Development and Validation of a Predictive Nomogram for 30-Day Mortality in Sepsis Patients Coexisting with Malignant Tumors ： a Retrospective Cohort Study Using the MIMIC-IV Database

Development and Validation of a Predictive Nomogram for 30-Day Mortality in Sepsis Patients Coexisting with Malignant Tumors ： a Retrospective Cohort Study Using the MIMIC-IV Database | 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 Development and Validation of a Predictive Nomogram for 30-Day Mortality in Sepsis Patients Coexisting with Malignant Tumors ： a Retrospective Cohort Study Using the MIMIC-IV Database Lin Qian, Weiting Sun, Peng Ding, Song Zhang, Kunlan Long This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6007779/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Sepsis is the main cause of death for cancer patients, and our study aims to evaluate risk factors and develop a model to predict the 30-Day mortality in sepsis patients coexisting with malignant tumors. Methods We obtained 4196 sepsis patients coexisting with malignant tumors from the MIMIC-IV database and randomly split them into a training set (2937 patients) and a validation set (1259 patients) at a ratio of 7:3. A multivariable logistic regression model was used to identify independent risk factors for predicting mortality, and the model's performance was evaluated. Results Multivariable logistic regression analysis showed that age, gender, CPD, diabetes, AKI, heart rate, APACHE III, cardiovascular system, anion gap, BUN, calcium, creatinine, bilirubin, pH, and PCO2 were independent risk factors. The nomogram achieved optimal performance in discrimination, calibration, and clinical application. Conclusion The nomogram effectively predicts the 30-Day mortality in sepsis patients coexisting with malignant tumors, and internal validation confirms its effectiveness. The study results can help clinical doctors make decisions on the treatment of these patients, thus reducing the risk of sepsis and death for cancer patients. malignant tumors sepsis nomogram 30-Day mortality MIMIC-IV Figures Figure 1 Figure 2 Figure 3 Figure 4 1 Introduction Sepsis and septic shock affect millions of people worldwide each year, with a mortality rate of 1/4 [ 1 ]. The mortality rate is even higher, exceeding 40%, for septic shock [ 2 ]. Important risk factors for sepsis include advanced age, low socioeconomic status, obesity, presence of chronic diseases (including chronic obstructive pulmonary disease, cancer, kidney or liver disease, and diabetes), as well as impaired immune function due to human immunodeficiency virus (HIV) infection or use of immunosuppressive drugs [ 3 ]. The risk of sepsis in cancer patients is tenfold higher than in non-cancer patients, and sepsis is the most common reason for their admission to the intensive care unit (ICU), with higher in-hospital mortality rates [ 4 , 5 ]. Several large retrospective analyses have shown that nearly 20% of patients admitted to the ICU for sepsis have underlying malignant tumors [ 6 – 8 ]. Malignant tumors are also independently associated with an increased risk of in-hospital mortality for sepsis patients [ 7 ]. Despite significant improvements in the survival rates of sepsis patients with coexisting malignant tumors in the past two decades, sepsis remains a major cause of death for cancer patients [ 9 , 10 ]. The predictive value of existing critical illness scoring systems for cancer patients is still unclear. Our aim is to establish a predictive model by combining multiple prognostic factors to evaluate the outcomes of these patients. 2 Materials and methods 2.1 Database and study population The data employed in this research were procured from the Medical Information Mart for Intensive Care (MIMIC)-IV, version 2.2. MIMIC-IV, an accessible online clinical database, is renowned in the field of critical care research, incorporating information from in excess of 50,000 intensive care unit (ICU) admissions for adult patients at the Beth Israel Deaconess Medical Center, Boston, Massachusetts, spanning the period from 2008 to 2019. Patient informed consent was not necessary for this research as the database received approval from the Institutional Review Committee at MIT and the Beth Israel Deaconess Medical Center. To safeguard patient confidentiality, the MIMIC-IV database has undergonethorough de-identification of all personal information. In the MIMIC-IV v2.2 dataset, a total of 299712 patients generated 431231 hospitalization records, of which 73181 were ICU admissions. According to the Sepsis 3.0 criteria, 33177 patients within the ICU were diagnosed with sepsis, and among them, 5312 patients had concurrent malignant tumors. After excluding patients younger than 18 years of age, those with non-initial ICU admissions, and those who passed away within 24 hours of ICU admission, a cohort of 4196 septic patients with malignant tumors was established for this study. These patients were randomly divided into training and validation sets at a 7:3 ratio, comprising 2937 and 1259 patients, respectively (Fig. 1 ). 