Association between stress hyperglycemia ratio and all-cause mortality in critically ill patients with mitral valve disease Authors: Nanru Peng, Peizhe Li, Yimeng Liu Published: 2025 ABSTRACT AbstractThe study of stress hyperglycemia ratio (SHR) aims to further investigate the relationship between chronic glucose factors and adverse clinical events, particularly cardiovascular outcomes, in critically ill patients. However, prior research has not explored the correlation between SHR and the prognosis of patients with mitral valve disease (MVD). This study evaluates the relationship between SHR and 28-day all-cause mortality in critically ill MVD patients admitted to the intensive care unit. Patient data were extracted from the MIMIC-IV (medical information mart for intensive care IV) database. Participants were categorized into four groups based on SHR quartiles. The primary outcome was 28-day all-cause mortality, and Kaplan-Meier curves were used to compare outcomes across different glucose metabolism states. Cox proportional hazards regression and restricted cubic splines (RCS) were employed to assess the relationship between the SHR index and the outcomes. Five machine learning algorithms were utilized for mortality risk prediction, with SHapley additive explanations (SHAP) applied to identify key predictors. Among 3506 participants, 1986 (56.65%) were male and 1520 (43.35%) were female. Kaplan-Meier curves showed that higher SHR levels were associated with increased 28-day all-cause mortality (log-rank P = 0.01). This correlation persisted across glucose metabolism groups. Cox regression analysis revealed that the highest SHR quartile was linked to higher mortality risk, particularly in non-diabetic patients. RCS analysis confirmed a U-shaped relationship between SHR and mortality. Subgroup analysis showed that SHR was associated with mortality risk across various subgroups, with stronger associations in females and prediabetic patients. Among eight machine learning models developed based on eight RFE-selected variables, the XGBoost model achieved the best predictive performance for 28-day mortality (AUC = 0.819). SHR can serve as a key prognostic tool for mortality in mitral valve disease (MVD) patients, especially among women and patients with prediabetes. This integrated approach may inform personalized glucose management strategies, potentially improving clinical outcomes.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-27975-x.Keywords: Mitral valve disease, Critical illness, Machine learning, Stress hyperglycemia ratioSubject terms: Cardiology, Diseases, Endocrinology, Medical research, Risk factors Introduction Mitral valve disease (MVD) encompasses heart conditions caused by dysfunction of the mitral valve, including mitral stenosis and mitral regurgitation, which disrupt normal heart function and hemodynamics1,2. The causes of MVD vary regionally: in developed countries, degenerative changes such as mitral valve prolapse and annular calcification are common, while in developing countries, rheumatic heart disease (RHD) remains the primary cause of mitral stenosis3,4. Despite advancements in treatment, including medication, valve replacement surgery, and catheter-based interventions, the rapid progression of MVD, particularly in elderly patients, continues to result in high mortality rates5,6.Therefore, early identification and treatment of critically ill patients with mitral valve disease are crucial. Stress-induced hyperglycemia (SHR) refers to elevated blood glucose levels triggered by acute stress7,8, which can lead to microvascular damage, myocardial fibrosis, myocardial hypertrophy, and mitochondrial dysfunction9–11. The American Heart Association (AHA) has proposed that SHR could serve as an important marker not only to quantify glucose fluctuations under stress but also to predict adverse clinical outcomes12. Elevated SHR has been linked to increased mortality risk following myocardial infarction, ischemic stroke, and other cardiovascular diseases, such as heart failure13–15, atrial fibrillation16,17, coronary artery disease18, and post-cardiac surgery19,20. However, the prognostic value of SHR in critically ill patients with heart valve disease, particularly MVD, remains unexplored. Patients with severe MVD often experience hemodynamic instability, multi-organ dysfunction, and metabolic stress, which complicate early treatment and disease assessment. Thus, this study aims to fill the gap by evaluating whether SHR can serve as a prognostic marker for mortality in critically ill patients with mitral valve disease. This evaluation