Construction and Validation of a Risk Prediction Model for Sepsis-indu

Introduction

Sepsis, as a major challenge in global emergency and critical care medicine, is a life-threatening organ dysfunction caused by a dysregulated host immune response to infection.1 Its incidence and mortality rates remain consistently high, and it has become one of the leading causes of death among emergency and critically ill patients.2,3 According to statistics, annual global sepsis cases exceed 48.9 million, with related deaths reaching 11 million, accounting for 19.7% of global total deaths.4 It is estimated that from 2017 to 2019, there were 4.8 million to 6.1 million hospitalized cases of sepsis in China each year.5 In the multiple organ dysfunction caused by sepsis, myocardial injury is a common and severe complication, referred to as SMCI, with an incidence as high as 13.8% to 79%.6–8 The occurrence of SMCI not only significantly increases the difficulty of treatment but also is closely associated with poor prognosis. Studies have shown that, as an independent risk factor for mortality in sepsis, patients with SMCI have a mortality rate of 32.6% to 43.7%, which is significantly higher than that in patients without myocardial injury.8–10 Although the clinical significance of SMCI has been widely recognized, its early identification still faces numerous challenges. Current diagnosis mainly relies on elevations in myocardial biomarkers (such as troponin T/I, brain natriuretic peptide) or the evidence of cardiac dysfunction demonstrated by echocardiography.8,11 However, elevated troponin levels in sepsis may be influenced by multiple factors such as inflammation and renal injury, leading to insufficient specificity.11 Although echocardiography can assess ventricular systolic and diastolic function, it is affected by factors such as equipment conditions, operator experience, and patients’ hemodynamic status, making it difficult to be applied in real-time and dynamically in emergency departments.12 More importantly, the condition of sepsis patients progresses rapidly. When typical clinical symptoms or definite biomarker abnormalities manifest, myocardial injury has often entered the moderate to advanced stage, missing the optimal timing for intervention. Therefore, the early identification of high-risk patients for SMCI and the timely implementation of targeted interventions have become a key link in improving the prognosis of sepsis patients.

As a quantitative tool that integrates multiple factors, the risk prediction model can provide clinicians with individualized risk assessments and assist in early decision-making by screening independent risk factors and developing visual prediction models (eg, nomograms). It has shown significant value in risk stratification and prognostic evaluation for various diseases.13,14 Currently, existing studies have explored the risk factors for SMCI and found that lactate levels, septic shock, age, hemoglobin levels, etc. are independent risk factors.7,15,16 However, Risk factor analysis fails to intuitively, visually, and conveniently calculate the risk of SMCI occurrence, thus having limited guiding value for clinical practice. Nomograms, as visual tools of the model, intuitively demonstrate the relationship between various clinical variables and the probability of SMCI. By developing a nomogram, rapid risk assessment of myocardial injury can be performed via user-friendly digital interfaces, which facilitates rapid prediction of the probability of myocardial injury in sepsis patients, thereby enabling timely implementation of interventions to reduce morbidity and mortality.17,18 Although Jiangquan Yu et al19 constructed a risk prediction model for SMCI, their study had two limitations: First, the indicators included in model development were dichotomous variables, converting continuous variables into dichotomous variables is not conducive to the accurate calculation of the probability of SMCI occurrence. Second, it lacked model validation and evaluation of clinical applicability. Other studies have addressed these limitations.17,20 However, they included sepsis patients admitted to the intensive care unit (ICU); patients admitted to the ICU generally have a longer disease course, which may result in a certain delay in prediction. In addition, the mechanism of SMCI involve immune dysregulation, cytokine storm, and oxidative stress induced by infection and insufficient tissue perfusion.21 However, their studies incorporated relatively few inflammatory indicators, only routinely including white blood cell count (WBC) and procalcitonin (PCT), but not C-reactive protein (CRP), IL-6, or serum amyloid A (SAA) etc. Thus, this study systematically collected the initial clinical data of sepsis patients at the time of their presentation to the emergency department, including indicators reflecting immunity and cytokine storm, indicators reflecting tissue perfusion, and other commonly used clinical indicators. Using these indicators, an SMCI risk prediction model was constructed. The model’s discrimination, calibration, and clinical applicability were evaluated and internally validated, aiming to provide clinical evidence for the early identification of SMCI. This may facilitate the early detection, early diagnosis, and early treatment of SMCI, thereby improving patient prognosis.

