic Stroke Patients: A Retrospective Analysis of the MIMIC-IV and eICU-CRD Databases

The C-Reactive Protein-to-Albumin Ratio (CAR) and All-Cause Mortality in Critically Ill Ischemic Stroke Patients: A Retrospective Analysis of the MIMIC-IV and eICU-CRD Databases | 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 The C-Reactive Protein-to-Albumin Ratio (CAR) and All-Cause Mortality in Critically Ill Ischemic Stroke Patients: A Retrospective Analysis of the MIMIC-IV and eICU-CRD Databases Wang Binyang, Zhong Jing, Shao Lu, Fan shuochen, Wang Shiping, and 6 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7185510/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract The C-Reactive Protein-to-Albumin Ratio (CAR) demonstrates associations with cerebrovascular disease outcomes. However, its prognostic value in critically ill ischemic stroke (IS) patients intensive care unit (ICU) admission remains unclear. This study aimed to investigate the association between CAR and clinical prognosis in critically ill IS patients. In this retrospective cohort study, clinical data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database (serving as the training set) and externally validated using the eICU Collaborative Research Database (eICU-CRD). The primary outcomes were 28-day, 60-day, and 90-day all-cause mortality. The association between CAR and mortality was evaluated using multivariable logistic regression and restricted cubic splines (RCS). Machine learning algorithms were employed to develop prediction models incorporating CAR. Model performance was assessed using the Boruta algorithm for feature importance and the Integrated Discrimination Improvement (IDI). A total of 2,664 critically ill IS patients were analyzed (mean CAR: 22.173 ± 26.011). After adjusting for confounders, multivariable logistic regression confirmed CAR as an independent predictor of mortality: the adjusted odds ratios (95% confidence intervals) were 1.006 (1.003–1.010, P = 0.033) for 28-day, 1.005 (1.001–1.008, P = 0.005) for 60-day, and 1.004 (1.001–1.007, P = 0.016) for 90-day mortality. RCS analysis indicated a monotonically increasing association between CAR and mortality risk. Machine learning models incorporating CAR demonstrated superior fit and higher area under the curve (AUC) values compared to models without it. In conclusions,In critically ill patients with ischemic stroke, a higher CAR is significantly associated with increased short- and medium-term all-cause mortality risk. ischemic stroke C-reactive protein albumin Prognosis Machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Stroke represents the second leading cause of global mortality and third leading cause of disability worldwide. In 2019 alone, stroke affected over 100 million individuals, with 12 million new cases and approximately 7 million deaths [ 1 ]. Stroke causes significant macroeconomic losses. In 2019, the total welfare loss value by stroke worldwide was 2,059.67 billion US dollars, accounting for 1.66% of the global GDP, while the global welfare loss value by ischemic stroke was 964.51 billion US dollars, accounting for 0.78% of the global GDP [ 2 ]. Projections indicate stroke-related direct medical costs in the United States will surge from 36.7 billionin in 2015 to 94.3 billion by 2035 [ 3 ]. Complications such as post-stroke cognitive impairment, post-stroke depression, hemorrhagic transformation, gastrointestinal dysfunction, cardiovascular events, and post-stroke infection often occur after ischemic stroke, which affect the prognosis of the disease and lead to progressive neurological deficits and high mortality [ 4 ]. Despite the current use of treatment methods such as intravenous thrombolysis, endovascular thrombectomy, cell protection or adjuvant drugs, the risk of adverse clinical outcomes in IS patients remains high, especially in critically ill patients [ 5 , 6 ]. Early and effective intervention and management are crucial. Timely treatment based on monitoring relevant indicators significantly enhances patient prognosis. Utilizing these laboratory parameters to predict clinical outcomes in IS patients facilitates optimized clinical management and prognosis assessment. Neuroinflammation after stroke is a major factor leading to poor functional outcomes and death. In clinical practice, it is necessary to monitor the inflammatory levels of patients [ 7 ]. C-reactive protein (CRP) is an acute-phase protein used as an inflammatory marker. In stroke patients, elevated CRP levels may indicate an enhanced inflammatory response, which is associated with poor prognosis and increased risk of death [ 8 ].Current data suggest that a high level of CRP within 24 hours after the onset of IS is independently associated with poor functional outcomes after acute ischemic events [ 9 ]. Albumin is the most widely studied protein for diagnosing malnutrition. Hypoalbuminemia (< 3.5 g/dL) is defined as an indicator of malnutrition to screen for malnourished patients, and hypoalbuminemia may provide as a potential predictive biomarker for inflammation [ 10 ]. There is evidence that for every one-unit increase in serum albumin level (g/L), the impairment of activities of daily living in stroke patients decreases by 7% [ 11 ]. Recent studies confirm that ​​elevated C-reactive protein (CRP) and hypoalbuminemia​​ independently predict mortality and are widely utilized to anticipate complications and fatal outcomes in critically ill patients [ 12 , 13 ]. Notably​​, the CRP/Albumin ratio (CAR) surpasses either biomarker alone in ​​accurately reflecting systemic inflammation and predicting prognosis​​ across cardiovascular, cerebrovascular, and oncological diseases [ 14 – 17 ]. ​As a novel inflammatory prognostic score​​, CAR has further demonstrated utility in forecasting outcomes for diverse inflammatory conditions—including sepsis, pneumonia, multiple arthritides, perforated appendicitis, and COVID-19 [ 18 – 22 ]. Nevertheless​​, the association between CAR and ​​all-cause mortality in critically ill IS patients remains unestablished​​. ​ ​ Leveraging the MIMIC-IV database, this study investigates the association between the CAR and 28-, 60-, and 90-day all-cause mortality in critically ill IS patients. We will develop a machine learning-based predictive model and employ the SHAP framework to interpret feature importance. This approach aims to assist clinicians in quantitatively assessing disease severity and implementing personalized early interventions to mitigate mortality risk in this vulnerable population Methods Study Population The original data were obtained from the MIMIC-IV database, which is a contemporary electronic health record dataset established by Beth Israel Deaconess Medical Center and the Massachusetts Institute of Technology, covering information on over 70,000 adult patients admitted to the emergency department or intensive care unit at Beth Israel Deaconess Medical Center in Boston, Massachusetts, USA, from 2008 to 2019 [ 23 ]. eICU is a multi-center intensive care unit database containing high-resolution data on over 200,000 patients admitted to 208 different ICUs in the United States from 2014 to 2015. The source hospital of the MIMIC-IV did not participate in the eICU program [ 24 ]. The author (Binyang Wang: 13488206) obtained access to the databases. All patient records in both databases were fully de-identified, so informed consent and ethical approval were not required.. Inclusion criteria Patients diagnosed with IS according to the 9th and 10th editions of the International Classification of Diseases were included in this study. Exclusion criteria were as follows: (1) For patients with multiple hospitalizations, only the data from the first hospitalization were included; (2) patients who died within 28 days after admission; (3) patients with severe diseases such as end-stage renal failure, liver cirrhosis, or cancer; (4) patients with an ICU stay < 3 hours; and (5) patients lacking CRP and albumin data at admission. Finally, a total of 2,664 patients were included in this study and were divided into four groups based the quartiles of CAR index (Q1 n = 667, Q2 n = 665, Q3 n = 666, and Q4 n = 666). Data Extraction CAR is a continuous variable calculated as the serum CRP concentration (mg/dl) closest to admission divided by the serum albumin concentration (g/dl) closest to admission. CAR was then divided into quartiles and classified as follows: Q1: 1.92, Q2: 12.19, Q3: 33.58, and Q4: 155.00. The extracted potential variables can be classified into five major categories: (1) Demographic data, such as age and gender. (2) Comorbidities, including myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic lung disease, uncomplicated diabetes, complicated diabetes, paraplegia, kidney disease, and sepsis. (3) Laboratory indicators, including the C-reactive protein, albumin level, C-reactive protein-to-albumin ratio, international normalized ratio, prothrombin time, partial thromboplastin time, hemoglobin level, hematocrit, mean corpuscular ,mean corpuscular hemoglobin concentration level, mean corpuscular volume, platelet count, red blood cell count, red blood cell distribution width, white blood cell count, anion gap, bicarbonate, blood urea nitrogen, glucose, serum creatinine, serum calcium, serum chloride, serum sodium, and serum potassium. (4) Admission disease severity scores, including the Glasgow Coma Scale (GCS), Simplified Acute Physiology Score II (SAPS-II), Systemic Inflammatory Response Syndrome (SIRS), and Charlson. (5) Hospitalization and mortality-related time, such as length of hospital stay, length of ICU stay, 28-day all-cause mortality, 60-day all-cause mortality, and 90-day all-cause mortality. All laboratory parameters extracted from the database were measured for the first time after admission to the ICU. For missing data, first, feature columns with a missing data rate over 30% were excluded as they contained low information. Then,the remaining missing values of the retained features were subsequently filled using the mode imputation method. This approach eliminates high-missingness dimensions through threshold screening and selects the mode for filling based on the distribution characteristics of the variables, balancing data integrity and distribution stability. Clinical Outcomes The primary outcome of this study was all-cause mortality at 28, 60, and 90 days. Model Construction The Boruta algorithm is a feature selection and wrapping algorithm based on random forests. It assesses the importance of features by generating values for each feature in the dataset and comparing them with the values of the corresponding "shadow features" [ 25 ]. We used the Boruta algorithm to select clinical features and built six machine learning algorithm prediction models: Random Forest, AdaBoost, Gradient Boosting, XGBoost, LightGBM, and Neural Network. The models were established on the training set, and internal and external validation sets were used to validate the best model. The performance of the prediction models was evaluated using the area under the receiver operating characteristic curve (AUC) and accuracy. Additionally, decision curve analysis (DCA) and calibration curves were plotted to demonstrate the true clinical utility, and the integrated discrimination improvement (IDI) was calculated. The optimal model was used to analyze all-cause mortality at 28, 60, and 90 days, and the SHAP method was employed to visualize the contribution of each feature to the prediction results. Statistical Analysis The data collected for analysis were divided into two major categories: categorical variables and continuous variables. Continuous variables were presented as mean (standard deviation(SD) and were compared using tests or no-parametric tests, as appropriate. Categorical variables were presented as frequency and percentage(%) and were compared between groups using chi-square tests or Fisher's exact tests. To assess the correlation between CAR and the risk of death at 28, 60, and 90 days, multivariate logistic regression analysis was conducted to calculate odds ratios (ORs) and their corresponding 95% confidence intervals (CIs) to quantify the impact of CAR on these outcomes, with adjustment for confounding variables. The restricted cubic spline (RCS) method was used to explore the potential non-linear association between CAR and the outcome. Subgroup analyses were also conducted to verify the association between CAR and 28-day, 60-day, and 90-day mortality within each subgroup. Six machine learning models were trained on the training set and validation set, and the performance of the models was compared. SHAP values were used to explain the best-performing model. Statistical significance was defined as a two-sided P value < 0.05. All the statistical analyses were performed using R software and Python. Results Baseline characteristics This study included 2,664 eligible participants stratified by CAR quartiles from the MIMIC-IV database. The baseline characteristics of the participants are shown in Table 1. The average age of the participants was 67.493 ± 14.746 years, 53.23% were male, and the average CAR index was 22.173 ± 26.011. The 28-day, 60-day, and 90-day mortality rates were 42.00%, 46.21%, and 55.43%, respectively. There were significant differences in gender, length of hospital stay, length of ICU stay, CRP, albumin, international normalized ratio, prothrombin time, partial thromboplastin time, blood urea nitrogen, serum calcium, serum chloride, serum sodium, serum potassium, myocardial infarction, congestive heart failure, peripheral vascular disease, chronic lung disease, diabetes with chronic complications, paraplegia, sepsis, kidney disease, hematocrit, hemoglobin, mean corpuscular hemoglobin concentration, platelet count, red blood cell count, red blood cell distribution width, white blood cell count, and CAR among the different CAR quartiles (p < 0.05). Compared