could help identify high-risk patients for closer monitoring and timely intervention. Methods Data source This study is a retrospective analysis using data from the publicly available MIMIC-IV (version 3.1) database21, which contains clinical information for over 190,000 patients and 450,000 hospital admissions recorded at Beth Israel Deaconess Medical Center (BIDMC) in Boston, Massachusetts, between 2008 and 2022. The database provides a wide range of patient-related information22, including demographic details, laboratory results, vital signs, medication records, surgical interventions, disease diagnoses, drug management, and follow-up survival status, all while ensuring patient privacy. Since the database does not provide protected health information (e.g. patient names or admission dates), a waiver of informed consent was granted. All the data can be accessed publicly at https://mimic.mit.edu/docs/iv/. Study design and population Our analysis included patients aged 18 years or older with mitral valve disease (Table 1). Table 1Baseline characteristics of patients grouped according to outcome.variablesSurvival(N = 3285)Dead (N = 221) p Population informationGender, n(%)Female1416 (43.1%)104 (47.1%) 0.281 Male1869 (56.9%)117 (52.9%)Age, (years)69.52 ± 12.7976.44 ± 13.12 < 0.001 ComorbiditiesMI, n(%)NO2443 (74.4%)140 (63.3%) < 0.001 YES842 (25.6%)81 (36.7%)CHF, n(%)NO1475 (44.9%)59 (26.7%) < 0.001 YES1810 (55.1%)162 (73.3%)CVD, n(%)NO2819 (85.8%)155 (70.1%) < 0.001 YES466 (14.2%)66 (29.9%)COPD, n(%)NO2340 (71.2%)147 (66.5%) 0.156 YES945 (28.8%)74 (33.5%)LD, n(%)NO3130 (95.3%)194 (87.8%) < 0.001 YES155 (4.7%)27 (12.2%)DM, n(%)NO2279 (69.4%)123 (55.7%) < 0.001 YES1006 (30.6%)98 (44.3%)AF, n(%)NO1461 (44.5%)88 (39.8%) 0.201 YES1824 (55.5%)133 (60.2%)HBP, n(%)NO863 (26.3%)43 (19.5%) 0.031 YES2422 (73.7%)178 (80.5%)HLP, n(%)NO1455 (44.3%)102 (46.2%) 0.639 YES1830 (55.7%)119 (53.8%)AKI, n(%)NO883 (26.9%)23 (10.4%) < 0.001 YES2402 (73.1%)198 (89.6%)CAD, n(%)NO1446 (44%)92 (41.6%) 0.533 YES1839 (56%)129 (58.4%) Score APSIII40.37 ± 17.9257.86 ± 23.21 < 0.001 GCS14.12 ± 2.9114.14 ± 2.34 0.889 CCI5.03 ± 2.647.78 ± 2.74 < 0.001 SOFA2.65 ± 2.422.80 ± 2.89 0.453 Laboratory InspectionHR, (bpm)82.09 ± 13.1987.27 ± 16.64 < 0.001 MBP, (mmHg)86.32 ± 8.9986.10 ± 11.82 0.792 RR, (bpm)18.68 ± 3.2820.89 ± 4.08 < 0.001 Tm, (℃)36.72 ± 0.4536.70 ± 0.59 0.568 Spo2, (%)97.37 ± 1.7296.27 ± 3.26 < 0.001 WBC, (K/uL)12.88 ± 5.7012.58 ± 6.78 0.526 Plt, (109/L)171.48 ± 73.40202.31 ± 113.96 < 0.001 Hb, (g/dL)10.40 ± 1.6910.30 ± 2.10 0.483 AG, (mEq/L)12.72 ± 3.4216.14 ± 5.02 < 0.001 Na, (mEq/L)137.96 ± 3.47138.54 ± 5.75 0.138 Ca, (mg/dL)8.41 ± 0.658.64 ± 0.89 < 0.001 BUN, (mEq/L)22.89 ± 16.4940.53 ± 26.72 < 0.001 Crn, (mg/dL)1.25 ± 1.172.01 ± 1.57 < 0.001 PT, (s)15.56 ± 5.4318.91 ± 9.35 < 0.001 PTT, (s)38.11 ± 17.5047.06 ± 27.68 < 0.001 HbA1c, (%)6.11 ± 1.266.24 ± 1.30 0.142 SHR1.03 ± 0.261.20 ± 0.35 < 0.001 TreatmentACEI/ARB, n(%)NO2781 (84.7%)194 (87.8%) 0.247 YES504 (15.3%)27 (12.2%)Insulin, n(%)NO1795 (54.6%)143 (64.7%) 0.004 YES1490 (45.4%)78 (35.3%)Statin, n(%)NO2534 (77.1%)161 (72.9%) 0.167 YES751 (22.9%)60 (27.1%)Warfarin, n(%)NO2520 (76.7%)196 (88.7%) < 0.001 YES765 (23.3%)25 (11.3%)Mechanical Ventilation, n(%)NO313 (9.5%)29 (13.1%) 0.104 YES2972 (90.5%)192 (86.9%)CRRT, n(%)NO3184 (96.9%)187 (84.6%) < 0.001 YES101 (3.1%)34 (15.4%)ICU time, (d)28.00 ± 0.0010.41 ± 7.80 < 0.001 SHR group, n(%)Q1845 (25.7%)33 (14.9%) < 0.001 Q2839 (25.5%)30 (13.6%)Q3834 (25.4%)49 (22.2%)Q4767 (23.3%)109 (49.3%)Glucose Metabolism State, n(%)DM728 (22.2%)66 (29.9%) < 0.001 NGR1379 (42%)62 (28.1%)Pre-DM1178 (35.9%)93 (42.1%)MI: myocardial infarction, CHF: chronic heart failure, CVD: cerebrovascular disease, COPD: chronic obstructive pulmonary disease, LD: liver disease, DM: diabetes mellitus, AF: atrial fibrillation, HBP: hypertension, HLP: hyperlipidemia, AKI: acute kidney injury, CAD: coronary artery disease. HR: heart rate, RR: respiratory rate, MBP: mean blood pressure, Tm: temperature, SpO2: pulse blood oxygen saturation, WBC: white blood cell count, PLT: platelet, Na: sodium, Ca: calcium, AG: anion gap, PT: prothrombin time, PTT: partial thromboplastin time, BUN: blood urea nitrogen, Crn: creatinine, HbA1c: glycated hemoglobin, APSIII: acute physiology score, SOFA: sequential organ failure assessment, ACEI/ARB: angiotensin-converting enzyme Inhibitors /angiotensin receptor blockers, CRRT: continuous renal replacement therapy, SHR: stress-induced hyperglycemia, NGR: normal Glucose Regulation, Pre-DM: prediabetes. Baseline characteristics of patients grouped according to outcome. MI: myocardial