Methods
Patients

A retrospective analysis was performed on the clinical data of 370 sepsis patients admitted to the Emergency Resuscitation Area of the First Affiliated Hospital of Xinjiang Medical University from September 2022 to December 2024. The study was conducted in accordance with the 1964 Declaration of Helsinki and its subsequent amendments, and was approved by the Ethics Committee of the First Affiliated Hospital of Xinjiang Medical University (K202505-32). Because this study was retrospective, informed consent was waived by the Ethics Committee of the First Affiliated Hospital of Xinjiang Medical University. To protect patient privacy, all personally identifiable information (eg, names, hospital identification numbers, national ID numbers) was de-identified during data collection and analysis, thus ensuring that the data were utilized exclusively by the research team in an anonymous format.

Inclusion criteria: 1) Age ≥ 18 years; 2) Diagnosis of sepsis, based on the Sepsis-3 diagnostic criteria:1 patients with infection or suspected infection, combined with an increase of ≥ 2 points in the Sequential Organ Failure Assessment (SOFA) score from baseline at admission; 3) Hospitalization duration >24 hours. Exclusion criteria were as follows: 1) Malignant tumor; 2) History of myocardial infarction, chronic heart failure, chronic renal failure, and chronic hepatic failure; 3) Decompensated liver cirrhosis; 4) Incomplete data. We included 520 patients with sepsis. A total of 150 patients were excluded according to the exclusion criteria. Finally, 370 patients were included in the statistical analysis. The diagnosis of SMCI was based on the following criterion: two consecutive cardiac troponin I (cTnI) levels > 0.04 μg/L in sepsis patients.22 Based on this criterion, patients were divided into the SMCI group (n = 170) and Non-SMCI group (n = 200). The cTnI levels in the two groups were 0.151 (0.074, 0.684) μg/L and 0.012 (0.012, 0.021) μg/L. These patients were randomly assigned to the training cohort or internal validation cohort. A stratified sampling method was adopted, with “serial_number” as the stratification variable, and the dataset was divided into a training set and a validation set at a 7:3 ratio. Specifically, the create Data Partition function in the caret package was used for data splitting, and the random seed was set to 123 to ensure reproducibility. A flowchart of patient enrollment is shown in Figure 1.

Figure 1 Flow diagram of study design.

Abbreviations: LASSO, least absolute shrinkage and selection operator; Myo, myoglobin; BNP, B-type natriuretic peptide; IL-6, interleukin-6.

Data Collection

Comprehensive clinical data were obtained on the first day of admission to emergency resuscitation area, which included: age, gender, SOFA score, Acute Physiology and Chronic Health Evaluation II (APACHE-II), Glasgow Coma Scale (GCS), comorbidities, infection site, temperature, mean arterial pressure (MAP), heart rate, respiratory rate, pH, PaCO2, PaO2, HCO3−, oxygenation index, standard base excess (SBE), K⁺, Na⁺, Ca2⁺, glucose, lactic acid, WBC, absolute neutrophil count (ANC), absolute lymphocyte count (ALC), absolute monocyte count (AMC), hemoglobin, platelet count, fibrinogen, prothrombin time (PT), international normalized ratio (INR), activated partial thromboplastin time (APTT), thrombin time (TT), D-dimer, cTnI, creatine kinase isoenzyme MB (CK-MB), Myo, BNP, CRP, IL-6, PCT, SAA. Since the levels of D-dimer, Myo, BNP, and IL-6 vary by thousands of folds among patients, we performed log transformation on these indicators to reduce the impact of extreme values and improve the applicability of subsequent statistical analyses and the reliability of the results. First, LASSO regression was used to screen for predictive variables. Then, univariate and multivariate logistic regression analyses were performed to identify the independent risk factors for SMCI. Subsequently, a nomogram was constructed using these independent risk factors. The AUC was used to evaluate the discriminative ability of the model, the Hosmer-Lemeshow tests and calibration curve were used to assess calibration, and DCA and CIC were used to evaluate clinical applicability. Meanwhile, internal validation was performed.