with the other groups, the highest CAR level (Q4) had longer hospital and ICU stays, higher levels of CRP, international normalized ratio, prothrombin time, partial thromboplastin time, serum calcium, and serum sodium;,and red blood cell distribution width; and higher white blood cell count, while lower levels of albumin, blood urea nitrogen, serum potassium, hematocrit, hemoglobin, mean corpuscular hemoglobin concentration, and red blood cell count were detected. The SAPS II score, SIRS score, and Charlson score at admission were higher, and the GCS score was lower in the Q4 group, indicating that as CAR level, the severity of disease scores also increased. Additionally, the Q4 group had the longest hospital and ICU stays and higher prevalence of myocardial infarction, congestive heart failure, peripheral vascular disease, chronic lung disease, diabetes with chronic complications, paraplegia, sepsis, and kidney disease. Compared with the other groups, the Q4 group had significantly higher mortality rates at 28 days (45.195% vs. 45.796% vs. 44.361% vs. 32.684%, P < 0.001), 60 days (51.351% vs. 49.850% vs. 47.970% vs. 35.682%, P < 0.001), and 90 days (52.853% vs. 50.901% vs. 48.722% vs. 35.982%, P < 0.001). The differences in baseline characteristics between survivors and non-survivors at 28, 60 and 90 days during hospitalization are shown in Table 2. Patients in the non-survivor group were older and more likely to be male, had a higher severity of disease, and had shorter hospital and intensive care unit (ICU) stays. The prevalence of myocardial infarction, congestive heart failure, peripheral vascular disease, chronic lung disease, diabetes with chronic comorbidities, paraplegia, sepsis, and kidney disease was significantly higher in the non-survivor group. Compared with the survivor group, the non-survivor group had worse coagulation function, more severe anemia, and more prominent electrolyte disorders, with higher anion gap, CRP, WBC, and creatinine values. Additionally, CAR level in non-survivors was significantly higher than that in survivors (28 days: 24.359 vs 20.590,60 days: 24.754 vs 19.954,90 days: 24.859 vs 19.781,P < 0.001). The impact of CAR index on clinical outcomes We used continuous and categorical variable logistic regression models to study the independent effect of CAR on mortality and found that a higher CAR was positively correlated with an increased risk of death in critically ill IS patients (Table 3). Model 1 was unadjusted, Model 2 was adjusted for age, gender, hospital stay, and ICU stay; model 3: adjusted for age, gender, hospital stay, ICU stay, GCS score, SAPSll score, SIRS score, and Charlson score; model 4: age, gender, hospital stay, ICU stay, GCS score, SAPSll score, SIRS score, Charlson score, myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, chronic lung disease, diabetes without chronic comorbidities, diabetes with chronic comorbidities, paraplegia, and kidney disease were adjusted. The results showed that when CAR was a continuous variable, it was independently associated with increased 28-day mortality (OR = 1.006, 95% CI = 1.003–1.010, P = 0.033), 60-day mortality (OR = 1.005, 95% CI = 1.001–1.008, P = 0.005), and 90-day mortality (OR = 1.004; 95% CI = 1.001–1.007, P = 0.016), and all were significant risk factors. These results were further confirmed in the fully adjusted model 4. Specifically, when CAR was a nominal variable, the OR for 28-day mortality in the highest CAR quartile was 1.510, 95% CI: 1.163–1.962, P = 0.002; the OR for 60-day all-cause mortality was 1.368, 95% CI: 1.065–1.759, P = 0.014; and the OR for 90-day all-cause mortality was 1.331, 95% CI: 1.038–1.709, P = 0.024, all compared with the lowest quartile. Moreover, the risk of 28-day mortality, 60-day mortality, and 90-day mortality increased with the increase in CAR index quartiles. (Detailed information is shown in Table 3.) Nonlinear relationship between CAR index and clinical outcomes We used RCS curve analysis to reveal the nonlinear relationship between CAR and all-cause mortality at different time points (28 days, 60 days, and 90 days) (Fig. 1). When evaluating relationship between CAR and 28-day all-cause mortality in critically ill IS patients was evaluated, all the models showed significant nonlinearity (model 1: P-nonlinear < 0.001; Model 2: P-nonlinear < 0.001; Model 3: P-nonlinear = 0.0018; Model 4: P-nonlinear = 0.0139). CAR showed a significant nonlinear relationship with 28-, 60-, and 90-day all-cause mortality in the Model 1 and in partially adjusted Models 2 and 3. As the confounding factors were gradually adjusted, the nonlinear relationship weakened. However, the nonlinear correlation analysis results of CAR and 60-day and 90-day mortality indicated a linear association in the fully adjusted Model 4. That is, the all-cause mortality of critical IS patients at 60 and 90 days increased linearly with the increase of CAR. (P-nonlinear = 0.1916, P-nonlinear = 0.1921) Subgroup analysis To investigate potential variations within specific populations, logistic regression analysis was conducted for various subgroups, including age, gender, congestive heart failure, kidney disease, chronic lung disease, dementia, cerebrovascular disease, diabetes with or without chronic complications, sepsis, peripheral vascular disease, myocardial infarction, GCS Score, SAPSII Score, SIRS Score, and Charlson Score. The forest plot (Fig. 2) showed that CAR was significantly associated with 28-(Fig. 22A), 60-(Fig. 2B), and 90-day(Fig. 2C) all-cause mortality in critical IS patients aged < = 65 years, SAPSII Score < = 35, and Charlson Score < = 5 (P < 0.05). Additionally, CAR was significantly associated with 28-day all-cause mortality in IS patients with cerebrovascular disease (P = 0.038). The interaction analysis indicated that CAR seemed to more accurately predict 28-day all-cause mortality in patients aged < = 65, SAPSII Score < = 35, and Charlson Score < = 5 (P-interaction = 0.049, 0.002, 0.004). Moreover, CAR more accurately predicted 60-day and 90-day all-cause mortality in patients with a SAPSII Score < = 35 and a Charlson score < = 5 (P-interaction < 0.05). There were no significant differences in 28-, 60-, and 90-day all-cause mortality based on gender, presence of congestive heart failure, kidney disease, chronic lung disease, dementia, cerebrovascular disease, diabetes with or without chronic complications, sepsis, peripheral vascular disease, myocardial infarction, GCS Score, and SIRS Score (P > 0.05). Model construction and performance comparison Feature selection based on the Boruta algorithm was used to select clinical features and construct six machine learning models: Random Forest, AdaBoost, Gradient Boosting, XGBoost, LightGBM, and Neural Network to predict the mortality risk of critical IS patients. Figure 3 shows the performance of various models. The results indicated that among the six prediction models for 28-, 60-, and 90-day all-cause mortality in the training set, the machine learning models containing CAR had better fit and higher AUC values. Regarding 28-day all-cause mortality (Fig. 3A-M), in the training cohort, Random Forest with CAR showed the best model fit, with an area under the curve (AUC) of 0.80(Fig. 4A); without the CAR(Fig. 3B), Gradient Boosting showed the best model fit, with an AUC of 0.77. In the eICU-CRD validation cohort(Fig. 3G), XGBoost had the best fit (AUC = 0.79). Regarding the 60-day all-cause mortality rate, in the training cohort(Fig. 3C), the Gradient Boosting model was the best in CAR model (AUC = 0.75), and in the model without the CAR(Fig. 3D), Gradient Boosting was also the best (AUC = 0.72). In the eICU-CRD validation cohort(Fig. 3H), LightGBM was the best (AUC = 0.73), followed by XGBoost and Gradient Boosting (AUC = 0.72). Regarding the 90-day all-cause mortality rate, in the training cohort(Fig. 3E), Random Forest and Gradient Boosting were the best in CAR model (AUC = 0.75), and in the model without the CAR(Fig. 3F), Neural Network was the best (AUC = 0.73). In the eICU-CRD validation cohort(Fig. 4I), Random Forest was the best (AUC = 0.80), followed by Gradient Boosting (AUC = 0.79). We evaluated the accuracy of each model in predicting the 28-day, 60-day, and 90-day all-cause mortality probabilities of critical IS patients by analyzing the calibration curves and clinical decision curves of the training set(Fig. 3J-O). The results showed that the calibration curve of Gradient Boosting had a good fit, indicating high consistency between the model prediction and the actual incidence rate. In terms of clinical applicability, the Gradient Boosting model demonstrated a robust net benefit across a wide range of threshold probabilities. Table 4 shows the performance of the six models in the training and test sets. In the 28-day model, Gradient Boosting had the highest accuracy in the training set (71.5%), and in the validation set, Gradient Boosting had an accuracy of 88.37%, second only to Adaboost (89.53%). In the 60-day model, Gradient Boosting had the highest accuracy in the training set (69.25%), and in the validation set, Random Forest had the highest accuracy (91.86%). In the 90-day model, Random Forest had the highest accuracy in both the training set (69%) and the validation set (93.02%). Although the accuracy of Gradient Boosting was not the highest in the validation set, it was all above 80%, which could provide strong support for doctors' clinical decisions. In summary, the Gradient Boosting model demonstrated excellent overall performance. The incremental effect of CAR index We calculated the IDI of the scoring tools (GCS, SAPSII, SIRS, OASIS, Charlson) and analyzed the impact of CAR on their predictive ability (Table 5). This study showed that in predicting 28-day and 60-day all-cause mortality rates, CAR significantly enhanced the incremental of the GCS, SIRS, OASIS, and Charlson scores (P 0.05). In predicting the 90-day all-cause mortality rate, CAR significantly enhanced the predictive accuracy of all the scoring tools (P < 0.05). Model interpretation In recent years, many machine learning models have been applied to predict adverse outcomes in IS patients. The lack of interpretability limits the application of more powerful machine learning methods in medical decision support. SHAP combines optimal credit assignment with local explanations, enabling intuitive interpretation of the importance of individual variables in the model [26]. Therefore, we used the SHAP algorithm to graphically demonstrate the important influencing features of the Gradient Boosting model in predicting mortality (Fig. 4). The results Show that length of hospital stay, Charlson Score, SAPS-II score, red blood cell distribution width, blood urea nitrogen, age, partial thromboplastin time, platelet count, prothrombin time, albumin, CRP, glucose, red blood cell count, congestive heart failure, and CAR were the main influencing factors for predicting 28-day, 60-day, and 90-day mortality. To determine the main predictors of all-cause mortality in IS patients, we calculated the 15 most important features of the best machine learning model. SHAP summary bar charts B, E, and H show the top 15 most important features in the 28-day, 60-day, and 90-day prediction models, with variables listed in descending order of importance. Length of hospital stay is the most important predictor of 28-day all-cause mortality(Fig. 4A,B), followed by Charlson Score, SAPS-II score, red blood cell distribution width, and blood urea nitrogen. Compared with other factors, the SAPS-II score has a greater impact on 60-day and 90-day all-cause mortality(Fig. 4D-E,G-H), than other factors followed by the Charlson Score, red blood cell distribution width, and blood urea nitrogen level. Positive SHAP values indicate that the feature will have a positive impact on the output, i.e., a higher risk of death. Red indicates a higher feature value, and blue indicates a lower feature value. SHAP summary dot plots (Figs. 4A, D, and G) visually display the direction and intensity of each feature's impact on the model prediction: advanced age, high blood urea nitrogen, and high blood glucose levels significantly increase the risk of death in critically ill IS patients. Notably, a high CAR increases the risk of 28-day, 60-day, and 90-day all-cause mortality in the prediction model for critically ill IS patients. To further explore the contribution of these features to specific individual patients and clinical applications, we conducted local model explanations (Figs. 4C, F, and I). The red area indicates that the feature increases the risk of death, while the blue area indicates that the feature reduces the risk of death. The wider the color area is, the greater the impact of the feature on death. The results show that the length of hospital stay, SAPS-II score, red blood cells, and platelet count are positive contributors to 28-day and 60-day mortality prediction. SAPS-II score, hematocrit, platelet count, and CRP are positive contributors to 90-day mortality. Discussion This study analyzed 2,664 critically ill ischemic stroke (IS) patients and 43 potential variables to evaluate the association between the C-reactive protein to albumin ratio (CAR) and clinical outcomes. Our findings demonstrate that elevated CAR levels are significantly associated with increased 28-day, 60-day, and 90-day all-cause mortality in this population.Notably, significant differences in clinical history variables were observed between survivors and non-survivors. Non-survivors exhibited higher frequencies of comorbidities including myocardial infarction, congestive heart failure, peripheral vascular disease, chronic lung disease, diabetes, paraplegia, sepsis, and kidney disease. They also presented with higher SAPS II, SIRS, and Charlson scores, alongside lower GCS scores—indicative of greater clinical severity. Importantly, CAR was significantly elevated in