infarction, CHF: chronic heart failure, CVD: cerebrovascular disease, COPD: chronic obstructive pulmonary disease, LD: liver disease, DM: diabetes mellitus, AF: atrial fibrillation, HBP: hypertension, HLP: hyperlipidemia, AKI: acute kidney injury, CAD: coronary artery disease. HR: heart rate, RR: respiratory rate, MBP: mean blood pressure, Tm: temperature, SpO2: pulse blood oxygen saturation, WBC: white blood cell count, PLT: platelet, Na: sodium, Ca: calcium, AG: anion gap, PT: prothrombin time, PTT: partial thromboplastin time, BUN: blood urea nitrogen, Crn: creatinine, HbA1c: glycated hemoglobin, APSIII: acute physiology score, SOFA: sequential organ failure assessment, ACEI/ARB: angiotensin-converting enzyme Inhibitors /angiotensin receptor blockers, CRRT: continuous renal replacement therapy, SHR: stress-induced hyperglycemia, NGR: normal Glucose Regulation, Pre-DM: prediabetes. Patients were excluded based on the following criteria: Lack of HbA1c data; Lack of HbA1c data; Absence of admission fasting blood glucose data; Absence of admission fasting blood glucose data; Non-first hospitalizations; Non-first hospitalizations; Hospital stay of less than 1 day. Hospital stay of less than 1 day. Ultimately, a total of 3,506 patients met the inclusion criteria and were divided into four groups based on the quartiles of the SHR index (Fig. 1). Flow of included patients through the trial. Data extraction Demographic information Age, gender. Comorbidities (Extracted based on the International Classification of Diseases coding system) including myocardial infarction (MI), chronic heart failure (CHF), cerebrovascular disease (CVD), chronic obstructive pulmonary disease (COPD), liver disease (LD), diabetes mellitus (DM), atrial fibrillation (AF), hypertension (HBP), hyperlipidemia (HLP), acute kidney injury (AKI), and coronary artery disease (CAD). Vital signs Heart rate(HR), mean blood pressure (MBP), respiratory rate(RR), arterial oxygen saturation (SpO₂). Laboratory test results White blood cell count (WBC), platelet (PLT), hemoglobin(Hb), sodium (Na), potassium (K), calcium (Ca), anion gap(AG), prothrombin time (PT), partial thromboplastin time (PTT), blood urea nitrogen (BUN), creatinine (Crn), admission blood glucose (ABG) and glycated hemoglobin (HbA1c). Disease scoring systems Sequential Organ Failure Assessment (SOFA), Acute Physiology Score III (APSIII), Glasgow Coma Scale (GCS), Charlson Comorbidity Index (CCI). Treatment information Angiotensin-converting enzyme Inhibitors /angiotensin receptor blockers (ACEI/ARB), insulin (INS), statins, warfarin (WAR), mechanical ventilation (MV), continuous renal replacement therapy (CRRT). Stress-induced hyperglycemia ratio (SHR) Defined as the index calculated using the formula: SHR = (Admission blood glucose (mg/dl)) / (28.7 × HbA1c (%) − 46.7)23. Outcomes The primary outcomes of the present study were 28-day all-cause mortality. Statistical analysis Firstly, the missing data proportion for all variables was calculated. Variables with a missing data rate exceeding 20% were excluded from the analysis. For variables with a missing data rate below 20%, multiple imputation was used to address the missing data24. However, due to the possibility that the missing data was not randomly generated, potential biases cannot be ruled out. Therefore, this should be taken into account when interpreting the results. The variance inflation factor (VIF) was then assessed to evaluate multicollinearity, and variables with a VIF greater than 5 were excluded from the multivariable model25,26. To ensure the robustness of the final model, we also conducted further analysis of potential multicollinearity through correlation matrices and theoretical correlations. When the collinear variables represent similar constructs, only one of the variables is retained to maintain simplicity without reducing explanatory power. Based on the SHR quartile method, the patients included in the study were divided into four groups: Q1 (SHR < 0.90) with 877 patients, Q2 (0.90 ≤ SHR < 1.01) with 876 patients, Q3 (1.01 ≤ SHR < 1.14) with 876 patients, and Q4 (SHR ≥ 1.14) with 877 patients. Additionally, participants were classified into three distinct cohorts based on established glucose metabolism diagnostic criteria: Normal Glucose Regulation (NGR), Prediabetes (Pre-DM), and Diabetes Mellitus (DM)27. The NGR group consisted of individuals with HbA1c levels < 5.7% and no history of diabetes. The Pre-DM group included patients with HbA1c levels ranging from 5.7% (inclusive) to 6.5% and no prior history of diabetes. The DM group included patients with a history