2.2 Data collection To enhance the simplicity of the model, we selected variables that are readily available in clinical settings. The collected data encompasses patient demographic information (age, gender, weight), complications (myocardial infarction, congestive heart failure, chronic pulmonary disease, liver disease, diabetes, renal disease, malignant cancer, metastatic solid tumor, acute kidney injury), vital signs (heart rate, respiratory rate, temperature, SpO 2 ), severity of illness (Sequential Organ Failure Assessment, Acute Physiology and Chronic Health Evaluation III), extracorporeal life support measures (cardiovascular, renal replacement therapy, ventilation), and laboratory indicators (white blood cells, hemoglobin, platelets, anion gap, blood urea nitrogen, calcium, creatinine, glucose, sodium, potassium, international normalized ratio, prothrombin time, activated partial thromboplastin time, bilirubin, lactate, pH, PO 2 , PCO 2 ). All clinical-related covariates are based on the patient's indicators on the first day of ICU admission. If the same indicator has multiple measurements on the first day of ICU admission, the worst result from that day is chosen. All of the aforementioned data were extracted from the MIMIC-IV database using Structured Query Language (SQL) and facilitated by Navicat Premium 16. 2.3 Variables Selection To ensure that our model neither overfits nor underfits, we employed five advanced statistical strategies for variable selection in the training cohort: the best subsets regression (BSR), the least absolute shrinkage and selection operator (LASSO), and the forward stepwise regression (FSR). For BSR, we used the Bayesian Information Criterion (BIC) as the basis for variable selection, while for FSR, we utilized the Akaike Information Criterion (AIC). Both random forests and gradient boosting trees are ensemble learning methods, which enhance prediction accuracy and mitigate overfitting by building and combining multiple decision trees. In the variable selection process, we also employed the feature importance derived from Random Forest (RF) and Gradient Boosting Trees (GBT). 2.4 Statistical analysis In the data preprocessing stage, we handled missing values to ensure the robustness of subsequent statistical analyses. Given the potential nonlinear relationships among variables, we adopted a random forest-based imputation method to deal with the missing data. We partitioned the entire dataset into two subsets by randomly splitting it in a 7:3 ratio. To ensure the accuracy of the model construction, 70% of the data was designated as the training cohort for model selection and development, while the remaining 30% served as the validation cohot. Baseline characteristics are presented separately for the original cohort, training set, and validation set. Categorical variables are described as percentages (%). Continuous variables with non-normal distributions are represented by the median and interquartile ranges (IQRs), while those with normal distributions are expressed as the mean and standard deviation (Mean ± SD). Differences between categorical variables were assessed using the chi-square test, while differences between continuous variables were evaluated using the t-test or non-parametric tests. The performance of the model was evaluated in terms of calibration and discrimination in both the training and validation sets. The calibration capability of the prediction plot was assessed using calibration curves, while the Harrell's Consistency Index (C-index) and the Area Under the Receiver Operating Characteristic (ROC) Curve (AUC) were both employed to assess discrimination. Additionally, the clinical utility of the model for septic patients with malignant tumors was assessed using Decision Curve Analysis (DCA). Furthermore, the three models selected based on BSR, LASSO, and FSR were compared in terms of discrimination, calibration, and clinical decision-making. The variables that were ultimately determined as independent risk factors were incorporated into the final logistic regression model, and the corresponding nomogram was plotted. All statistical analyses were performed using the Python software (version 3.9, https://www.python.org ), R software (version 4.3.0, http://www.R-project.org ), and Free Statistics software version 1.9. This study was conceptualized and conducted in alignment with the guidelines provided by the TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis) statement. 