non-survivors with these conditions. Notably, the CAR of non-survivors with the aforementioned conditions was significantly higher than that of survivors. In a prospective study [ 27 ], it was indicated that elevated CRP and decreased albumin levels were associated with poor prognosis in IS patients 3 months later, which is similar to our research results. In our study, there were significant differences in CRP (34.0 vs 49.1, 32.6 vs 50.1, 31.8 vs 50.2, p < 0.001) and albumin (3.6 vs 3.4, p < 0.001) between survivors and non-survivors at 28 days, 60 days, and 90 days. This is consistent with previous studies showing that high inflammatory levels and low albumin levels are associated with a worse prognosis in IS patients [ 28 – 31 ]. Previous studies have found that elevated CRP is significantly associated with the severity and mortality of IS patients [ 32 – 33 ]. Bucci et al. [ 34 ] conducted a retrospective observational study within TriNetX and found that IS patients with CRP levels > 3 mg/L had a significantly increased risk of cardiac complications such as death, heart failure, ischemic heart disease, atrial fibrillation, and ventricular arrhythmia. Low albumin is a marker of malnutrition, and critically ill IS patients are in a state of high catabolism. Without appropriate nutritional intervention, patients are prone to malnutrition, leading to poor clinical outcomes [ 35 ]. Additionally, albumin has neuroprotective effects due to its antioxidant, anti-apoptotic, and anti-inflammatory properties, which can reduce the recurrence rate and mortality of acute ischemic stroke patients and the occurrence of complications such as hemorrhagic transformation [ 36 , 37 ]. Thuemmler et al. [ 38 ] found that acute ischemic stroke patients with albumin < 37 g/L had a 48% increased risk of in-hospital mortality and a twofold increased risk of long-term mortality. Zhou et al. [ 39 ], through data analysis and meta-analysis from the Third China National Stroke Registry, found that low albumin levels increased the risk of functional disability and death in acute ischemic stroke patients at 3 months and 1 year. CAR not only reflects the inflammatory level of patients but also their nutritional status. An increase in CAR indicates greater inflammation, lower nutrition, and weaker neuroprotective effects, which may synergistically increase the risk of death in IS patients. Currently, studies have shown that CAR is associated with adverse outcomes such as hemorrhagic transformation after thrombolysis, post-stroke complications such as stroke-associated pneumonia, and dysphagia in the elderly [ 40 – 42 ]. Previous studies have shown [ 43 ] that the composite index CAR has a higher predictive value for weaning from the ventilator in patients with traumatic brain injury than CRP or albumin alone. Therefore, we included 2,664 critically ill IS patients to explore the predictive value of CAR in predicting mortality in critically ill IS patients. The study results showed that an elevated CAR was associated with 28-day, 60-day, and 90-day all-cause mortality in critically ill IS patients. These findings remained consistent in subgroup analyses, enhancing the robustness of the results. Logistic regression analysis revealed that CAR was independently associated with an increased risk of death, with odds ratios (ORs) and 95% confidence intervals (CIs) for 28-day mortality (OR = 1.006, 95% CI = 1.003–1.010, P = 0.033), 60-day mortality (OR = 1.005, 95% CI = 1.001–1.008, P = 0.005), and 90-day mortality (OR = 1.004, 95% CI = 1.001–1.007, P = 0.016). Even after adjusting for confounding factors, the dose-response relationship between CAR levels and mortality risk persisted. We also observed significant age differences. In stratified analysis, CAR was significantly associated with increased mortality in IS critical patients aged 65 years and younger. This is consistent with the findings of DU et al. [ 44 ], who included 69 stroke patients aged 18–50 years and found that increased CAR was independently associated with an increased risk of adverse outcomes at 30 and 60 days (mRS score 2–6) in young stroke patients. In this study, we used six machine learning models and found that models containing CAR had better discrimination values for 28-day, 60-day, and 90-day all-cause mortality in IS patients compared to models without CAR. The AUC of machine learning models containing CAR was consistently higher than those without CAR. After analyzing AUC, DCA, and calibration curves, the Gradient Boosting model demonstrated superior overall performance compared to other models. We applied the SHAP method to the Gradient Boosting regression model to achieve the best predictive performance and interpretability. We identified several important variables related to mortality in IS patients, further validating the importance of CAR in predicting 28-day, 60-day, and 90-day all-cause mortality in IS patients. In summary, our study results and previous studies support CAR as a useful predictive biomarker for mortality in IS patients [ 45 – 48 ]. By monitoring CAR levels, doctors can identify individuals at higher risk of death earlier and make timely treatment decisions for intervention. This study has some limitations. First, it is based on retrospective data, which may introduce information bias. Second, although we adjusted for multiple potential confounding factors in the analysis, other unmeasured or uncontrolled confounding factors may still exist. For example, the use of anti-inflammatory drugs may affect CAR levels, but the MIMIC-IV and eICU-CRD databases do not contain information on pre-hospital medications. Additionally, the data in this study come from databases with racial restrictions, which means that the study results may have certain limitations when applied to multiple countries and ethnic groups. Although the predictive ability of CAR for mortality has been validated in both internal and external cohorts, further validation through other cohorts and prospective studies is needed to enhance the level of evidence. Declarations Ethics approval and consent to participate Approval for MIMIC-IV and eICU-CRD databases were granted by Physionet (https://physionet.org/). Given the retrospective design of this study, which utilized only existing data and did not influence clinical decision-making, the requirement for informed consent was waived. Therefore, no additional ethics approval or consent was needed for the analysis. All MIMIC-IV data are deidentified to comply with the Health Information Portability and Accountability Act (HIPAA) Safe Harbor provision. eICU-CRD data are deidentified to meet the safe harbor provision of the US Health Insurance Portability and HIPAA. The whole data collection and processing method obey in accordance with the Declaration of Helsinki Consent for publication Not applicable Availability of Data and Material Publicly available datasets used in this study can be downloaded from the mimic-IV database and eICU-CRD database. Further inquiries can be directed to the corresponding author. Competing interests Not applicable Funding Funded by the University-College Joint Foundation of Yunnan University of Chinese Medicine：NO.XYLH2024021. Authors' contributions BY W and L P conceived the study idea. L S, SC F and J Z performed the data analysis. G L, SP W, XL C and JF L participated in the preparation of figures and tables. BY W and J Z wrote the manuscript. L P, CX L and XL C revised the manuscript. All authors approved the submitted version. Acknowledgements Not applicable References Hilkens NA, Casolla B, Leung TW, de Leeuw FE. Stroke[J] Lancet. 2024;403(10446):2820–36. Gerstl JVE, Blitz SE, Qu QR, Yearley AG, Lassarén P, Lindberg R, et al. Global, Regional, and National Economic Consequences of Stroke[J]. Stroke. 2023;54(9):2380–9. Saini V, Guada L, Yavagal DR. Global Epidemiology of Stroke and Access to Acute Ischemic Stroke Interventions[J]. Neurology. 2021;97(20 Suppl 2):S6–16. Zhang J, Ling L, Xiang L, Li W, Bao P, Yue W. Role of the gut microbiota in complications after ischemic stroke[J]. Front Cell Infect Microbiol. 2024;14:1334581. Sluis WM, Hinsenveld WH, Goldhoorn RB, Potters LH, Bruggeman AA, van der Hoorn A, et al. Timing and causes of death after endovascular thrombectomy in patients with acute ischemic stroke[J]. Eur Stroke J. 2023;8(1):215–23. Nguyen TN, Abdalkader M, Fischer U, Qiu Z, Nagel S, Chen HS, et al. Endovascular management of acute stroke[J]. Lancet. 2024;404(10459):1265–78. Alsbrook DL, Di Napoli M, Bhatia K, Biller J, Andalib S, Hinduja A, et al. Neuroinflammation in Acute Ischemic and Hemorrhagic Stroke[J]. Curr Neurol Neurosci Rep. 2023;23(8):407–31. Eckart A, Struja T, Kutz A, Baumgartner A, Baumgartner T, Zurfluh S, et al. Relationship of Nutritional Status, Inflammation, and Serum Albumin Levels During Acute Illness: A Prospective Study[J]. Am J Med. 2020;133(6):713–e722717. Bian J, Guo S, Huang T, Li X, Zhao S, Chu Z, et al. CRP as a potential predictor of outcome in acute ischemic stroke[J]. Biomed Rep. 2023;18(2):17. Jiang Y, Yang Z, Wu Q, Cao J, Qiu T. The association between albumin and C-reactive protein in older adults[J]. Med (Baltim). 2023;102(34):e34726. Bi R, Shi Y, Li M, Liu X, Ma Z, Huang Y, et al. Association between serum albumin and severe impairment of activities of daily living in patients with stroke: a cross-sectional study[J]. Front Neurol. 2024;15:1501294. Bucci T, Pastori D, Pignatelli P, Ntaios G, Abdul-Rahim AH, Violi F, et al. Albumin Levels and Risk of Early Cardiovascular Complications After Ischemic Stroke: A Propensity-Matched Analysis of a Global Federated Health Network[J]. Stroke. 2024;55(3):604–12. Bafei SEC, Yang S, Chen C, Gu X, Mu J, Liu F, et al. Sex and age differences in the association between high sensitivity C-reactive protein and all-cause mortality: A 12-year prospective cohort study[J]. Mech Ageing Dev. 2023;211:111804. Bao Y, Yang J, Duan Y, Chen Y, Chen W, Sun D. The C-reactive protein to albumin ratio is an excellent prognostic predictor for gallbladder cancer[J]. Biosci Trends. 2021;14(6):428–35. Sabanoglu C, Inanc IH. C-reactive protein to albumin ratio predicts for severity of coronary artery disease and ischemia[J]. Eur Rev Med Pharmacol Sci. 2022;26(20):7623–31. Sonsöz MR, Karadamar N, Yılmaz H, Eroğlu Z, Şahin KK, Özateş Y, et al. C-Reactive Protein to Albumin Ratio Predicts In-hospital Mortality in Patients with Acute Heart Failure[J]. Turk Kardiyol Dern Ars. 2023;51(3):174–81. Wändell P, Carlsson AC, Larsson AO, Ärnlöv J, Ruge T, Rydell A. The predictive value of cardiovascular outcomes and mortality assessed by the C-reactive protein to albumin ratio in the UK Biobank[J]. BMC Cardiovasc Disord. 2024;24(1):326. Yuksel ME, Ozkan N, Avci E. C-reactive protein/albumin ratio greater than 7.1 is a good candidate to be used as an inflammation biomarker to predict perforation in appendicitis[J]. Eur Rev Med Pharmacol Sci. 2022;26(22):8333–41. Kaplan H, Cengiz G, Şaş S, Eldemir Y. Is the C-reactive protein-to-albumin ratio the most remarkable simple inflammatory marker showing active disease in patients with axial spondyloarthritis, psoriatic arthritis, and rheumatoid arthritis?[J]. Clin Rheumatol. 2023;42(11):2959–69. Kunutsor SK, Laukkanen JA. Serum C-reactive protein-to-albumin ratio is a potential risk indicator for pneumonia: Findings from a prospective cohort study[J]. Respir Med. 2022;199:106894. Zhou X, Fu S, Wu Y, Guo Z, Dian W, Sun H, et al. C-reactive protein-to-albumin ratio as a biomarker in patients with sepsis: a novel LASSO-COX based prognostic nomogram[J]. Sci Rep. 2023;13(1):15309. Rathore SS, Oberoi S, Iqbal K, Bhattar K, Benítez-López GA, Nieto-Salazar MA, et al. Prognostic value of novel serum biomarkers, including C-reactive protein to albumin ratio and fibrinogen to albumin ratio, in COVID-19 disease: A meta-analysis[J]. Rev Med Virol. 2022;32(6):e2390. Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, et al. MIMIC-IV, a freely accessible electronic health record dataset[J]. Sci Data. 2023;10(1):1. Pollard TJ, Johnson AEW, Raffa JD, Celi LA, Mark RG, Badawi O. The eICU Collaborative Research Database, a freely available multi-center database for critical care research[J]. Sci Data. 2018;5:180178. Sun T, Liu J, Yuan H, Li X, Yan H. Construction of a risk prediction model for lung infection after chemotherapy in lung cancer patients based on the machine learning algorithm[J]. Front Oncol. 2024;14:1403392. Lundberg SM, Erion G, Chen H, DeGrave A, Prutkin JM, Nair B, et al. From Local Explanations to Global Understanding with Explainable AI for Trees[J]. Nat Mach Intell. 2020;2(1):56–67. Bielewicz JE, Kurzepa J, Kamieniak P, Daniluk B, Szczepańska-Szerej A, Rejdak K. Clinical and biochemical predictors of late-outcome in patients after ischemic stroke[J]. Ann Agric Environ Med. 2020;27(2):290–4. Eto F, Nezu T, Aoki S, Kamimura T, Naito H, Shiga Y, et al. The association between nutritional intake one week after admission and outcome in acute ischemic stroke patients[J]. J Stroke Cerebrovasc Dis. 2022;31(11):106812. Shi K, Tian DC, Li ZG, Ducruet AF, Lawton MT, Shi FD. Global brain inflammation in stroke[J]. Lancet Neurol. 2019;18(11):1058–66. Wang R, Cao L, He Y, Zhang P, Feng L. Nutrition-associated markers and outcomes among patients receiving enteral nutrition after ischemic stroke: a retrospective cohort study[J]. BMC Neurol. 2024;24(1):303. Zhang G, Pan Y, Zhang R, Wang M, Meng X, Li Z, et al. Prevalence and Prognostic Significance of Malnutrition Risk in Patients With Acute Ischemic Stroke: Results From the Third China National Stroke Registry[J]. Stroke. 2022;53(1):111–9. Wang L, Yang L, Liu H, Pu J, Li Y, Tang L et al. C-Reactive Protein Levels and Cognitive Decline following Acute Ischemic Stroke: A Systematic Review and Meta-Analysis[J]. Brain Sci, 2023, 13(7). Jiang J, Tan C, Zhou W, Peng W, Zhou X, Du J, et al. Plasma C-Reactive Protein Level and Outcome of Acute Ischemic Stroke Patients Treated by Intravenous Thrombolysis: A Systematic Review and Meta-Analysis[J]. Eur Neurol. 