of diabetes or HbA1c levels ≥ 6.5%28,29. The Shapiro-Wilk test was used to assess the normality of continuous variables. Normally distributed variables were analyzed using Student’s t-test or one-way ANOVA (presented as mean ± SD). Non-normally distributed variables were compared using the Wilcoxon rank-sum test (presented as median [IQR]). Chi-square or Fisher’s exact test was applied to evaluate categorical variables (presented as counts and percentages). Kaplan-Meier (KM) curves were generated to estimate the overall risk of all-cause mortality and the risk of all-cause mortality stratified by different SHR index quartiles. Cox proportional hazards regression models were used to assess the hazard ratio (HR) and 95% confidence interval (95% CI) for event occurrence, in three consecutive steps: Model 1 (unadjusted), Model 2 (adjusted for age and gender), and Model 3 (adjusted for age, gender, APSIII, AG, Ca, SpO₂, AKI, CHF, INS, WAR, CRRT, BUN, Crn, PT, PLT). In all three models, the lowest SHR quartile was used as the reference group. The selection of covariates was based on clinically relevant factors and those showing statistical significance in univariate analysis, with multicollinearity controlled (VIF > 5). A restricted cubic spline (RCS) analysis was conducted to explore the dose-response relationship of SHR. Subgroup analyses, stratified by gender, age, comorbidities, and treatment methods were visualized using hazard ratio forest plots (95% CI). For machine learning-based prediction, Feature Recursive Elimination (RFE) was used to select the most important features for predicting 28-day mortality. The dataset was randomly divided into training (70%) and testing (30%) subsets. Eight models were developed: Support Vector Machine (SVM) was chosen for its effectiveness in classification tasks. Neural Network (NN) and Multilayer Perceptron (MLP) was selected for its ability to model non-linear relationships. Gaussian Process (GP) was used for its flexibility in capturing uncertainty. Gradient Boosting Machine (GBM) was preferred for its strong performance in complex data patterns. Logistic Regression (LR) was included for its simplicity and interpretability. Naive Bayes (NB) was selected for its speed and efficiency with categorical data. Extreme Gradient Boosting (XGB)was chosen for its superior predictive accuracy. The performance of the model was evaluated using metrics including the Area Under the Curve (AUC), accuracy, specificity, sensitivity, and the F1 score. The best-performing model was further explained using SHapley Additive Explanations (SHAP) to identify key predictive factors. All analyses were conducted in RStudio, with a two-tailed p-value of 0.05 considered statistically significant. Results Baseline characteristics Data from 3,506 patients diagnosed with mitral valve disease were selected from the MIMIC-IV database (Fig. 1). Table S1 presents the variance inflation factors, confirming the absence of multicollinearity, and also shows the missing data proportion for each variable. Table 1 summarizes the baseline characteristics of the participants, with 1,869 male patients (56.9%). Among the patients, 842 (25.6%) had myocardial infarction, 1,006 (30.6%) had diabetes mellitus, 1,810 (55.1%) had heart failure, 466 (14.2%) had cerebrovascular disease, 2,402 (73.1%) had acute kidney injury, 2,402 (73.1%) had chronic obstructive pulmonary disease, 2,422 (73.7%) had hypertension, 1,830 (55.7%) had hyperlipidemia, and 155 (4.7%) had liver disease. Additionally, 1,824 (55.5%) had atrial fibrillation, and 1,839 (56%) had coronary artery disease. Compared to survivors, non-survivors were older and had significantly higher APSIII, GCS, CCI, and SOFA scores. The use of mechanical ventilation and renal replacement therapy was also more frequent among non-survivors. As shown in Table S2, higher SHR groups were associated with increased heart rate, temperature, APS, AG, BUN, Crn, PT, and PTT, while SpO₂ levels were lower. Renal replacement therapy was used more frequently in these patients. Relationship between SHR and clinical outcomes of patients with MVD The Kaplan-Meier curve indicates that, regardless of the overall patient population or the three different glucose metabolic states, the 28-day survival rate gradually decreases as SHR quartiles increase (Fig. 2). Kaplan-Meier survival analysis curves for all-cause mortality of 28-day. Kaplan-Meier curves for all patients(A); and for NGR(B); and for Pre-DM (C); and for DM (D). In the adjusted Cox proportional hazards regression analysis for the entire study population (Table 2), patients in