Statistical Analyses

Data analyses were performed using R software (version 4.5.1). Quantitative data were expressed as mean ± standard deviation or median (interquartile range [IQR], 25th–75th percentiles). For normally distributed quantitative data, comparisons between two groups were performed using the t-test; for non-normally distributed data, intergroup comparisons were performed using the Mann–Whitney U-test. Categorical data were presented as n (%), and intergroup comparisons were made using the χ2-test. P<0.05 is considered statistically significant, and Benjamini-Hochberg procedure was applied the to control the false discovery rate (FDR) at 5% for multiple comparisons across all clinical variables. Based on multiple linear regression, variance inflation factor (VIF) was calculated to assess the multicollinearity of the variables. To prevent overfitting, LASSO regression was employed to select optimal features. LASSO feature selection was performed using the “glmnet” package. Univariate and multivariate logistic regression analyses were performed using the “glm” package. Model stability was assessed using the events-per-variable (EPV) ratio. Nomograms were plotted using the “regplot” package. ROC curve plotting and AUC calculation were performed using the “pROC” package. Hosmer-Lemeshow tests and calibration plots were generated using the “rms” package. DCAs were performed using the “rmda” package. CICs were plotted using the “ggplot2” package.

Results
Clinical Characteristics

Of the 370 sepsis, 226 were men and 144 were women; 170 had myocardial injury and 200 did not. The numerous statistical tests increase the risk of type I error. Therefore, we use the Benjamini-Hochberg method to calculate the FDR to reduce the risk of type I error in multiple testing. The clinical characteristics of patients in the two groups are presented in Table 1. These patients were randomly assigned to the training cohort or validation cohort at a ratio of 7:3, resulting in 262 patients in the training cohort and 108 in the validation cohort. No significant differences were observed in age, SOFA score, APACHE II score, GCS score, comorbidities, or infection site between the training and validation cohorts, indicating that the two cohorts were comparable (stable 1).

Table 1 A Comparison of the Clinical Characteristics of SMCI and Non-SMCI Groups

Independent Risk Factors in the Training Cohort

As shown in Table 1, there were many variables with statistically significant differences between the SMCI and Non-SMCI groups. Prior to LASSO regression, we calculated the VIF of the variables; since the VIF values of multiple variables exceeded 10, there was a problem of severe multicollinearity. Therefore, we used LASSO regression to eliminate redundant variables and reduce the risks of collinearity and overfitting. In the training cohort, LASSO regression analysis was performed with cross-validation, and 4 variables with non-zero coefficients (Log D-dimer, Log Myo, Log BNP, Log IL-6) were selected based on optimal λ and 1 standard error (Figures 2A and 2B). The VIF values of Log D-dimer, Log Myo, Log BNP, and Log IL-6 were 1.05, 1.03, 1.09, and 1.11, respectively, indicating no substantial multicollinearity in our final model. In our training cohort (n=262), the number of patients with the outcome event (SMCI) was 121, and the number of predictor parameters retained in the final LASSO model was 4. This yields an EPV of 30.3 (121÷4), which substantially exceeds the widely accepted minimum threshold of 10–20 events per variable. This high EPV provides strong evidence for the stability of our prediction model and indicates a very low risk of overfitting. Through univariate and multivariate logistic regression analysis, we determined that Log Myo, Log BNP, and Log IL-6 were independent risk factors for SMCI (Table 2).

Table 2 Results of the Univariate and Multivariate Logistic Regression Analysis Using the Training Cohort

Figure 2 Variable selection for the LASSO regression model. (A) coefficient profile plot was constructed based on the log(λ) sequence. (B) By deriving the optimal lambda, 4 variables with non-zero coefficients were selected. After validating the optimal λ in the LASSO model, we plotted the partial likelihood deviance (binomial deviance) curve against log(lambda), with dashed vertical lines drawn according to the 1-standard error criterion.

Abbreviation: LASSO, least absolute shrinkage and selection operator.

Development of SMCI‑Predicting Nomogram

Based on the results of multivariable Logistic regression analysis: Score = −11.354 + 1.130×Log Myo + 1.403×Log BNP + 0.901×Log IL-6, a nomogram was constructed to predict the risk of SMCI (Figure 3).