2021;84(3):145–50. Bucci T, Sagris D, Harrison SL, Underhill P, Pastori D, Ntaios G, et al. C-reactive protein levels are associated with early cardiac complications or death in patients with acute ischemic stroke: a propensity-matched analysis of a global federated health from the TriNetX network[J]. Intern Emerg Med. 2023;18(5):1329–36. Kiboshi R, Satoh S, Mikami K, Kitajima M, Urushizaka M, Metoki N, et al. Serum Albumin, Body Mass Index, and Preceding Xa and P2Y12 Inhibitors Predict Prognosis of Recurrent Ischemic Stroke[J]. J Stroke Cerebrovasc Dis. 2021;30(5):105681. Wang C, Deng L, Qiu S, Bian H, Wang L, Li Y, et al. Serum Albumin Is Negatively Associated with Hemorrhagic Transformation in Acute Ischemic Stroke Patients[J]. Cerebrovasc Dis. 2019;47(1–2):88–94. Zhang Q, Lei YX, Wang Q, Jin YP, Fu RL, Geng HH, et al. Serum albumin level is associated with the recurrence of acute ischemic stroke[J]. Am J Emerg Med. 2016;34(9):1812–6. Thuemmler RJ, Pana TA, Carter B, Mahmood R, Bettencourt-Silva JH, Metcalf AK et al. Serum Albumin and Post-Stroke Outcomes: Analysis of UK Regional Registry Data, Systematic Review, and Meta-Analysis[J]. Nutrients, 2024, 16(10). Zhou H, Wang A, Meng X, Lin J, Jiang Y, Jing J, et al. Low serum albumin levels predict poor outcome in patients with acute ischaemic stroke or transient ischaemic attack[J]. Stroke Vasc Neurol. 2021;6(3):458–66. Xu T, Xia L, Wu Y, Xu Y, Xu X, Zhang W, et al. High ratio of C-reactive protein to albumin is associated with hemorrhagic transformation and poor functional outcomes in acute ischemic stroke patients after thrombolysis[J]. Front Aging Neurosci. 2023;15:1109144. Huang L, Zhang R, Ji J, Long F, Wang Y, Lu J, et al. Hypersensitive C-reactive protein-albumin ratio is associated with stroke-associated pneumonia and early clinical outcomes in patients with acute ischemic stroke[J]. Brain Behav. 2022;12(7):e2675. Guo C, Zheng P, Chen S, Wei L, Fu X, Fu Y, et al. Association between the C-reactive protein/albumin ratio and mortality in older Japanese patients with dysphagia[J]. Front Nutr. 2024;11:1370763. Zeng J, Liu J, Lu Y, Fu J, Han D, Chen J, et al. C-reactive protein to albumin ratio and Glasgow Coma Scale score as the predictors for weaning outcomes in traumatic brain injury[J]. Ann Med. 2025;57(1):2472866. Du Y, Zhang J, Li N, Guo J, Liu X, Bian L, et al. Association between the C-reactive protein to albumin ratio and adverse clinical prognosis in patients with young stroke[J]. Front Neurol. 2022;13:989769. Jang JH, Hong S, Ryu JA. Prognostic Value of C-Reactive Protein and Albumin in Neurocritically Ill Patients with Acute Stroke[J]. J Clin Med, 2022, 11(17). Kocatürk M, Kocatürk Ö. Assessment of relationship between C-reactive protein to albumin ratio and 90-day mortality in patients with acute ischaemic stroke[J]. Neurol Neurochir Pol. 2019;53(3):205–11. Yu D, Guo G, Wan F, Hu B. The association between C-reactive protein to albumin ratio and adverse outcomes in acute ischemic stroke patients: A study in the Korean population[J]. Heliyon. 2024;10(20):e39212. Yuan J, Cheng Y, Han X, Zhu N, Ma W, Li J, et al. Association between C-reactive protein/albumin ratio and all-cause mortality in patients with stroke: Evidence from NHANES cohort study[J]. Nutr Metab Cardiovasc Dis. 2024;34(10):2305–14. Tables Tables 1 to 5 are available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files table1.xlsx Table 1 Baseline Information. Baseline clinic features by quantile of CAR index. table2.xlsx Table 2 Baseline Information. Baseline clinic features by survivors and non-survivors at 28, 60 and 90 days during hospitalization . table3.xlsx Table 3 Logistic regression for 28-day, 60-day and 90-day all-cause mortality incidence by different adjusted models. table4.xlsx Table 4 Internal and External validation. The model performance index for different machine learning models for 28-day, 60-day and 90-day all-cause morality, respective in train and validation sets. table5.xlsx Table 5 Integrated discrimination improvement(IDI) analysis. For 28-day, 60-day and 90-day all-cause morality, Compared CAR index with GCS score, SASPII score, OASIS score and Charlson Score. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 05 Sep, 2025 Editor invited by journal 10 Aug, 2025 Editor assigned by journal 28 Jul, 2025 Submission checks completed at journal 28 Jul, 2025 First submitted to journal 22 Jul, 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-7185510","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":512985996,"identity":"cc62c8da-238f-4351-adf5-c5723f2a6705","order_by":0,"name":"Wang Binyang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAxElEQVRIiWNgGAWjYBACfv7mAwcSKmrkGNsbiNQiOeNY4oEHZ44ZM/ccIFKLwYEc44MPW5gT22ckEGtLwxmDA4kNbIm9Mx9vvMFQYxNNUAs/c1vBgcQdMsYzZ6cVWzAcS8ttIGzL4Q0HEs+wyW6cnWMmwdhwmLAWgwMJQIe1MTPuv3mGaC0pYC2KjTN4iNQCDOSEAwnAQGbsAfolgRi/AKPy8Mcf4Kg8vPHGhxobwlpQHCmRQIpyiBZSdYyCUTAKRsHIAAB9s0tvrRUQQgAAAABJRU5ErkJggg==","orcid":"","institution":"Kunming Medical University","correspondingAuthor":true,"prefix":"","firstName":"Wang","middleName":"","lastName":"Binyang","suffix":""},{"id":512985997,"identity":"672da4d6-3b48-4850-9fb7-d85fac519685","order_by":1,"name":"Zhong Jing","email":"","orcid":"","institution":"Yunnan University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Zhong","middleName":"","lastName":"Jing","suffix":""},{"id":512985998,"identity":"c6b9bc4a-21a6-4a3a-b284-30d8b0de6ef4","order_by":2,"name":"Shao Lu","email":"","orcid":"","institution":"Yunnan University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Shao","middleName":"","lastName":"Lu","suffix":""},{"id":512985999,"identity":"480c554c-37b0-4df2-b652-ec38697a0130","order_by":3,"name":"Fan shuochen","email":"","orcid":"","institution":"Yunnan University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Fan","middleName":"","lastName":"shuochen","suffix":""},{"id":512986000,"identity":"a6ceeced-47f1-4c5b-878a-cd8a51d287e8","order_by":4,"name":"Wang Shiping","email":"","orcid":"","institution":"The Yunnan University Affiliated Hospital","correspondingAuthor":false,"prefix":"","firstName":"Wang","middleName":"","lastName":"Shiping","suffix":""},{"id":512986001,"identity":"03894190-29dc-42af-bd36-819983575b45","order_by":5,"name":"Li Geng","email":"","orcid":"","institution":"Yunnan University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Li","middleName":"","lastName":"Geng","suffix":""},{"id":512986002,"identity":"f6043dda-57d3-481e-a4f5-20c909a5a4d8","order_by":6,"name":"Li jianfeng","email":"","orcid":"","institution":"Second Affiliated Hospital of Yunnan University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Li","middleName":"","lastName":"jianfeng","suffix":""},{"id":512986003,"identity":"5d8f424c-08f4-4b4d-9a43-58b270e1e545","order_by":7,"name":"Zhang yuping","email":"","orcid":"","institution":"The Cooperation of Chinese and Western Medicine Hospitital of Yunnan Province","correspondingAuthor":false,"prefix":"","firstName":"Zhang","middleName":"","lastName":"yuping","suffix":""},{"id":512986004,"identity":"c3b5ac55-f8c0-497f-af88-3542328c17f5","order_by":8,"name":"Li chuanxiong","email":"","orcid":"","institution":"The Yunnan University Affiliated Hospital","correspondingAuthor":false,"prefix":"","firstName":"Li","middleName":"","lastName":"chuanxiong","suffix":""},{"id":512986005,"identity":"2fd4d3f4-65d2-42e8-b594-d8c9ef6b0d2a","order_by":9,"name":"Chen Xiaolin","email":"","orcid":"","institution":"Second Affiliated Hospital of Yunnan University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Chen","middleName":"","lastName":"Xiaolin","suffix":""},{"id":512986006,"identity":"28f9628a-96fb-4320-ae6c-c9b6a7940fb8","order_by":10,"name":"Pan Lei","email":"","orcid":"","institution":"Second Affiliated Hospital of Yunnan University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Pan","middleName":"","lastName":"Lei","suffix":""}],"badges":[],"createdAt":"2025-07-22 09:53:23","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7185510/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7185510/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":91190210,"identity":"c0b45148-eda4-44e3-a290-5ccd3bed72a3","added_by":"auto","created_at":"2025-09-12 14:34:30","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":620485,"visible":true,"origin":"","legend":"\u003cp\u003eRestricted cubic spline curves for the CAR index. MODEL1 (A,E,I):None;MODEL2(B,F,J):age, gender, length of stay in hospital, length of stay in ICU; MODEL3(C,G,K):age, gender, length of stay in hospital, length of stay in ICU,GCS score, SAPSII score, SIRS score, Charlson Score;MODEL4(D,H,L):age, gender, length of stay in hospital, length of stay in ICU,GCS score, SAPSII score, SIRS score, Charlson Score, Myocardial Infarct, Congestive Heart Failure, Peripheral Vascular Disease, Cerebrovascular Disease, Chronic Pulmonary Disease, Diabetes without Chronic Comorbities, Diabetes with Chronic Comorbities, Paraplegia, Renal Disease\u003c/p\u003e","description":"","filename":"fig1RCS.png","url":"https://assets-eu.researchsquare.com/files/rs-7185510/v1/9f8d304d9022479be10c84b1.png"},{"id":91194138,"identity":"8a251829-a8bb-4c28-a3ef-810bb6c6a915","added_by":"auto","created_at":"2025-09-12 14:50:30","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1860847,"visible":true,"origin":"","legend":"\u003cp\u003eForest plots for three outcomes. A:Subgroup analysis for 28-days all-cause morality; B:Subgroup analysis for 60-days all-cause morality; C:Subgroup analysis for 90-days all-cause morality.\u003c/p\u003e","description":"","filename":"fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-7185510/v1/9f8896f4f1223ab014e1a75c.png"},{"id":91190225,"identity":"f6cc3634-2c58-4597-843f-d884f7236084","added_by":"auto","created_at":"2025-09-12 14:34:30","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":2048195,"visible":true,"origin":"","legend":"\u003cp\u003eModel Performance.(A-E) MIMIC-IV cohort for 28-day, 60-day and 90-day all-cause morality, and train with CAR index; (B-F) MIMIC-IV cohort for 28-day, 60-day and 90-day all-cause morality, and train without CAR index; (G-I) eICU-CRD cohort for 28-day, 60-day and 90-day all-cause morality, and train with CAR index;(J-L) The calibration curve for different for 28-day, 60-day and 90-day all-cause morality;(M-O) The decision curve for different for 28-day, 60-day and 90-day all-cause morality.\u003c/p\u003e","description":"","filename":"fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-7185510/v1/c1fc97667b1795ddbbf5d77b.png"},{"id":91192528,"identity":"fd8202dd-1f95-4124-bfa3-0d1fa607f43a","added_by":"auto","created_at":"2025-09-12 14:42:30","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1222705,"visible":true,"origin":"","legend":"\u003cp\u003eInternal and External validation. The model performance index for different machine learning models for 28-day, 60-day and 90-day all-cause morality, respective in train and validation sets.\u003c/p\u003e","description":"","filename":"fig4SHAP.png","url":"https://assets-eu.researchsquare.com/files/rs-7185510/v1/e9b83ccd1afcfbc7755062b0.png"},{"id":91196798,"identity":"3a7b6d22-e2ef-4d0d-8122-b3108d831898","added_by":"auto","created_at":"2025-09-12 15:06:35","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6310527,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7185510/v1/810ae7c2-a1b0-423a-9d83-327b4697c07a.pdf"},{"id":91190203,"identity":"fe57aa52-457b-43ee-acc3-090e1152c0f5","added_by":"auto","created_at":"2025-09-12 14:34:30","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":16395,"visible":true,"origin":"","legend":"\u003cp\u003eTable 1 Baseline Information. Baseline clinic features by quantile of CAR index.\u003c/p\u003e","description":"","filename":"table1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7185510/v1/8a71e72cd75aa55987121ec6.xlsx"},{"id":91190205,"identity":"db2a8b8d-f615-4b01-86b1-78e1c5d5da68","added_by":"auto","created_at":"2025-09-12 14:34:30","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":16967,"visible":true,"origin":"","legend":"\u003cp\u003eTable 2 Baseline Information. Baseline clinic features by survivors and non-survivors at 28, 60 and 90 days during hospitalization .\u003c/p\u003e","description":"","filename":"table2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7185510/v1/d13418e56cc11393bbded23f.xlsx"},{"id":91190208,"identity":"0974786a-c3fb-4230-8e91-2755cfcbabb5","added_by":"auto","created_at":"2025-09-12 14:34:30","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":11877,"visible":true,"origin":"","legend":"\u003cp\u003eTable 3 Logistic regression for 28-day, 60-day and 90-day all-cause mortality incidence by different adjusted models.\u003c/p\u003e","description":"","filename":"table3.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7185510/v1/83fee67d310aac6480766802.xlsx"},{"id":91192527,"identity":"8a1dcf5e-38db-45ff-ae1f-27fb14d20db3","added_by":"auto","created_at":"2025-09-12 14:42:30","extension":"xlsx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":10260,"visible":true,"origin":"","legend":"\u003cp\u003eTable 4 Internal and External validation. The model performance index for different machine learning models for 28-day, 60-day and 90-day all-cause morality, respective in train and validation sets.\u003c/p\u003e","description":"","filename":"table4.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7185510/v1/39b207593d19cd3fda7ca74c.xlsx"},{"id":91192531,"identity":"4596c398-346e-4b21-a2c9-954cfc669607","added_by":"auto","created_at":"2025-09-12 14:42:30","extension":"xlsx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":22241,"visible":true,"origin":"","legend":"\u003cp\u003eTable 5 Integrated discrimination improvement(IDI) analysis. For 28-day, 60-day and 90-day all-cause morality, Compared CAR index with GCS score, SASPII score, OASIS score and Charlson Score.