the highest SHR quartile (SHR ≥ 1.14) had a 2.90-fold increased risk of death within 28 days (2.90 (1.95–4.31)). Furthermore, in the glucose metabolism subgroups, those in the highest SHR quartile had a 2.61-fold increased risk of all-cause mortality within 28 days (2.61 (1.56–4.36]). Notably, in the patient without DM group, patients in the highest SHR quartile had a 2.91-fold increased risk of all-cause mortality within 28 days (2.91 [1.45–5.84]), which was slightly higher than in patients with DM. The proportional hazards assumption was tested and satisfied for all Cox regression models. Overall, the correlation between SHR and mortality remained consistent across both the diabetic subgroups and the entire population. Table 2Cox proportional hazard models for 28-day all-cause mortality.VariablesModel 1Model 2Model 3HR (95% CI)PHR (95% CI)PHR (95% CI)POverallQ 11.00(reference)1.00(reference)1.00(reference)Q 20.94(0.57–1.53)0.7290.99(0.61–1.62)0.9731.37(0.83–2.25)0.219Q 31.47(0.94–2.29)0.0881.57(1.02–2.44)0.0481.92(1.22-3.00)0.004Q43.47(2.35–5.12)< 0.0013.72(2.52–5.50)< 0.0012.90(1.95–4.31)< 0.001P for trend1.90(1.65–2.19)< 0.0012.02(1.75–2.33)< 0.0011.75(1.45–2.12)< 0.001Patients with DMQ 11.00(reference)1.00(reference)1.00(reference)Q 21.24(0.63–2.46)0.5291.21(0.61–2.38)0.5891.46(0.73–2.94)0.285Q 32.08(1.12–3.86)0.0201.90(1.03–3.53)0.0411.87(1.00-3.50)0.051Q44.14(2.53–6.77)< 0.0014.05(2.48–6.63)< 0.0012.61(1.56–4.36)0.005P for trend2.10(1.69–2.59)< 0.0012.05(1.66–2.54)< 0.0011.78(1.27–2.49)< 0.001Patients without DMQ 11.00(reference)1.00(reference)1.00(reference)Q 21.07(0.49–2.31)0.8731.20(0.55–2.60)0.6451.27(0.57–2.80)0.557Q 31.70(0.83–3.48)0.1451.95(0.95-4.00)0.0671.84(0.88–3.83)0.105Q43.93(2.01–7.69)< 0.0014.52(2.31–8.86)< 0.0012.91(1.45–5.84)0.003P for trend1.84(1.52–2.24)< 0.0012.03(1.67–2.46)< 0.0011.76(1.35–2.30)< 0.001HR: Hazard Ratio, CI: Confidence Interval.Model1: unadjusted.Model2: Adjust: Age and Gender.Model3: Adjust: Age, Gender, APSIII, AG, Ca, SpO₂, AKI, CHF, INS, WAR, CRRT, BUN, Crn, PT, PLT. Cox proportional hazard models for 28-day all-cause mortality. 1.00 (reference) 1.00 (reference) 1.00 (reference) 0.94 (0.57–1.53) 0.99 (0.61–1.62) 1.37 (0.83–2.25) 1.47 (0.94–2.29) 1.57 (1.02–2.44) 1.92 (1.22-3.00) 3.47 (2.35–5.12) 3.72 (2.52–5.50) 2.90 (1.95–4.31) 1.90 (1.65–2.19) 2.02 (1.75–2.33) 1.75 (1.45–2.12) 1.00 (reference) 1.00 (reference) 1.00 (reference) 1.24 (0.63–2.46) 1.21 (0.61–2.38) 1.46 (0.73–2.94) 2.08 (1.12–3.86) 1.90 (1.03–3.53) 1.87 (1.00-3.50) 4.14 (2.53–6.77) 4.05 (2.48–6.63) 2.61 (1.56–4.36) 2.10 (1.69–2.59) 2.05 (1.66–2.54) 1.78 (1.27–2.49) 1.00 (reference) 1.00 (reference) 1.00 (reference) 1.07 (0.49–2.31) 1.20 (0.55–2.60) 1.27 (0.57–2.80) 1.70 (0.83–3.48) 1.95 (0.95-4.00) 1.84 (0.88–3.83) 3.93 (2.01–7.69) 4.52 (2.31–8.86) 2.91 (1.45–5.84) 1.84 (1.52–2.24) 2.03 (1.67–2.46) 1.76 (1.35–2.30) HR: Hazard Ratio, CI: Confidence Interval. Model1: unadjusted. Model2: Adjust: Age and Gender. Model3: Adjust: Age, Gender, APSIII, AG, Ca, SpO₂, AKI, CHF, INS, WAR, CRRT, BUN, Crn, PT, PLT. The detection of nonlinear relationships The RCS analysis of 28-day all-cause mortality revealed a U-shaped relationship between SHR and the risk of death (Fig. 3). The lowest point of the RCS curve occurred at SHR = 0.86, representing the point of minimal mortality risk. According to the Cox proportional hazards regression model, the inflection point is located in the Q1 group of SHR, where the mortality risk is at its lowest. These findings are consistent between both analyses. RCS of SHR with 28-day all-cause mortality. Subgroup analysis As shown in the Fig. 4, the association between SHR and mortality remains highly significant (1.90 (1.65–2.19), P < 0.001). This relationship is consistent across multiple subgroups, including those categorized by age, myocardial infarction (MI), warfarin (WAR), and mechanical ventilation treatment (MV), with no significant interactions observed (Fig. 4). However, interestingly, compared to male patients, SHR is a more significant predictor of all-cause mortality in female patients (Female: 7.16 (4.36–11.74) vs. Male: 1.74 (1.44–2.11), interaction p < 0.001). Meanwhile, in different subgroups of sugar metabolism, the all-cause mortality rate of SHR for Pre-DM patients was significantly higher than that of NGR and DM (NGR: 2.14 (1.69–2.59) vs. Pre-DM: 14.93 (7.52–29.67) vs. DM: 1.61 (1.26–2.07), interaction p < 0.001). Forest plots of Subgroups analyses of SHR and 28-day all-cause mortality. MI: myocardial infarction, DM: diabetes mellitus, WAR: warfarin, MV: mechanical ventilation treatment. Machine learning The