Figure 3 Independent risk factors of Log Myo, Log BNP, and Log IL-6 for nomogram prediction model. The significance of the asterisks beside each variable in part represent importance of all the risk factors.

Abbreviations: Myo, myoglobin; BNP, B-type natriuretic peptide; IL-6, interleukin-6.

Evaluation and Validation of Nomogram
Discrimination Ability

The AUC of the nomogram was 0.856 in the training cohort (Figure 4A) and 0.853 in the internal validation cohort (Figure 4B), indicating moderately good performance.

Figure 4 ROC validation of the SMCI risk nomogram prediction. (A) The performance of the nomogram in the training cohort. (B) The performance of the nomogram in the internal validation cohort.

Abbreviations: ROC, receiver operating characteristic; AUC, area under the curve; CI, confidence interval.

Calibration Ability

A calibration curve and Hosmer–Lemeshow test were used to assess the calibration of the predictive model. From the calibration curves, the model showed a good fit in both the training and validation cohorts. As shown by the Hosmer–Lemeshow test, the predicted and actual probabilities were highly consistent (training cohort, P = 0.6229, Figure 5A; validation cohort, P = 0.3553, Figure 5B).

Figure 5 Calibration curve of risk of SMCI. The y-axis represents actual diagnosed cases of SMCI, the x-axis represents the predicted risk of SMCI. (A) training cohort; (B) validation cohort.

Abbreviations: SMCI, sepsis-induced myocardial injury; ROC, receiver operating characteristic.

Clinical Usefulness

DCA showed that in the training cohort, if the threshold probability was within the range of 4%–98%, using this nomogram to predict the risk of SMCI could yield a net benefit (Figure 6A); in the validation cohort, the threshold probability range for net benefit was 13%–86% (Figure 6B). As indicated by the CIC, in the training cohort, when the threshold probability was 58.8% (Figure 6C), the number of high-risk individuals predicted by this nomogram was equal to the number of actual cases; in the validation cohort, this threshold probability was 40.5% (Figure 6D). Taken together, the DCA and CIC results indicate that this model has good clinical applicability.

Figure 6 DCA and CIC of the nomogram prediction model for the risk of cardiac dysfunction in patients with sepsis. (A) DCA for training cohort; (B) DCA for validation cohort; (C) CIC for training cohort; (D) CIC for validation cohort.

Abbreviations: DCA, decision curve analysis; CIC, clinical impact curve.

Discussion

Sepsis patients are at high risk of developing myocardial injury, and those with myocardial injury exhibit an increased risk of mortality.6,9 Developing a risk prediction model for myocardial injury in sepsis patients can help clinicians quickly and accurately assess such risk, thereby enabling them to formulate individualized, targeted intervention strategies to reduce the incidence of myocardial injury and improve patient survival. In this study, four predictive variables (Log D-dimer, Log Myo, Log BNP, and Log IL-6) were screened using LASSO regression. Through univariate and multivariate logistic regression, Log Myo, Log BNP, and Log IL-6 were identified as independent predictors. A nomogram for SMCI risk prediction was developed based on these three factors. For patients, scores of relevant predictive variables are obtained by drawing vertical lines from their variable values to the top score line; the sum of scores of all variables is the total score, and the corresponding predicted probability indicates the patient’s risk of developing SMCI. The AUC of the prediction model was 0.856 in the training cohort and 0.853 in the internal validation cohort, indicating that the model has good discriminative ability. Hosmer-Lemeshow tests and calibration curves revealed that the prediction model also has good calibration. Additionally, the DCAs and CICs further demonstrated that the model has good clinical applicability.