\u003c/p\u003e","description":"","filename":"table5.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7185510/v1/dceaeb5c9e56502eb5c0f1b1.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"The C-Reactive Protein-to-Albumin Ratio (CAR) and All-Cause Mortality in Critically Ill Ischemic Stroke Patients: A Retrospective Analysis of the MIMIC-IV and eICU-CRD Databases","fulltext":[{"header":"Introduction","content":"\u003cp\u003eStroke represents the second leading cause of global mortality and third leading cause of disability worldwide. In 2019 alone, stroke affected over 100\u0026nbsp;million individuals, with 12\u0026nbsp;million new cases and approximately 7\u0026nbsp;million deaths [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Stroke causes significant macroeconomic losses. In 2019, the total welfare loss value by stroke worldwide was 2,059.67\u0026nbsp;billion US dollars, accounting for 1.66% of the global GDP, while the global welfare loss value by ischemic stroke was 964.51\u0026nbsp;billion US dollars, accounting for 0.78% of the global GDP [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Projections indicate stroke-related direct medical costs in the United States will surge from 36.7 billionin in 2015 to 94.3\u0026nbsp;billion by 2035 [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eComplications such as post-stroke cognitive impairment, post-stroke depression, hemorrhagic transformation, gastrointestinal dysfunction, cardiovascular events, and post-stroke infection often occur after ischemic stroke, which affect the prognosis of the disease and lead to progressive neurological deficits and high mortality [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Despite the current use of treatment methods such as intravenous thrombolysis, endovascular thrombectomy, cell protection or adjuvant drugs, the risk of adverse clinical outcomes in IS patients remains high, especially in critically ill patients [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Early and effective intervention and management are crucial. Timely treatment based on monitoring relevant indicators significantly enhances patient prognosis. Utilizing these laboratory parameters to predict clinical outcomes in IS patients facilitates optimized clinical management and prognosis assessment.\u003c/p\u003e\u003cp\u003eNeuroinflammation after stroke is a major factor leading to poor functional outcomes and death. In clinical practice, it is necessary to monitor the inflammatory levels of patients [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. C-reactive protein (CRP) is an acute-phase protein used as an inflammatory marker. In stroke patients, elevated CRP levels may indicate an enhanced inflammatory response, which is associated with poor prognosis and increased risk of death [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].Current data suggest that a high level of CRP within 24 hours after the onset of IS is independently associated with poor functional outcomes after acute ischemic events [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Albumin is the most widely studied protein for diagnosing malnutrition. Hypoalbuminemia (\u0026lt; 3.5 g/dL) is defined as an indicator of malnutrition to screen for malnourished patients, and hypoalbuminemia may provide as a potential predictive biomarker for inflammation [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. There is evidence that for every one-unit increase in serum albumin level (g/L), the impairment of activities of daily living in stroke patients decreases by 7% [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Recent studies confirm that ​​elevated C-reactive protein (CRP) and hypoalbuminemia​​ independently predict mortality and are widely utilized to anticipate complications and fatal outcomes in critically ill patients [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Notably​​, the CRP/Albumin ratio (CAR) surpasses either biomarker alone in ​​accurately reflecting systemic inflammation and predicting prognosis​​ across cardiovascular, cerebrovascular, and oncological diseases [\u003cspan additionalcitationids=\"CR15 CR16\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e–\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. ​As a novel inflammatory prognostic score​​, CAR has further demonstrated utility in forecasting outcomes for diverse inflammatory conditions—including sepsis, pneumonia, multiple arthritides, perforated appendicitis, and COVID-19 [\u003cspan additionalcitationids=\"CR19 CR20 CR21\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e–\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Nevertheless​​, the association between CAR and ​​all-cause mortality in critically ill IS patients remains unestablished​​.\u003c/p\u003e\u003cp\u003e​\u003cb\u003e​\u003c/b\u003eLeveraging the MIMIC-IV database, this study investigates the association between the CAR and 28-, 60-, and 90-day all-cause mortality in critically ill IS patients. We will develop a machine learning-based predictive model and employ the SHAP framework to interpret feature importance. This approach aims to assist clinicians in quantitatively assessing disease severity and implementing personalized early interventions to mitigate mortality risk in this vulnerable population\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cb\u003eStudy Population\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe original data were obtained from the MIMIC-IV database, which is a contemporary electronic health record dataset established by Beth Israel Deaconess Medical Center and the Massachusetts Institute of Technology, covering information on over 70,000 adult patients admitted to the emergency department or intensive care unit at Beth Israel Deaconess Medical Center in Boston, Massachusetts, USA, from 2008 to 2019 [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. eICU is a multi-center intensive care unit database containing high-resolution data on over 200,000 patients admitted to 208 different ICUs in the United States from 2014 to 2015. The source hospital of the MIMIC-IV did not participate in the eICU program [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. The author (Binyang Wang: 13488206) obtained access to the databases. All patient records in both databases were fully de-identified, so informed consent and ethical approval were not required..\u003c/p\u003e\u003cp\u003e\u003cb\u003eInclusion criteria\u003c/b\u003e\u003c/p\u003e\u003cp\u003ePatients diagnosed with IS according to the 9th and 10th editions of the International Classification of Diseases were included in this study.\u003c/p\u003e\u003cp\u003eExclusion criteria were as follows: (1) For patients with multiple hospitalizations, only the data from the first hospitalization were included; (2) patients who died within 28 days after admission; (3) patients with severe diseases such as end-stage renal failure, liver cirrhosis, or cancer; (4) patients with an ICU stay \u0026lt; 3 hours; and (5) patients lacking CRP and albumin data at admission. Finally, a total of 2,664 patients were included in this study and were divided into four groups based the quartiles of CAR index (Q1 n = 667, Q2 n = 665, Q3 n = 666, and Q4 n = 666).\u003c/p\u003e\u003cp\u003e\u003cb\u003eData Extraction\u003c/b\u003e\u003c/p\u003e\u003cp\u003eCAR is a continuous variable calculated as the serum CRP concentration (mg/dl) closest to admission divided by the serum albumin concentration (g/dl) closest to admission. CAR was then divided into quartiles and classified as follows: Q1: 1.92, Q2: 12.19, Q3: 33.58, and Q4: 155.00.\u003c/p\u003e\u003cp\u003eThe extracted potential variables can be classified into five major categories: (1) Demographic data, such as age and gender. (2) Comorbidities, including myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic lung disease, uncomplicated diabetes, complicated diabetes, paraplegia, kidney disease, and sepsis. (3) Laboratory indicators, including the C-reactive protein, albumin level, C-reactive protein-to-albumin ratio, international normalized ratio, prothrombin time, partial thromboplastin time, hemoglobin level, hematocrit, mean corpuscular ,mean corpuscular hemoglobin concentration level, mean corpuscular volume, platelet count, red blood cell count, red blood cell distribution width, white blood cell count, anion gap, bicarbonate, blood urea nitrogen, glucose, serum creatinine, serum calcium, serum chloride, serum sodium, and serum potassium. (4) Admission disease severity scores, including the Glasgow Coma Scale (GCS), Simplified Acute Physiology Score II (SAPS-II), Systemic Inflammatory Response Syndrome (SIRS), and Charlson. (5) Hospitalization and mortality-related time, such as length of hospital stay, length of ICU stay, 28-day all-cause mortality, 60-day all-cause mortality, and 90-day all-cause mortality. All laboratory parameters extracted from the database were measured for the first time after admission to the ICU. For missing data, first, feature columns with a missing data rate over 30% were excluded as they contained low information. Then,the remaining missing values of the retained features were subsequently filled using the mode imputation method. This approach eliminates high-missingness dimensions through threshold screening and selects the mode for filling based on the distribution characteristics of the variables, balancing data integrity and distribution stability.\u003c/p\u003e\u003cp\u003e\u003cb\u003eClinical Outcomes\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe primary outcome of this study was all-cause mortality at 28, 60, and 90 days.\u003c/p\u003e\u003cp\u003e\u003cb\u003eModel Construction\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe Boruta algorithm is a feature selection and wrapping algorithm based on random forests. It assesses the importance of features by generating values for each feature in the dataset and comparing them with the values of the corresponding \"shadow features\" [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. We used the Boruta algorithm to select clinical features and built six machine learning algorithm prediction models: Random Forest, AdaBoost, Gradient Boosting, XGBoost, LightGBM, and Neural Network. The models were established on the training set, and internal and external validation sets were used to validate the best model. The performance of the prediction models was evaluated using the area under the receiver operating characteristic curve (AUC) and accuracy. Additionally, decision curve analysis (DCA) and calibration curves were plotted to demonstrate the true clinical utility, and the integrated discrimination improvement (IDI) was calculated. The optimal model was used to analyze all-cause mortality at 28, 60, and 90 days, and the SHAP method was employed to visualize the contribution of each feature to the prediction results.\u003c/p\u003e\u003ch2\u003eStatistical Analysis\u003c/h2\u003e\u003cp\u003eThe data collected for analysis were divided into two major categories: categorical variables and continuous variables. Continuous variables were presented as mean (standard deviation(SD) and were compared using tests or no-parametric tests, as appropriate. Categorical variables were presented as frequency and percentage(%) and were compared between groups using chi-square tests or Fisher's exact tests. To assess the correlation between CAR and the risk of death at 28, 60, and 90 days, multivariate logistic regression analysis was conducted to calculate odds ratios (ORs) and their corresponding 95% confidence intervals (CIs) to quantify the impact of CAR on these outcomes, with adjustment for confounding variables. The restricted cubic spline (RCS) method was used to explore the potential non-linear association between CAR and the outcome. Subgroup analyses were also conducted to verify the association between CAR and 28-day, 60-day, and 90-day mortality within each subgroup. Six machine learning models were trained on the training set and validation set, and the performance of the models was compared. SHAP values were used to explain the best-performing model. Statistical significance was defined as a two-sided P value \u0026lt; 0.05. All the statistical analyses were performed using R software and Python.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eBaseline characteristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study included 2,664 eligible participants stratified by CAR quartiles from the MIMIC-IV database. The baseline characteristics of the participants are shown in Table 1. The average age of the participants was 67.493\u0026thinsp;\u0026plusmn;\u0026thinsp;14.746 years, 53.23% were male, and the average CAR index was 22.173\u0026thinsp;\u0026plusmn;\u0026thinsp;26.011. The 28-day, 60-day, and 90-day mortality rates were 42.00%, 46.21%, and 55.43%, respectively.