final variables in the machine learning model were determined using recursive feature elimination (RFE) analysis, which selected the top 8 features (Fig. 5): SHR, CRRT, CVD, WAR, LD, CAD, ACEI_ARB, AKI. The dataset was split into training and validation sets in a 7:3 ratio. Eight models were developed using the selected features, including Support Vector Machine (SVM), Neural Network (NN) with Multilayer Perceptron (MLP), Gaussian Process (GP), Gradient Boosting Machine (GBM), Logistic Regression (LR), Naive Bayes (NB), and Extreme Gradient Boosting (XGB). They were all applied to individually analyze the filtered variables to predict the 28-day mortality risk in patients with mitral valve disease. Hyperparameter tuning was performed during the model development process. The training and validation sets were used for model training and evaluation, respectively. Model performance was assessed using ROC curves and the corresponding area under the curve (AUC) (Fig. 6). Decision curve analysis (DCA) was employed to evaluate clinical utility (Figure S1), while calibration curves were used to assess the accuracy of the model in predicting absolute risks (Figure S2). Among the models, XGBoost demonstrated the best performance (AUC = 0.819), and the DCA curve indicated its clinical value across a wide range of threshold probabilities. Feature selection based on the REF algorithm. ROC curves and DCA Curve of ML-based 28-day mortality prediction models. Support vector machine (SVM), neural network (NN) with multilayer perceptron) (MLP, Gaussian process (GP), Gradient boosting machine (GBM), logistic regression (LR), Naive Bayes (NB), and extreme gradient boosting (XGB). Meanwhile, the SHAP method is employed to evaluate the contribution of each feature to the overall prediction result (Fig. 7). To further validate the interpretability of the model, a single-sample prediction was performed using the best model. The results indicated that SHR had a negative impact on the prediction for survivors (Figure S3), while it had a positive impact on the prediction for non-survivors (Figure S4). Figure S5 shows the predictive ability of different scoring criteria for the 28-day mortality risk of critical mitral valve disease. Among them, the CCS score has the best discriminatory performance (AUC = 0.712). Feature importance evaluation graph. Discussion This retrospective study primarily investigates the correlation between SHR and the prognosis of patients with severe mitral valve disease. The results can be summarized as follows: (1) In patients with severe mitral valve disease, a high SHR index is significantly associated with a 28-day all-cause mortality rate, and this association is consistent across different glucose metabolism groups; (2) Further Cox proportional hazard modeling reveals a significant relationship between SHR and 28-day all-cause mortality, especially in the subgroup of patients without abnormal glucose metabolism, where a high SHR index appears to predict a higher mortality rate; (3) Subgroup analysis further demonstrates that the correlation between SHR and mortality remains consistent across various subgroups, including gender, age, glucose metabolism, warfarin use, and mechanical ventilation treatment. Notably, SHR’s predictive power for mortality is more pronounced in patients with prediabetes and in females; (4) In machine learning, a model incorporating eight feature variables, including SHR, also shows predictive value for the prognosis of patients with severe mitral valve disease. These findings may provide valuable guidance for the management of treatment strategies aimed at reducing the mortality risk in patients with severe mitral valve disease. It is important to emphasize that while SHR plays a significant role in predicting the 28-day mortality rate in patients with severe mitral valve disease, it should not be considered an independent determining factor. The feature selection process employed the recursive feature elimination method, which recursively constructs models and selects the best or worst features (based on weights), gradually eliminating less important features to identify the optimal feature subset30. However, this method does not automatically exclude the interactions between features but instead selects features based on their individual importance31,32. Therefore, while SHR is crucial in the predictive model, this does not necessarily imply it is the most decisive factor. Nonetheless, as shown in our Cox regression analysis, a higher SHR is associated with an increased risk of 28-day all-cause mortality in patients with