Previous studies have found that the following indicators are independent risk factors for SMCI: elevated plasma histone H4,23 elevated lactate levels,7,16 hemoglobin level,15 low systolic blood pressure,24 and elevated BNP.25 However, a simple risk factor analysis fails to calculate the probability of SMCI occurrence in a convenient, intuitive, and visually accessible manner. Although previous studies have developed risk prediction models for SMCI, their research exhibits the following limitations.17–19 The study by Jiangquan Yu et al19 neither validated the model nor evaluated its clinical applicability. Moreover, their study converted continuous variables into dichotomous categories, such as defining PCT > 40 ng/mL as an independent predictor. In clinical practice, there exists a considerable difference in the severity of infection and inflammatory response between PCT values of 0.1 ng/mL and 35 ng/mL, with PCT > 10 ng/mL generally considered a critical threshold. However, in their model, the risk of SMCI was assigned identically for PCT values of 0.1 ng/mL and 35 ng/mL, which may compromise the accurate assessment of SMCI probability. The studies by Yan Zhuang et al20 and Peng‑fei Sun et al17 primarily utilized clinical data from sepsis patients admitted to the ICU. However, since sepsis patients typically first present to the emergency department, early identification of high-risk individuals and implementation of preventive measures in the emergency setting would be more effective in reducing incidence rates and improving clinical outcomes. Furthermore, the pathogenesis of SMCI involves excessive inflammatory response, cytokine storm, and coagulation dysregulation, while their study only incorporated routine parameters such as WBC and PCT, without including cytokine profiles or coagulation markers. Consequently, it failed to establish the crucial link between pathogenic mechanisms and clinical phenotypes.

This study collected comprehensive data from sepsis patients, including cytokines along with inflammatory, coagulation, hemodynamic, and tissue perfusion parameters etc. Ultimately three key indicators were incorporated into the final model: cytokine (IL-6), myocardial injury marker (Myo), and cardiac dysfunction marker (BNP). These collectively reflect the dynamic progression from infection-induced cytokine storm to myocardial cell injury and cardiac dysfunction in sepsis, thereby bridging pathogenic mechanisms with clinical phenotypes. Cytokines are small proteins secreted by immune cells and stromal cells, which mediate intercellular signaling.21 Sepsis manifests as a dysregulated systemic inflammatory cascade, characterized predominantly by the overproduction of proinflammatory cytokines including tumor necrosis factor-α, IL-1, IL-6, IL-12, and interferons (IFN-α, IFN-β, IFN-γ).26 These cytokines exacerbate inflammation, endoplasmic reticulum stress, mitochondrial dysfunction, and oxidative stress, ultimately leading to cardiomyocyte edema, myocardial injury, and even cardiac dysfunction.21,27 This study has the following advantages: First, we employed LASSO regression to mitigate multicollinearity risks, and subsequent VIF confirmed the absence of multicollinearity among the four predictors. Additionally, we computed an EPV value of 30.3, which provides robust evidence for the stability of our prediction model and indicates a minimal risk of overfitting. Second, we collected commonly used laboratory indicators and disease severity scores within 24 hours of admission for sepsis patients in the emergency resuscitation area. Since sepsis patients primarily present in the emergency resuscitation room, using such commonly used clinical data from the emergency resuscitation room to predict the occurrence of myocardial injury offers advantages in clinical applicability and early prediction. Early identification of patients at high risk of myocardial injury in the emergency resuscitation room, coupled with timely prevention and intervention, can reduce the incidence and mortality of sepsis-induced myocardial injury.

This study has the following limitations. First, as a retrospective analysis, this model lacks real-time applicability and this study cannot prospectively control for biases. Second, although we included commonly used clinical laboratory indicators, disease severity scores, cytokines, coagulation parameters, hemodynamic parameters, and tissue perfusion parameters, factors such as fluid resuscitation, blood transfusion, antibiotic use, and oxygen therapy were not incorporated. Third, this study did not include traumatic sepsis, failing to explore the predictive value of these indicators in myocardial injury in traumatic sepsis. Fourth, although we performed internal validation and evaluated the model’s discriminative ability, calibration, and clinical applicability, the model lacks external or multicenter validation to assess its generalizability. In the future, it is necessary to conduct prospective, multicenter, and longitudinal studies, integrate variables such as treatment-related, hemodynamic, novel inflammatory, and cytokine factors to establish and validate an SMCI risk prediction model.

Conclusion

The proposed nomogram based on Log Myo, Log BNP, and Log IL-6 may serve as a practical tool for early risk assessment of myocardial injury in sepsis, though external validation is required before clinical implementation.