\u003c/p\u003e\n\u003cp\u003eThere were significant differences in gender, length of hospital stay, length of ICU stay, CRP, albumin, international normalized ratio, prothrombin time, partial thromboplastin time, blood urea nitrogen, serum calcium, serum chloride, serum sodium, serum potassium, myocardial infarction, congestive heart failure, peripheral vascular disease, chronic lung disease, diabetes with chronic complications, paraplegia, sepsis, kidney disease, hematocrit, hemoglobin, mean corpuscular hemoglobin concentration, platelet count, red blood cell count, red blood cell distribution width, white blood cell count, and CAR among the different CAR quartiles (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e\n\u003cp\u003eCompared with the other groups, the highest CAR level (Q4) had longer hospital and ICU stays, higher levels of CRP, international normalized ratio, prothrombin time, partial thromboplastin time, serum calcium, and serum sodium;,and red blood cell distribution width; and higher white blood cell count, while lower levels of albumin, blood urea nitrogen, serum potassium, hematocrit, hemoglobin, mean corpuscular hemoglobin concentration, and red blood cell count were detected. The SAPS II score, SIRS score, and Charlson score at admission were higher, and the GCS score was lower in the Q4 group, indicating that as CAR level, the severity of disease scores also increased. Additionally, the Q4 group had the longest hospital and ICU stays and higher prevalence of myocardial infarction, congestive heart failure, peripheral vascular disease, chronic lung disease, diabetes with chronic complications, paraplegia, sepsis, and kidney disease. Compared with the other groups, the Q4 group had significantly higher mortality rates at 28 days (45.195% vs. 45.796% vs. 44.361% vs. 32.684%, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), 60 days (51.351% vs. 49.850% vs. 47.970% vs. 35.682%, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and 90 days (52.853% vs. 50.901% vs. 48.722% vs. 35.982%, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e\n\u003cp\u003eThe differences in baseline characteristics between survivors and non-survivors at 28, 60 and 90 days during hospitalization are shown in Table 2. Patients in the non-survivor group were older and more likely to be male, had a higher severity of disease, and had shorter hospital and intensive care unit (ICU) stays. The prevalence of myocardial infarction, congestive heart failure, peripheral vascular disease, chronic lung disease, diabetes with chronic comorbidities, paraplegia, sepsis, and kidney disease was significantly higher in the non-survivor group. Compared with the survivor group, the non-survivor group had worse coagulation function, more severe anemia, and more prominent electrolyte disorders, with higher anion gap, CRP, WBC, and creatinine values. Additionally, CAR level in non-survivors was significantly higher than that in survivors (28 days: 24.359 vs 20.590,60 days: 24.754 vs 19.954,90 days: 24.859 vs 19.781,P\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThe impact of CAR index on clinical outcomes\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe used continuous and categorical variable logistic regression models to study the independent effect of CAR on mortality and found that a higher CAR was positively correlated with an increased risk of death in critically ill IS patients (Table\u0026nbsp;3). Model 1 was unadjusted, Model 2 was adjusted for age, gender, hospital stay, and ICU stay; model 3: adjusted for age, gender, hospital stay, ICU stay, GCS score, SAPSll score, SIRS score, and Charlson score; model 4: age, gender, hospital stay, ICU stay, GCS score, SAPSll score, SIRS score, Charlson score, myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, chronic lung disease, diabetes without chronic comorbidities, diabetes with chronic comorbidities, paraplegia, and kidney disease were adjusted.\u003c/p\u003e\n\u003cp\u003eThe results showed that when CAR was a continuous variable, it was independently associated with increased 28-day mortality (OR\u0026thinsp;=\u0026thinsp;1.006, 95% CI\u0026thinsp;=\u0026thinsp;1.003\u0026ndash;1.010, P\u0026thinsp;=\u0026thinsp;0.033), 60-day mortality (OR\u0026thinsp;=\u0026thinsp;1.005, 95% CI\u0026thinsp;=\u0026thinsp;1.001\u0026ndash;1.008, P\u0026thinsp;=\u0026thinsp;0.005), and 90-day mortality (OR\u0026thinsp;=\u0026thinsp;1.004; 95% CI\u0026thinsp;=\u0026thinsp;1.001\u0026ndash;1.007, P\u0026thinsp;=\u0026thinsp;0.016), and all were significant risk factors. These results were further confirmed in the fully adjusted model 4. Specifically, when CAR was a nominal variable, the OR for 28-day mortality in the highest CAR quartile was 1.510, 95% CI: 1.163\u0026ndash;1.962, P\u0026thinsp;=\u0026thinsp;0.002; the OR for 60-day all-cause mortality was 1.368, 95% CI: 1.065\u0026ndash;1.759, P\u0026thinsp;=\u0026thinsp;0.014; and the OR for 90-day all-cause mortality was 1.331, 95% CI: 1.038\u0026ndash;1.709, P\u0026thinsp;=\u0026thinsp;0.024, all compared with the lowest quartile. Moreover, the risk of 28-day mortality, 60-day mortality, and 90-day mortality increased with the increase in CAR index quartiles. (Detailed information is shown in Table\u0026nbsp;3.)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNonlinear relationship between CAR index and clinical outcomes\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe used RCS curve analysis to reveal the nonlinear relationship between CAR and all-cause mortality at different time points (28 days, 60 days, and 90 days) (Fig. 1). When evaluating relationship between CAR and 28-day all-cause mortality in critically ill IS patients was evaluated, all the models showed significant nonlinearity (model 1: P-nonlinear\u0026thinsp;\u0026lt;\u0026thinsp;0.001; Model 2: P-nonlinear\u0026thinsp;\u0026lt;\u0026thinsp;0.001; Model 3: P-nonlinear\u0026thinsp;=\u0026thinsp;0.0018; Model 4: P-nonlinear\u0026thinsp;=\u0026thinsp;0.0139). CAR showed a significant nonlinear relationship with 28-, 60-, and 90-day all-cause mortality in the Model 1 and in partially adjusted Models 2 and 3. As the confounding factors were gradually adjusted, the nonlinear relationship weakened. However, the nonlinear correlation analysis results of CAR and 60-day and 90-day mortality indicated a linear association in the fully adjusted Model 4. That is, the all-cause mortality of critical IS patients at 60 and 90 days increased linearly with the increase of CAR. (P-nonlinear\u0026thinsp;=\u0026thinsp;0.1916, P-nonlinear\u0026thinsp;=\u0026thinsp;0.1921)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSubgroup analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo investigate potential variations within specific populations, logistic regression analysis was conducted for various subgroups, including age, gender, congestive heart failure, kidney disease, chronic lung disease, dementia, cerebrovascular disease, diabetes with or without chronic complications, sepsis, peripheral vascular disease, myocardial infarction, GCS Score, SAPSII Score, SIRS Score, and Charlson Score.\u003c/p\u003e\n\u003cp\u003eThe forest plot (Fig. 2) showed that CAR was significantly associated with 28-(Fig. 22A), 60-(Fig. 2B), and 90-day(Fig. 2C) all-cause mortality in critical IS patients aged\u0026thinsp;\u0026lt;\u0026thinsp;=\u0026thinsp;65 years, SAPSII Score\u0026thinsp;\u0026lt;\u0026thinsp;=\u0026thinsp;35, and Charlson Score\u0026thinsp;\u0026lt;\u0026thinsp;=\u0026thinsp;5 (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Additionally, CAR was significantly associated with 28-day all-cause mortality in IS patients with cerebrovascular disease (P\u0026thinsp;=\u0026thinsp;0.038).\u003c/p\u003e\n\u003cp\u003eThe interaction analysis indicated that CAR seemed to more accurately predict 28-day all-cause mortality in patients aged\u0026thinsp;\u0026lt;\u0026thinsp;=\u0026thinsp;65, SAPSII Score\u0026thinsp;\u0026lt;\u0026thinsp;=\u0026thinsp;35, and Charlson Score\u0026thinsp;\u0026lt;\u0026thinsp;=\u0026thinsp;5 (P-interaction\u0026thinsp;=\u0026thinsp;0.049, 0.002, 0.004). Moreover, CAR more accurately predicted 60-day and 90-day all-cause mortality in patients with a SAPSII Score\u0026thinsp;\u0026lt;\u0026thinsp;=\u0026thinsp;35 and a Charlson score\u0026thinsp;\u0026lt;\u0026thinsp;=\u0026thinsp;5 (P-interaction\u0026thinsp;\u0026lt;\u0026thinsp;0.05). There were no significant differences in 28-, 60-, and 90-day all-cause mortality based on gender, presence of congestive heart failure, kidney disease, chronic lung disease, dementia, cerebrovascular disease, diabetes with or without chronic complications, sepsis, peripheral vascular disease, myocardial infarction, GCS Score, and SIRS Score (P\u0026thinsp;\u0026gt;\u0026thinsp;0.05).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel construction and performance comparison\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFeature selection based on the Boruta algorithm was used to select clinical features and construct six machine learning models: Random Forest, AdaBoost, Gradient Boosting, XGBoost, LightGBM, and Neural Network to predict the mortality risk of critical IS patients. Figure 3 shows the performance of various models. The results indicated that among the six prediction models for 28-, 60-, and 90-day all-cause mortality in the training set, the machine learning models containing CAR had better fit and higher AUC values.\u003c/p\u003e\n\u003cp\u003eRegarding 28-day all-cause mortality (Fig. 3A-M), in the training cohort, Random Forest with CAR showed the best model fit, with an area under the curve (AUC) of 0.80(Fig. 4A); without the CAR(Fig. 3B), Gradient Boosting showed the best model fit, with an AUC of 0.77. In the eICU-CRD validation cohort(Fig. 3G), XGBoost had the best fit (AUC\u0026thinsp;=\u0026thinsp;0.79). Regarding the 60-day all-cause mortality rate, in the training cohort(Fig. 3C), the Gradient Boosting model was the best in CAR model (AUC\u0026thinsp;=\u0026thinsp;0.75), and in the model without the CAR(Fig. 3D), Gradient Boosting was also the best (AUC\u0026thinsp;=\u0026thinsp;0.72). In the eICU-CRD validation cohort(Fig. 3H), LightGBM was the best (AUC\u0026thinsp;=\u0026thinsp;0.73), followed by XGBoost and Gradient Boosting (AUC\u0026thinsp;=\u0026thinsp;0.72).\u003c/p\u003e\n\u003cp\u003eRegarding the 90-day all-cause mortality rate, in the training cohort(Fig. 3E), Random Forest and Gradient Boosting were the best in CAR model (AUC\u0026thinsp;=\u0026thinsp;0.75), and in the model without the CAR(Fig. 3F), Neural Network was the best (AUC\u0026thinsp;=\u0026thinsp;0.73). In the eICU-CRD validation cohort(Fig. 4I), Random Forest was the best (AUC\u0026thinsp;=\u0026thinsp;0.80), followed by Gradient Boosting (AUC\u0026thinsp;=\u0026thinsp;0.79). We evaluated the accuracy of each model in predicting the 28-day, 60-day, and 90-day all-cause mortality probabilities of critical IS patients by analyzing the calibration curves and clinical decision curves of the training set(Fig. 3J-O). The results showed that the calibration curve of Gradient Boosting had a good fit, indicating high consistency between the model prediction and the actual incidence rate. In terms of clinical applicability, the Gradient Boosting model demonstrated a robust net benefit across a wide range of threshold probabilities.\u003c/p\u003e\n\u003cp\u003eTable 4 shows the performance of the six models in the training and test sets. In the 28-day model, Gradient Boosting had the highest accuracy in the training set (71.5%), and in the validation set, Gradient Boosting had an accuracy of 88.37%, second only to Adaboost (89.53%). In the 60-day model, Gradient Boosting had the highest accuracy in the training set (69.25%), and in the validation set, Random Forest had the highest accuracy (91.86%). In the 90-day model, Random Forest had the highest accuracy in both the training set (69%) and the validation set (93.02%). Although the accuracy of Gradient Boosting was not the highest in the validation set, it was all above 80%, which could provide strong support for doctors\u0026apos; clinical decisions. In summary, the Gradient Boosting model demonstrated excellent overall performance.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThe incremental effect of CAR index\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe calculated the IDI of the scoring tools (GCS, SAPSII, SIRS, OASIS, Charlson) and analyzed the impact of CAR on their predictive ability (Table\u0026nbsp;5). This study showed that in predicting 28-day and 60-day all-cause mortality rates, CAR significantly enhanced the incremental of the GCS, SIRS, OASIS, and Charlson scores (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05). However, the SAPSII score did not have a statistically significant incremental improvement (P\u0026thinsp;\u0026gt;\u0026thinsp;0.05). In predicting the 90-day all-cause mortality rate, CAR significantly enhanced the predictive accuracy of all the scoring tools (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel interpretation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn recent years, many machine learning models have been applied to predict adverse outcomes in IS patients. The lack of interpretability limits the application of more powerful machine learning methods in medical decision support. SHAP combines optimal credit assignment with local explanations, enabling intuitive interpretation of the importance of individual variables in the model [26]. Therefore, we used the SHAP algorithm to graphically demonstrate the important influencing features of the Gradient Boosting model in predicting mortality (Fig. 4).\u003c/p\u003e\n\u003cp\u003eThe results Show that length of hospital stay, Charlson Score, SAPS-II score, red blood cell distribution width, blood urea nitrogen, age, partial thromboplastin time, platelet count, prothrombin time, albumin, CRP, glucose, red blood cell count, congestive heart failure, and CAR were the main influencing factors for predicting 28-day, 60-day, and 90-day mortality.\u003c/p\u003e\n\u003cp\u003eTo determine the main predictors of all-cause mortality in IS patients, we calculated the 15 most important features of the best machine learning model. SHAP summary bar charts B, E, and H show the top 15 most important features in the 28-day, 60-day, and 90-day prediction models, with variables listed in descending order of importance. Length of hospital stay is the most important predictor of 28-day all-cause mortality(Fig. 4A,B), followed by Charlson Score, SAPS-II score, red blood cell distribution width, and blood urea nitrogen. Compared with other factors, the SAPS-II score has a greater impact on 60-day and 90-day all-cause mortality(Fig. 4D-E,G-H), than other factors followed by the Charlson Score, red blood cell distribution width, and blood urea nitrogen level.