severe mitral valve disease, which is consistent with the RFE algorithm, further confirming that SHR is an important feature. Additionally, incorporating the most predictive variables into various machine learning algorithms, the results indicate that these models exhibit strong performance, further suggesting that SHR can serve as a predictive factor for 28-day all-cause mortality in patients with severe mitral valve disease. SHR typically refers to transient hyperglycemia during the course of a disease, and its underlying mechanism is driven by a highly complex interplay of counter-regulatory hormones, including catecholamines, growth hormone, cortisol, and cytokines33–35. In patients with severe mitral valve disease, further exacerbation of cardiac burden, progressive heart failure, and hemodynamic instability activate the sympathetic nervous system36, leading to increased secretion of adrenaline and noradrenaline37. These stress hormones promote hepatic glucose production via gluconeogenesis, resulting in elevated blood glucose levels38,39. Moreover, in MVD patients experiencing heart failure or undergoing diuretic treatment, the resulting electrolyte disturbances can impair insulin secretion and action, thereby increasing insulin resistance40,41. Elevated blood glucose levels not only damage myocardial cells and impair myocardial contractility but also increase the risk of adverse cardiovascular events in patients with heart valve disease23. In a retrospective study conducted by Huang H, on secondary mitral regurgitation, it was confirmed that higher SHR levels increase the risk of heart failure42. Additionally, in patients with severe aortic stenosis undergoing TAVR, SHR has been shown to exhibit a linear relationship with all-cause mortality, the risk of heart failure-related rehospitalizations and major adverse cardiovascular events43. Similarly, in other cardiac diseases, stress-induced hyperglycemia has been proven to be associated with poor outcomes in patients with acute myocardial infarction15, coronary artery disease, and atrial fibrillation17. In the management of critically ill patients, although previous studies suggested that overly aggressive blood glucose control does not lead to better outcomes, further research has highlighted the significant impact of controlling stress-induced hyperglycemia on prognosis44. A retrospective animal cohort study observed that acute hyperglycemia reduces extracellular Na+, increases intracellular Na+, and elevates pH, thereby exacerbating myocardial injury45. Another study found that acute blood glucose elevation leads to elevated reactive oxygen species (ROS), activated NOGPs, and inhibits myocardial cells’ glucose uptake, improper calcium ion handling, endoplasmic reticulum stress, mitochondrial defects, as well as increased lipid oxidation and triglyceride accumulation in cardiomyocytes, all contributing to myocardial cell damage46,47. Additionally, multiple studies have confirmed that acute hyperglycemia causes the deposition of hexosamine biosynthesis products and further induces cardiomyocyte apoptosis48. Moreover, stress-induced hyperglycemia leads to elevated circulating levels of inflammatory markers such as interleukins (IL-1, IL-6, IL-18) and C-reactive protein (CRP), which in turn promote the upregulation of pro-inflammatory mediators (high-mobility group box 1 (HMGB1) and calprotectin), and enhance the expression of receptors related to stress (TLR2, TLR4, and advanced glycation end product receptors (RAGE)), further amplifying the inflammatory immune response and triggering additional myocardial cell death49,50. In experimental settings, hyperglycemia induces abnormal activation and a pro-inflammatory phenotype shift in endothelial cells and valve interstitial cells, upregulating the expression of cell adhesion molecules (CAMs) and subsequent monocyte adhesion, further worsening valve dysfunction51. Our research further confirmed that in non-diabetic patients, elevated SHR was also significantly associated with an increased 28-day mortality rate, suggesting that SHR not only reflects the degree of glucose metabolism disorder, but is also a comprehensive indicator of acute stress response and systemic metabolic imbalance. For the non-diabetic population, a higher SHR may indicate a more severe metabolic stress state and a higher risk of death. The same finding also exists in other heart diseases such as acute myocardial infarction52. Therefore, the significant correlation between a larger SHR and adverse outcomes in patients with mitral valve disease, regardless of whether they have