3 Results 3.1 Clinical Characteristics The baseline clinical and demographic characteristics of the training and validation cohorts can be found in Table 1. The primary cohort comprised 4196 sepsis patients with concurrent malignant tumors, of which 2937 were assigned to the training set and 1,259 to the validation set. Within this primary cohort, the general age distribution was approximately 66.6 ± 13 years, with males accounting for 61.7% of the population. A review of the baseline characteristics revealed commendable consistency in demographic characteristics, comorbidities, vital signs, severity of illness scores, extracorporeal life support measures, and laboratory results between the training and validation sets. Among the comparative analyses of the variables between the two groups, only the history of renal disease presented a significant difference (p = 0.035), whereas the p-values for the remaining variables exceeded 0.05. Before imputation using the random forest method, various baseline variables exhibited certain levels of missingness. Specific proportions and counts of missing data for each variable can be found in Supplementary Table 1. 3.2 Variable Selection via BSR The BSR method excels in variable selection due to its comprehensive computation of all possible variable combinations. The ultimate selection criterion is based on the minimum Bayesian Information Criterion (BIC). As illustrated in Figs. 2 A and 2 B, the report encompasses an evaluation of all 38 parameters, resulting in a BIC value of -662. The final number of selected variables is 8, which is based on the minimum BIC value in Fig. 2 A. The variables in Fig. 2 B with a BIC value less than − 660 represent the coefficients that constitute the final variable combination determined through BSR. This combination includes 8 variables from the training cohort: weight, acute kidney injury, heart rate, temperature, Apache II score, BUN, creatinine, and glucose. The relationship of all selected variables with the outcomes is detailed in Table 2 and Supplementary Table 2. Subsequently, Model 1 was established based on the variables screened by BSR. 3.3 Variable Selection via LASSO To identify the pivotal variables within our predictive model, we further employed the LASSO logistic regression method. LASSO achieves variable selection and complexity adjustment by penalizing coefficients, enhancing both predictive accuracy and interpretability of the model. This approach is particularly suited for addressing multicollinearity and high-dimensional data. We initiated the process by standardizing all predictive variables, and through cross-validation, we ascertained two critical regularization parameters (λ values): lambda.min and lambda.1se. Figure 2 C illustrates the coefficient paths of individual variables, demonstrating which variables enter the model first and which ones persist as λ values change. Figure 2 D portrays the relationship between model mean squared error and log(λ), with 23 variables included in the model at lambda.1se = 0.01: Age, Gender, Weight, CPD, Diabetes, AKI, Heart Rate, Temperature, SpO 2 , APACHE III, Cardiovascular, RRT, Ventilation, Anion gap, Bun, Calcium, Creatinine, Glucose, Sodium, Bilirubin, pH, PO 2 , PCO 2 . The relationships of all selected variables with the outcomes are detailed in Table 2 and Supplementary Table 2. Subsequently, Model 2 was established based on the variables screened by LASSO. 3.4 Variable Selection via FSR Stepwise regression automates the selection of independent variables most relevant to the dependent variable, thereby enhancing the predictive performance of the model. By iteratively adding or removing independent variables, it aids in constructing a more concise and high-performance model. First, we executed a forward stepwise regression based on the Akaike Information Criterion (AIC), commencing with zero variables and assessing the impact of introducing a new variable into the model at each step. Figure 2 E displays the AIC values at each step of the stepwise regression process. AIC values exhibited a decreasing trend with the addition of each new predictive variable, reflecting the incremental model fit. In the 10th step, following the inclusion of the 10th variable 'Creatinine,' the model's AIC reached its minimum at 2983.10. Figure 2 F illustrates the variation in P-values for each variable throughout the stepwise regression process. The red dashed line indicates the significance threshold of 0.05. Notably, each variable in the final model, at the step of its inclusion, displayed P-values significantly below the 0.05 threshold, implying their significant predictive power within the model. The stepwise regression analysis identified the following 10 variables as predictive factors for the final logistic regression model: Weight, AKI, Heart Rate, Temperature, SpO 2 , APACHE III, Ventilation, Bun, Creatinine, Glucose. These variables consistently exhibited significant p-values in their respective steps and substantially reduced the AIC upon inclusion in the model, thereby confirming their inclusion in Model 3 (Table 2 and Supplementary Table 2). 3.5 Variable Selection via RF and GBT Random Forest and Gradient Boosting Trees are both ensemble learning methods that improve prediction accuracy and control overfitting by constructing and combining multiple decision trees. We utilized the Random Forest Classifier and Gradient Boosting Classifier classes from the Scikit-learn library to train the Random Forest and Gradient Boosting Tree models, respectively. Feature importance is computed based on the frequency and depth of splits within the trees, offering a quantitative measure of each feature's contribution to the model's predictive capabilities. From Supplementary Fig. 1A and Supplementary Fig. 1B, it is evident that the top three most important variables are APACHE III, Temperature, and Heart Rate. These variables were consistently included in all three aforementioned models. 