Abbreviations

SMCI, sepsis-induced myocardial injury; LASSO, Least Absolute Shrinkage and Selection Operator; DCA, Decision Curve Analysis; CIC, Clinical Impact Curve; cTnI, cardiac troponin I; SOFA, Sequential Organ Failure Assessment; APACHE-II, Acute Physiology and Chronic Health Evaluation II; GCS, Glasgow Coma Scale; MAP, mean arterial pressure; SBE, standard base excess; WBC, white blood cell count; ANC, absolute neutrophil count; ALC, absolute lymphocyte count; AMC, absolute monocyte count; PT, prothrombin time; INR, international normalized ratio; APTT, activated partial thromboplastin time; TT, thrombin time; CK-MB, creatine kinase isoenzyme; MB, Myo myoglobin; BNP, N-terminal pro-B-type natriuretic peptide; CRP, C-reactive protein; IL-6, interleukin-6; SAA, serum amyloid A; log, logarithmic; FDR, false discovery rate; VIF, variance inflation factor; EPV, events-per-variable; IFN, interferons.

Data Statement

Data are available upon request from Jian-Zhong Yang ([email protected]).

Ethics Statement

The study was approved by the Ethics Committee of the First Affiliated Hospital of Xinjiang Medical University (K202505-32). Because this study was retrospective, so informed consent was waived by the Ethics Committee of the First Affiliated Hospital of Xinjiang Medical University. To protect patient privacy, all personally identifiable information (eg, names, hospital identification numbers, national ID numbers) was de-identified during data collection and analysis, thus ensuring that the data were utilized exclusively by the research team in an anonymous format.

Funding

This study was supported by the In-hospital Program for the Cultivation of Excellent Talents and Innovative Teams of The First Affiliated Hospital of Xinjiang Medical University (cxtd202408) and the National Natural Science Foundation of China grant (82260379).

Disclosure

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References

1. Singer M, Deutschman CS, Seymour CW, et al. The third international consensus definitions for sepsis and septic shock (Sepsis-3). JAMA. 2016;315(8):801–810. doi:10.1001/jama.2016.0287

2. Bréchot N, Hajage D, Kimmoun A, et al. Venoarterial extracorporeal membrane oxygenation to rescue sepsis-induced cardiogenic shock: a retrospective, multicentre, international cohort study. Lancet. 2020;396(10250):545–552. doi:10.1016/S0140-6736(20)30733-9

3. Fan D, Wu R. Mechanisms of the septic heart: from inflammatory response to myocardial edema. J Mol Cell Cardiol. 2024;195:73–82. doi:10.1016/j.yjmcc.2024.08.003

4. Rudd KE, Johnson SC, Agesa KM, et al. Global, regional, and national sepsis incidence and mortality, 1990–2017: analysis for the global burden of disease study. Lancet. 2020;395(10219):200–211. doi:10.1016/S0140-6736(19)32989-7

5. Weng L, Xu Y, Yin P, et al. National incidence and mortality of hospitalized sepsis in China. Crit Care. 2023;27(1):84. doi:10.1186/s13054-023-04385-x

6. Zhou YT, Wang GS, Gao XC, et al. Sepsis-associated myocardial injury: incidence and mortality. Medicine. 2025;104(25):e42513. doi:10.1097/MD.0000000000042513

7. Sato R, Kuriyama A, Takada T, et al. Prevalence and risk factors of sepsis-induced cardiomyopathy: a retrospective cohort study. Medicine. 2016;95(39):e5031. doi:10.1097/MD.0000000000005031

8. Xu KZ, Xu P, Li JJ, et al. Predictors and nomogram of in-hospital mortality in sepsis-induced myocardial injury: a retrospective cohort study. BMC Anesthesiol. 2023;23(1):230. doi:10.1186/s12871-023-02189-8

9. Zhang M, Duan M, Zhi D, et al. Risk factors for 28-day mortality in patients with sepsis-related myocardial injury in intensive care units. J Int Med Res. 2021;49(4):03000605211004759. doi:10.1177/03000605211004759

10. Hanumanthu BKJ, Nair AS, Katamreddy A, et al. Sepsis-induced cardiomyopathy is associated with higher mortality rates in patients with sepsis. Acute Crit Care. 2021;36(3):215–222. doi:10.4266/acc.2021.00234