diabetes or not. It is noteworthy that in the subgroup analysis, we observed a more pronounced correlation between SHR and mortality in patients with mitral valve disease, particularly in the female population. This gender difference has also been highlighted in other studies on heart valve diseases, where it appears that women exhibit a different phenotype of common cardiovascular conditions (including heart failure, arrhythmias, and valve diseases) compared to men53,54. This may be attributed to the less effective cardiovascular interventions in women, as well as factors such as psychological and hormonal changes55. The research also suggests that this disparity may be due to women presenting with more severe symptoms prior to admission, leading to an underestimation of the disease severity in females53,56. This phenomenon has been observed in a cardiovascular disease study on Native Americans, where female patients with diabetes mellitus (DM) exhibited more severe left ventricular abnormalities than their male counterparts57. Additionally, the correlation between SHR and mortality in patients with mitral valve disease is more significant in those with prediabetes, a finding also reported in other studies58. In the research by Capes et al., it was confirmed that SHR is a stronger predictor of in-hospital mortality in diabetic acute coronary syndrome (ACS) patients compared to non-diabetic patients, with a significant difference in in-hospital mortality rates between the two groups59. This may be due to the latter’s increased sensitivity to enhanced glucose toxicity, whereas chronic diabetic patients may have developed antioxidant defense mechanisms in their cells and tissues60,61. Significance of the study Our study demonstrates that SHR, as an easily accessible and uncomplicated measurement, is an effective prognostic indicator. It can assist clinicians in better identifying high-risk patients, thereby reducing mortality and improving patient outcomes. This makes SHR an important aspect of managing critically ill patients in the intensive care unit (ICU) setting. Additionally, the use of machine learning for validation and SHAP analysis further confirmed the incremental prognostic contribution of SHR. We found that a higher SHR is associated with an increased risk of all-cause mortality in patients with severe mitral valve disease. Limitations of the study This study has several limitations. First, it is a retrospective analysis based on observational data from the MIMIC-IV database, conducted at a single medical center, which may introduce information bias. Second, not all potential confounding factors could be controlled, and residual confounding may still affect our results. Third, the blood glucose levels used in the study were the first recorded values at ICU admission, and it remains unclear whether these measurements were taken in fasting patients. Moreover, this study cannot determine the biological validity of the SHR index in predicting outcomes for non-critical mitral valve disease patients. Fourth, because we were unable to identify the primary cause of death in critically ill mitral valve disease patients, this limitation could significantly affect the comprehensive assessment of factors influencing all-cause mortality in MVD patients. These limitations may to some extent affect the robustness and generalizability of the research conclusions. To reduce the influence of these potential biases, future studies can adopt a prospective design, standardize variable definitions and measurements during the data collection stage; at the same time, expand the sample source through multi-center collaborative research to enhance the representativeness and external validity of the results. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Author contributions PNR contributed to the study concept and study design. PNR and LPZ performed the statistical analysis and data interpretation. LPZ, PNR and LYM were responsible for the quality control of the data and algorithms. PNR and LYM performed the literature research and data extraction. All authors contributed to the writing of the manuscript and approved the final version. Data availability The datasets generated /or analysed and during the current study are available in the MIMIC-IV (version 3.1) database, [https://mimic.mit.edu/docs/iv/](https:/mimic.mit.edu/docs/iv) ]. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The datasets generated /or analysed and during the current study are available in the MIMIC-IV (version 3.1) database, [https://mimic.mit.edu/docs/iv/](https:/mimic.mit.edu/docs/iv) ].