\u003c/p\u003e\n\u003cp\u003ePositive SHAP values indicate that the feature will have a positive impact on the output, i.e., a higher risk of death. Red indicates a higher feature value, and blue indicates a lower feature value. SHAP summary dot plots (Figs. 4A, D, and G) visually display the direction and intensity of each feature\u0026apos;s impact on the model prediction: advanced age, high blood urea nitrogen, and high blood glucose levels significantly increase the risk of death in critically ill IS patients. Notably, a high CAR increases the risk of 28-day, 60-day, and 90-day all-cause mortality in the prediction model for critically ill IS patients.\u003c/p\u003e\n\u003cp\u003eTo further explore the contribution of these features to specific individual patients and clinical applications, we conducted local model explanations (Figs. 4C, F, and I). The red area indicates that the feature increases the risk of death, while the blue area indicates that the feature reduces the risk of death. The wider the color area is, the greater the impact of the feature on death. The results show that the length of hospital stay, SAPS-II score, red blood cells, and platelet count are positive contributors to 28-day and 60-day mortality prediction. SAPS-II score, hematocrit, platelet count, and CRP are positive contributors to 90-day mortality.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study analyzed 2,664 critically ill ischemic stroke (IS) patients and 43 potential variables to evaluate the association between the C-reactive protein to albumin ratio (CAR) and clinical outcomes. Our findings demonstrate that elevated CAR levels are significantly associated with increased 28-day, 60-day, and 90-day all-cause mortality in this population.Notably, significant differences in clinical history variables were observed between survivors and non-survivors. Non-survivors exhibited higher frequencies of comorbidities including myocardial infarction, congestive heart failure, peripheral vascular disease, chronic lung disease, diabetes, paraplegia, sepsis, and kidney disease. They also presented with higher SAPS II, SIRS, and Charlson scores, alongside lower GCS scores\u0026mdash;indicative of greater clinical severity. Importantly, CAR was significantly elevated in non-survivors with these conditions.\u003c/p\u003e\u003cp\u003eNotably, the CAR of non-survivors with the aforementioned conditions was significantly higher than that of survivors. In a prospective study [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e], it was indicated that elevated CRP and decreased albumin levels were associated with poor prognosis in IS patients 3 months later, which is similar to our research results. In our study, there were significant differences in CRP (34.0 vs 49.1, 32.6 vs 50.1, 31.8 vs 50.2, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and albumin (3.6 vs 3.4, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) between survivors and non-survivors at 28 days, 60 days, and 90 days. This is consistent with previous studies showing that high inflammatory levels and low albumin levels are associated with a worse prognosis in IS patients [\u003cspan additionalcitationids=\"CR29 CR30\" citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Previous studies have found that elevated CRP is significantly associated with the severity and mortality of IS patients [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Bucci et al. [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e] conducted a retrospective observational study within TriNetX and found that IS patients with CRP levels\u0026thinsp;\u0026gt;\u0026thinsp;3 mg/L had a significantly increased risk of cardiac complications such as death, heart failure, ischemic heart disease, atrial fibrillation, and ventricular arrhythmia. Low albumin is a marker of malnutrition, and critically ill IS patients are in a state of high catabolism. Without appropriate nutritional intervention, patients are prone to malnutrition, leading to poor clinical outcomes [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eAdditionally, albumin has neuroprotective effects due to its antioxidant, anti-apoptotic, and anti-inflammatory properties, which can reduce the recurrence rate and mortality of acute ischemic stroke patients and the occurrence of complications such as hemorrhagic transformation [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. Thuemmler et al. [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e] found that acute ischemic stroke patients with albumin\u0026thinsp;\u0026lt;\u0026thinsp;37 g/L had a 48% increased risk of in-hospital mortality and a twofold increased risk of long-term mortality. Zhou et al. [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e], through data analysis and meta-analysis from the Third China National Stroke Registry, found that low albumin levels increased the risk of functional disability and death in acute ischemic stroke patients at 3 months and 1 year.\u003c/p\u003e\u003cp\u003eCAR not only reflects the inflammatory level of patients but also their nutritional status. An increase in CAR indicates greater inflammation, lower nutrition, and weaker neuroprotective effects, which may synergistically increase the risk of death in IS patients. Currently, studies have shown that CAR is associated with adverse outcomes such as hemorrhagic transformation after thrombolysis, post-stroke complications such as stroke-associated pneumonia, and dysphagia in the elderly [\u003cspan additionalcitationids=\"CR41\" citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. Previous studies have shown [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e] that the composite index CAR has a higher predictive value for weaning from the ventilator in patients with traumatic brain injury than CRP or albumin alone. Therefore, we included 2,664 critically ill IS patients to explore the predictive value of CAR in predicting mortality in critically ill IS patients.\u003c/p\u003e\u003cp\u003eThe study results showed that an elevated CAR was associated with 28-day, 60-day, and 90-day all-cause mortality in critically ill IS patients. These findings remained consistent in subgroup analyses, enhancing the robustness of the results. Logistic regression analysis revealed that CAR was independently associated with an increased risk of death, with odds ratios (ORs) and 95% confidence intervals (CIs) for 28-day mortality (OR\u0026thinsp;=\u0026thinsp;1.006, 95% CI\u0026thinsp;=\u0026thinsp;1.003\u0026ndash;1.010, P\u0026thinsp;=\u0026thinsp;0.033), 60-day mortality (OR\u0026thinsp;=\u0026thinsp;1.005, 95% CI\u0026thinsp;=\u0026thinsp;1.001\u0026ndash;1.008, P\u0026thinsp;=\u0026thinsp;0.005), and 90-day mortality (OR\u0026thinsp;=\u0026thinsp;1.004, 95% CI\u0026thinsp;=\u0026thinsp;1.001\u0026ndash;1.007, P\u0026thinsp;=\u0026thinsp;0.016). Even after adjusting for confounding factors, the dose-response relationship between CAR levels and mortality risk persisted. We also observed significant age differences. In stratified analysis, CAR was significantly associated with increased mortality in IS critical patients aged 65 years and younger. This is consistent with the findings of DU et al. [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e], who included 69 stroke patients aged 18\u0026ndash;50 years and found that increased CAR was independently associated with an increased risk of adverse outcomes at 30 and 60 days (mRS score 2\u0026ndash;6) in young stroke patients.\u003c/p\u003e\u003cp\u003eIn this study, we used six machine learning models and found that models containing CAR had better discrimination values for 28-day, 60-day, and 90-day all-cause mortality in IS patients compared to models without CAR. The AUC of machine learning models containing CAR was consistently higher than those without CAR. After analyzing AUC, DCA, and calibration curves, the Gradient Boosting model demonstrated superior overall performance compared to other models. We applied the SHAP method to the Gradient Boosting regression model to achieve the best predictive performance and interpretability. We identified several important variables related to mortality in IS patients, further validating the importance of CAR in predicting 28-day, 60-day, and 90-day all-cause mortality in IS patients. In summary, our study results and previous studies support CAR as a useful predictive biomarker for mortality in IS patients [\u003cspan additionalcitationids=\"CR46 CR47\" citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e]. By monitoring CAR levels, doctors can identify individuals at higher risk of death earlier and make timely treatment decisions for intervention.\u003c/p\u003e\u003cp\u003eThis study has some limitations. First, it is based on retrospective data, which may introduce information bias. Second, although we adjusted for multiple potential confounding factors in the analysis, other unmeasured or uncontrolled confounding factors may still exist. For example, the use of anti-inflammatory drugs may affect CAR levels, but the MIMIC-IV and eICU-CRD databases do not contain information on pre-hospital medications. Additionally, the data in this study come from databases with racial restrictions, which means that the study results may have certain limitations when applied to multiple countries and ethnic groups. Although the predictive ability of CAR for mortality has been validated in both internal and external cohorts, further validation through other cohorts and prospective studies is needed to enhance the level of evidence.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eApproval for MIMIC-IV and eICU-CRD databases were granted by Physionet (https://physionet.org/). Given the retrospective design of this study, which utilized only existing data and did not influence clinical decision-making, the requirement for informed consent was waived. Therefore, no additional ethics approval or consent was needed for the analysis. All MIMIC-IV data are deidentified to comply with the Health Information Portability and Accountability Act (HIPAA) Safe Harbor provision. eICU-CRD data are deidentified to meet the safe harbor provision of the US Health Insurance Portability and HIPAA. The whole data collection and processing method obey in accordance with the Declaration of Helsinki\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of Data and Material\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePublicly available datasets used in this study can be downloaded from the mimic-IV database and eICU-CRD database. Further inquiries can be directed to the corresponding author.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFunded by the University-College Joint Foundation of Yunnan University of Chinese Medicine：NO.XYLH2024021.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBY W and L P conceived the study idea. L S, SC F and J Z performed the data analysis. G L, SP W, XL C and JF L participated in the preparation of figures and tables. BY W and J Z wrote the manuscript. L P, CX L and XL C revised the manuscript. All authors approved the submitted version.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eHilkens NA, Casolla B, Leung TW, de Leeuw FE. Stroke[J] Lancet. 2024;403(10446):2820\u0026ndash;36.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGerstl JVE, Blitz SE, Qu QR, Yearley AG, Lassar\u0026eacute;n P, Lindberg R, et al. Global, Regional, and National Economic Consequences of Stroke[J]. Stroke. 2023;54(9):2380\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSaini V, Guada L, Yavagal DR. Global Epidemiology of Stroke and Access to Acute Ischemic Stroke Interventions[J]. Neurology. 2021;97(20 Suppl 2):S6\u0026ndash;16.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang J, Ling L, Xiang L, Li W, Bao P, Yue W. Role of the gut microbiota in complications after ischemic stroke[J]. Front Cell Infect Microbiol. 2024;14:1334581.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSluis WM, Hinsenveld WH, Goldhoorn RB, Potters LH, Bruggeman AA, van der Hoorn A, et al. Timing and causes of death after endovascular thrombectomy in patients with acute ischemic stroke[J]. Eur Stroke J. 2023;8(1):215\u0026ndash;23.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNguyen TN, Abdalkader M, Fischer U, Qiu Z, Nagel S, Chen HS, et al. Endovascular management of acute stroke[J]. Lancet. 2024;404(10459):1265\u0026ndash;78.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAlsbrook DL, Di Napoli M, Bhatia K, Biller J, Andalib S, Hinduja A, et al. Neuroinflammation in Acute Ischemic and Hemorrhagic Stroke[J]. Curr Neurol Neurosci Rep. 2023;23(8):407\u0026ndash;31.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEckart A, Struja T, Kutz A, Baumgartner A, Baumgartner T, Zurfluh S, et al. Relationship of Nutritional Status, Inflammation, and Serum Albumin Levels During Acute Illness: A Prospective Study[J]. Am J Med. 2020;133(6):713\u0026ndash;e722717.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBian J, Guo S, Huang T, Li X, Zhao S, Chu Z, et al. CRP as a potential predictor of outcome in acute ischemic stroke[J]. Biomed Rep. 2023;18(2):17.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJiang Y, Yang Z, Wu Q, Cao J, Qiu T. The association between albumin and C-reactive protein in older adults[J]. Med (Baltim). 2023;102(34):e34726.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBi R, Shi Y, Li M, Liu X, Ma Z, Huang Y, et al. Association between serum albumin and severe impairment of activities of daily living in patients with stroke: a cross-sectional study[J]. Front Neurol. 2024;15:1501294.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBucci T, Pastori D, Pignatelli P, Ntaios G, Abdul-Rahim AH, Violi F, et al. Albumin Levels and Risk of Early Cardiovascular Complications After Ischemic Stroke: A Propensity-Matched Analysis of a Global Federated Health Network[J]. Stroke. 