11. Kim JS, Kim M, Kim YJ, et al. Troponin testing for assessing sepsis-induced myocardial dysfunction in patients with septic shock. JCM. 2019;8(2):239. doi:10.3390/jcm8020239

12. Cavefors O, Einarsson F, Holmqvist J, et al. Cardiac biomarkers for screening and prognostication of cardiac dysfunction in critically ill patients. ESC Heart Failure. 2024;11(6):4009–4018. doi:10.1002/ehf2.14980

13. Min JW, Min JH, Chang SH, et al. A risk prediction model (CMC-AKIX) for postoperative acute kidney injury using machine learning: algorithm development and validation. J Med Internet Res. 2025;27:e62853. doi:10.2196/62853

14. Xu Z, Li F, Xin Y, et al. Prognostic risk prediction model for patients with acute exacerbation of chronic obstructive pulmonary disease (AECOPD): a systematic review and meta-analysis. Respir Res. 2024;25(1):410. doi:10.1186/s12931-024-03033-4

15. Liang YW, Zhu YF, Zhang R, et al. Incidence, prognosis, and risk factors of sepsis-induced cardiomyopathy. WJCC. 2021;9(31):9452–9468. doi:10.12998/wjcc.v9.i31.9452

16. Han YY, Yang JL, Meng HM, et al. Identifying risk factors for myocardial injury in elderly patients with sepsis. Med Sci Monit. 2025;31. doi:10.12659/MSM.947840

17. Sun PF, Wang CJ, Du Y, et al. Development and validation of a nomogram to predict risk of septic cardiomyopathy in the intensive care unit. Sci Rep. 2024;14(1):14114. doi:10.1038/s41598-024-64965-x

18. Huang L, Wang Z, Zhang Y, et al. Development and validation of a nomogram prediction model for the risk of in-hospital death in patients with sepsis complicated by acute pulmonary embolism. Chin Critical Care Med. 2025;37(2):123–127. doi:10.3760/cma.j.cn121430-20240918-00778

19. Yu J, Zheng R, Yang P, et al. Construction of a predictive model and prognosis of left ventricular systolic dysfunction in patients with sepsis based on the diagnosis using left ventricular global longitudinal strain. J Intensive Care. 2022;10(1):29. doi:10.1186/s40560-022-00621-8

20. Zhuang Y, Zhang H, Dai L, et al. Risk factors and development of a clinical prediction model for myocardial dysfunction in sepsis. Chin J TCM WM Crit Care. 2022;29(3):262–267. doi:10.3969/j.issn.1008-9691.2022.03.002

21. Yang W, Cao Y, Li J, et al. Pathogenesis and treatment strategies of sepsis-induced myocardial injury: modern and traditional medical perspectives. Int J Biol Sci. 2025;21(8):3478–3504. doi:10.7150/ijbs.111288

22. Beesley SJ, Weber G, Sarge T, et al. Septic cardiomyopathy. Crit Care Med. 2018;46(4):625–634. doi:10.1097/CCM.0000000000002851

23. Lu NF, Jiang L, Zhu B, et al. Elevated plasma histone H4 levels are an important risk factor for the development of septic cardiomyopathy. Balkan Med J. 2019. doi:10.4274/balkanmedj.galenos.2019.2019.8.40

24. Song MJ, Lee SH, Leem AY, et al. Predictors and outcomes of sepsis-induced cardiomyopathy in critically ill patients. Acute Crit Care. 2020;35(2):67–76. doi:10.4266/acc.2020.00024

25. Jeong HS, Lee TH, Bang CH, Kim JH, Hong SJ. Risk factors and outcomes of sepsis-induced myocardial dysfunction and stress-induced cardiomyopathy in sepsis or septic shock: a comparative retrospective study. Medicine. 2018;97(13):e0263. doi:10.1097/MD.0000000000010263

26. You W. Roles of cytokine storm in sepsis progression: biomarkers, and emerging therapeutic strategies. Front Immunol. 2025;16:1696366. doi:10.3389/fimmu.2025.1696366

27. Boissier F, Aissaoui N. Septic cardiomyopathy: diagnosis and management. J Intensive Med. 2022;2(1):8–16. doi:10.1016/j.jointm.2021.11.004