3.6 Comparison of Model Performance and Establishment of the Final Model In the training cohort, each significant variable underwent an initial assessment through univariate logistic regression analysis (Supplementary Table 2). Subsequently, multivariable logistic regression analysis revealed that several variables, comprising Age, Gender, CPD, Diabetes, AKI, Heart Rate, APACHE III, Cardiovascular, Anion gap, BUN, Calcium, Creatinine, Bilirubin, pH, PCO 2 , emerged as independent risk factors, whereas Weight, Temperature, SpO 2 , RRT, Ventilation, Glucose, Sodium, PO 2 were identified as independent protective factors (Table 2). Across the three regression models examined, Model 2 consistently exhibited the highest C-index in both the training and validation cohorts. However, the inclusion of 23 variables in Model 2 resulted in a more cumbersome model. Model 1, on the other hand, displayed the lowest C-index in both cohorts, but it also comprised the fewest variables, totaling eight. Model 3's C-index and number of variables placed it between Models 1 and 2 (Table 2). In the training cohort, Model 2 achieved the highest area under the curve (AUC), followed by Model 3, and then Model 1, with specific values being AUC1: 0.785 (0.768, 0.803), AUC2: 0.802 (0.785, 0.819), and AUC3: 0.790 (0.773, 0.808). The comparison among the three yielded a p-value of less than 0.01 (Fig. 3 A). Similarly, in the validation cohort, Model 2 had the highest AUC, with Model 3 and Model 1 trailing, respectively. The exact values were AUC1: 0.789 (0.762, 0.815), AUC2: 0.793 (0.767, 0.819), and AUC3: 0.790 (0.764, 0.816), with a comparative p-value of less than 0.01 (Fig. 3 B). We subsequently compared the calibration of the three models in both the training and validation cohorts (Fig. 3 C-D). In Figs. 3 C and 3 D, the blue dashed line represents Model 1, the green solid line denotes Model 2, and the red dashed line signifies Model 3. The ideal calibration curve coincides with the 45° black dashed line in the figures, which represents a scenario where the model's predicted probabilities perfectly match the observed event rates. In practice, the closer a model's curve aligns with the ideal line, the more its predicted probabilities correspond to the actual event occurrences. As evident from the figures, all three models demonstrate high consistency in the training cohort. Although the calibration curves deviate somewhat from the ideal 45° line in the validation cohort, the variations among the three models are minimal. In the training cohort's Hosmer-Lemeshow (H-L) test, Model 1 yielded an X1-squared value of 11.84 with an H-L P-value of 0.159. For Model 2, the X2-squared value was 5.61 with an H-L P-value of 0.691, and for Model 3, the X3-squared value was 5.44 with an H-L P-value of 0.709. In the validation cohort's H-L test, Model 1 had an X1-squared value of 13.18 with an H-L P-value of 0.106, Model 2 presented an X2-squared value of 11.44 with an H-L P-value of 0.178, and Model 3 recorded an X3-squared value of 9.32 with an H-L P-value of 0.316. All P-values exceeded 0.05, suggesting that the observed event rates for the three models closely aligned with their predicted probabilities, indicating satisfactory calibration (Table 2). As depicted in Fig. 3 E-F, decision curve analysis demonstrates that for the training cohort with a threshold probability ranging from 8–82%, and the validation cohort ranging from 7–86%, the use of the line chart to predict the 30-day mortality rate offers a net benefit surpassing both the "treat all" and the "treat none" strategies. This underscores the clinical utility of the nomogram established by Model 1. The clinical relevance of the nomogram derived from Model 2 is evident in the training cohort with a threshold probability between 4% and 88%, and in the validation cohort between 4% and 98%. For Model 3, its nomogram displays clinical applicability within the training cohort for threshold probabilities of 5–88% and in the validation cohort from 5–92%. The research findings suggest that nomograms developed from all three models can benefit sepsis patients with concomitant malignant tumors. Model 2 exhibited the best performance in terms of discrimination, calibration, and clinical utility. However, its inclusion of 23 variables significantly hinders its future clinical application, as complex models carry a higher risk of overfitting. Model 1, incorporating only 8 variables, achieved model performance comparable to both Model 3 and Model 2 (Table 2). A more concise model is favorable for future clinical use, which is why we constructed the line chart based on Model 2 (Fig. 4 ). Additionally, the variables selected in