2024;55(3):604\u0026ndash;12.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBafei SEC, Yang S, Chen C, Gu X, Mu J, Liu F, et al. Sex and age differences in the association between high sensitivity C-reactive protein and all-cause mortality: A 12-year prospective cohort study[J]. Mech Ageing Dev. 2023;211:111804.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBao Y, Yang J, Duan Y, Chen Y, Chen W, Sun D. The C-reactive protein to albumin ratio is an excellent prognostic predictor for gallbladder cancer[J]. Biosci Trends. 2021;14(6):428\u0026ndash;35.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSabanoglu C, Inanc IH. C-reactive protein to albumin ratio predicts for severity of coronary artery disease and ischemia[J]. Eur Rev Med Pharmacol Sci. 2022;26(20):7623\u0026ndash;31.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSons\u0026ouml;z MR, Karadamar N, Yılmaz H, Eroğlu Z, Şahin KK, \u0026Ouml;zateş Y, et al. C-Reactive Protein to Albumin Ratio Predicts In-hospital Mortality in Patients with Acute Heart Failure[J]. Turk Kardiyol Dern Ars. 2023;51(3):174\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eW\u0026auml;ndell P, Carlsson AC, Larsson AO, \u0026Auml;rnl\u0026ouml;v J, Ruge T, Rydell A. The predictive value of cardiovascular outcomes and mortality assessed by the C-reactive protein to albumin ratio in the UK Biobank[J]. BMC Cardiovasc Disord. 2024;24(1):326.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYuksel ME, Ozkan N, Avci E. C-reactive protein/albumin ratio greater than 7.1 is a good candidate to be used as an inflammation biomarker to predict perforation in appendicitis[J]. Eur Rev Med Pharmacol Sci. 2022;26(22):8333\u0026ndash;41.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKaplan H, Cengiz G, Şaş S, Eldemir Y. Is the C-reactive protein-to-albumin ratio the most remarkable simple inflammatory marker showing active disease in patients with axial spondyloarthritis, psoriatic arthritis, and rheumatoid arthritis?[J]. Clin Rheumatol. 2023;42(11):2959\u0026ndash;69.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKunutsor SK, Laukkanen JA. Serum C-reactive protein-to-albumin ratio is a potential risk indicator for pneumonia: Findings from a prospective cohort study[J]. Respir Med. 2022;199:106894.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhou X, Fu S, Wu Y, Guo Z, Dian W, Sun H, et al. C-reactive protein-to-albumin ratio as a biomarker in patients with sepsis: a novel LASSO-COX based prognostic nomogram[J]. Sci Rep. 2023;13(1):15309.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRathore SS, Oberoi S, Iqbal K, Bhattar K, Ben\u0026iacute;tez-L\u0026oacute;pez GA, Nieto-Salazar MA, et al. Prognostic value of novel serum biomarkers, including C-reactive protein to albumin ratio and fibrinogen to albumin ratio, in COVID-19 disease: A meta-analysis[J]. Rev Med Virol. 2022;32(6):e2390.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJohnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, et al. MIMIC-IV, a freely accessible electronic health record dataset[J]. Sci Data. 2023;10(1):1.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePollard TJ, Johnson AEW, Raffa JD, Celi LA, Mark RG, Badawi O. The eICU Collaborative Research Database, a freely available multi-center database for critical care research[J]. Sci Data. 2018;5:180178.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSun T, Liu J, Yuan H, Li X, Yan H. Construction of a risk prediction model for lung infection after chemotherapy in lung cancer patients based on the machine learning algorithm[J]. Front Oncol. 2024;14:1403392.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLundberg SM, Erion G, Chen H, DeGrave A, Prutkin JM, Nair B, et al. From Local Explanations to Global Understanding with Explainable AI for Trees[J]. Nat Mach Intell. 2020;2(1):56\u0026ndash;67.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBielewicz JE, Kurzepa J, Kamieniak P, Daniluk B, Szczepańska-Szerej A, Rejdak K. Clinical and biochemical predictors of late-outcome in patients after ischemic stroke[J]. Ann Agric Environ Med. 2020;27(2):290\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEto F, Nezu T, Aoki S, Kamimura T, Naito H, Shiga Y, et al. The association between nutritional intake one week after admission and outcome in acute ischemic stroke patients[J]. J Stroke Cerebrovasc Dis. 2022;31(11):106812.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShi K, Tian DC, Li ZG, Ducruet AF, Lawton MT, Shi FD. Global brain inflammation in stroke[J]. Lancet Neurol. 2019;18(11):1058\u0026ndash;66.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang R, Cao L, He Y, Zhang P, Feng L. Nutrition-associated markers and outcomes among patients receiving enteral nutrition after ischemic stroke: a retrospective cohort study[J]. BMC Neurol. 2024;24(1):303.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang G, Pan Y, Zhang R, Wang M, Meng X, Li Z, et al. Prevalence and Prognostic Significance of Malnutrition Risk in Patients With Acute Ischemic Stroke: Results From the Third China National Stroke Registry[J]. Stroke. 2022;53(1):111\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang L, Yang L, Liu H, Pu J, Li Y, Tang L et al. C-Reactive Protein Levels and Cognitive Decline following Acute Ischemic Stroke: A Systematic Review and Meta-Analysis[J]. Brain Sci, 2023, 13(7).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJiang J, Tan C, Zhou W, Peng W, Zhou X, Du J, et al. Plasma C-Reactive Protein Level and Outcome of Acute Ischemic Stroke Patients Treated by Intravenous Thrombolysis: A Systematic Review and Meta-Analysis[J]. Eur Neurol. 2021;84(3):145\u0026ndash;50.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBucci T, Sagris D, Harrison SL, Underhill P, Pastori D, Ntaios G, et al. C-reactive protein levels are associated with early cardiac complications or death in patients with acute ischemic stroke: a propensity-matched analysis of a global federated health from the TriNetX network[J]. Intern Emerg Med. 2023;18(5):1329\u0026ndash;36.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKiboshi R, Satoh S, Mikami K, Kitajima M, Urushizaka M, Metoki N, et al. Serum Albumin, Body Mass Index, and Preceding Xa and P2Y12 Inhibitors Predict Prognosis of Recurrent Ischemic Stroke[J]. J Stroke Cerebrovasc Dis. 2021;30(5):105681.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang C, Deng L, Qiu S, Bian H, Wang L, Li Y, et al. Serum Albumin Is Negatively Associated with Hemorrhagic Transformation in Acute Ischemic Stroke Patients[J]. Cerebrovasc Dis. 2019;47(1\u0026ndash;2):88\u0026ndash;94.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang Q, Lei YX, Wang Q, Jin YP, Fu RL, Geng HH, et al. Serum albumin level is associated with the recurrence of acute ischemic stroke[J]. Am J Emerg Med. 2016;34(9):1812\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eThuemmler RJ, Pana TA, Carter B, Mahmood R, Bettencourt-Silva JH, Metcalf AK et al. Serum Albumin and Post-Stroke Outcomes: Analysis of UK Regional Registry Data, Systematic Review, and Meta-Analysis[J]. Nutrients, 2024, 16(10).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhou H, Wang A, Meng X, Lin J, Jiang Y, Jing J, et al. Low serum albumin levels predict poor outcome in patients with acute ischaemic stroke or transient ischaemic attack[J]. Stroke Vasc Neurol. 2021;6(3):458\u0026ndash;66.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXu T, Xia L, Wu Y, Xu Y, Xu X, Zhang W, et al. High ratio of C-reactive protein to albumin is associated with hemorrhagic transformation and poor functional outcomes in acute ischemic stroke patients after thrombolysis[J]. Front Aging Neurosci. 2023;15:1109144.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang L, Zhang R, Ji J, Long F, Wang Y, Lu J, et al. Hypersensitive C-reactive protein-albumin ratio is associated with stroke-associated pneumonia and early clinical outcomes in patients with acute ischemic stroke[J]. Brain Behav. 2022;12(7):e2675.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGuo C, Zheng P, Chen S, Wei L, Fu X, Fu Y, et al. Association between the C-reactive protein/albumin ratio and mortality in older Japanese patients with dysphagia[J]. Front Nutr. 2024;11:1370763.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZeng J, Liu J, Lu Y, Fu J, Han D, Chen J, et al. C-reactive protein to albumin ratio and Glasgow Coma Scale score as the predictors for weaning outcomes in traumatic brain injury[J]. Ann Med. 2025;57(1):2472866.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDu Y, Zhang J, Li N, Guo J, Liu X, Bian L, et al. Association between the C-reactive protein to albumin ratio and adverse clinical prognosis in patients with young stroke[J]. Front Neurol. 2022;13:989769.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJang JH, Hong S, Ryu JA. Prognostic Value of C-Reactive Protein and Albumin in Neurocritically Ill Patients with Acute Stroke[J]. J Clin Med, 2022, 11(17).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKocat\u0026uuml;rk M, Kocat\u0026uuml;rk \u0026Ouml;. Assessment of relationship between C-reactive protein to albumin ratio and 90-day mortality in patients with acute ischaemic stroke[J]. Neurol Neurochir Pol. 2019;53(3):205\u0026ndash;11.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYu D, Guo G, Wan F, Hu B. The association between C-reactive protein to albumin ratio and adverse outcomes in acute ischemic stroke patients: A study in the Korean population[J]. Heliyon. 2024;10(20):e39212.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYuan J, Cheng Y, Han X, Zhu N, Ma W, Li J, et al. Association between C-reactive protein/albumin ratio and all-cause mortality in patients with stroke: Evidence from NHANES cohort study[J]. Nutr Metab Cardiovasc Dis. 2024;34(10):2305\u0026ndash;14.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTables 1 to 5 are available in the Supplementary Files section.\u003c/p\u003e\n"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-neurology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"nurl","sideBox":"Learn more about [BMC Neurology](http://bmcneurol.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/nurl","title":"BMC Neurology","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"ischemic stroke, C-reactive protein, albumin, Prognosis, Machine learning","lastPublishedDoi":"10.21203/rs.3.rs-7185510/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7185510/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe C-Reactive Protein-to-Albumin Ratio (CAR) demonstrates associations with cerebrovascular disease outcomes. However, its prognostic value in critically ill ischemic stroke (IS) patients intensive care unit (ICU) admission remains unclear. This study aimed to investigate the association between CAR and clinical prognosis in critically ill IS patients.\u003c/p\u003e\u003cp\u003eIn this retrospective cohort study, clinical data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database (serving as the training set) and externally validated using the eICU Collaborative Research Database (eICU-CRD). The primary outcomes were 28-day, 60-day, and 90-day all-cause mortality. The association between CAR and mortality was evaluated using multivariable logistic regression and restricted cubic splines (RCS). Machine learning algorithms were employed to develop prediction models incorporating CAR. Model performance was assessed using the Boruta algorithm for feature importance and the Integrated Discrimination Improvement (IDI).\u003c/p\u003e\u003cp\u003eA total of 2,664 critically ill IS patients were analyzed (mean CAR: 22.173\u0026thinsp;\u0026plusmn;\u0026thinsp;26.011). After adjusting for confounders, multivariable logistic regression confirmed CAR as an independent predictor of mortality: the adjusted odds ratios (95% confidence intervals) were 1.006 (1.003\u0026ndash;1.010, P\u0026thinsp;=\u0026thinsp;0.033) for 28-day, 1.005 (1.001\u0026ndash;1.008, P\u0026thinsp;=\u0026thinsp;0.005) for 60-day, and 1.004 (1.001\u0026ndash;1.007, P\u0026thinsp;=\u0026thinsp;0.016) for 90-day mortality. RCS analysis indicated a monotonically increasing association between CAR and mortality risk. Machine learning models incorporating CAR demonstrated superior fit and higher area under the curve (AUC) values compared to models without it. In conclusions,In critically ill patients with ischemic stroke, a higher CAR is significantly associated with increased short- and medium-term all-cause mortality risk.\u003c/p\u003e","manuscriptTitle":"The C-Reactive Protein-to-Albumin Ratio (CAR) and All-Cause Mortality in Critically Ill Ischemic Stroke Patients: A Retrospective Analysis of the MIMIC-IV and eICU-CRD Databases","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-12 14:34:25","doi":"10.21203/rs.3.rs-7185510/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2025-09-05T04:52:59+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-08-10T20:05:21+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-07-28T08:24:43+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-07-28T08:22:25+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Neurology","date":"2025-07-22T09:41:44+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-neurology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"nurl","sideBox":"Learn more about [BMC Neurology](http://bmcneurol.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/nurl","title":"BMC Neurology","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"0b99e2a0-2b55-4015-ad2c-298fe4e20797","owner":[],"postedDate":"September 12th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-09-12T14:34:25+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-12 14:34:25","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7185510","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7185510","identity":"rs-7185510","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}