Model 1 had minimal missing data in the database. Only the body temperature had a missing rate of 1.35%, while the missing rates for the other variables were all below 0.67% (Supplementary Table 1). After developing the prediction model with Model 1 and categorizing the data based on risk scores, we observed that the high-risk group consistently exhibited the highest mortality risk at 30 days, 60 days, and 100 days (Supplementary Fig. 2). Concurrently, we presented a consolidated view of the relationships between the variables in Model 1 and the outcomes (Supplementary Fig. 3). Through decision curve analysis, we identified that at most threshold probabilities accepted by patients or clinicians, our prediction model offers a superior net benefit over other strategies. This suggests that, within these thresholds, employing our model for clinical decision-making can yield significant benefits for sepsis patients with concurrent malignant tumors (Supplementary Fig. 4). 4 Discussion Currently, there are almost no models available for predicting the risk of death within 30 days in septic patients with concomitant malignant tumors. In this study, we used multivariable logistic regression analysis to identify independent risk factors for death in septic patients with concomitant malignant tumors in MIMIC-IV and developed a simplified model consisting of 8 predictive factors: weight, AKI (acute kidney injury), heart rate, temperature, APACHE III (Acute Physiology and Chronic Health Evaluation III), blood urea nitrogen, creatinine, and blood glucose. These variables are easily obtainable clinical indicators. The model showed good predictive ability and discriminatory power with favorable results in both the training and validation datasets. However, it is important to note that our study primarily analyzed data from MIMIC-IV and is only applicable to ICU patients, and the potential for selection bias in the data cannot be ignored. The APACHE (Acute Physiology and Chronic Health Evaluation) scoring system is a tool used to assess and predict the severity and prognosis of patients. APACHE II and APACHE III are two commonly used versions. The APACHE II scoring system is mainly based on physiological parameters, age, and disease type, among other factors. It includes 12 physiological parameters such as heart rate, temperature, blood pressure, as well as disease type and age. The APACHE III scoring system builds upon APACHE II by including additional physiological parameters and disease diagnostic categories, including indicators related to hematology, liver function, and the nervous system [ 11 , 12 ]. Due to the consideration of a wider range of physiological parameters and disease categories, APACHE III may be more accurate in predicting patient prognosis compared to APACHE II. Several studies have confirmed that APACHE II score is an independent predictive factor for in-hospital and ICU survival rates in septic patients with malignant tumors [ 13 – 15 ]. However, there is limited research on the relationship between APACHE III score and sepsis in malignant tumor patients. Our study found that APACHE III score is an independent risk factor. Sepsis is a systemic inflammatory response caused by the introduction of bacteria or other infective microorganisms into the body. If sepsis is not treated promptly or inappropriately, the inflammatory response may lead to hemodynamic instability, tissue ischemia, cell damage, and other factors that can negatively affect the kidneys [ 16 ]. Increasing evidence suggests a higher risk of infection or sepsis following acute kidney injury (AKI), and some experts suggest considering AKI as an early sign of sepsis [ 17 , 18 ]. For septic patients with concomitant malignant tumors, it is important to prevent infection as much as possible, enhance immunity, and identify and treat complications early. If AKI occurs, measures such as renal replacement therapy should be promptly taken to maintain the patient's life, and monitoring and management of treatment should be intensified. Metabolic disturbances in glucose metabolism are common in septic patients. A large cohort study showed that hyperglycemia is associated with increased ICU mortality in septic patients, and the impact of blood glucose on mortality increases with the severity of sepsis [ 19 , 20 ]. Another multicenter prospective cohort study found that the mortality rate in septic patients with hypoglycemia is 2.5 times higher than that in septic patients with normal blood glucose levels [ 21 ]. The NICE-SUGAR study also reported a strong association between hypoglycemia and mortality [ 22 ]. It is evident that the mortality rate is higher in septic patients with hyperglycemia or hypoglycemia compared to patients with normal blood glucose levels, highlighting the importance of maintaining stable blood glucose concentrations. 5 Conclusion The predictive model includes 8 predictive factors, which are weight, AKI, heart rate, temperature, APACHE III, BUN, creatinine, and glucose. The model effectively predicts the risk of death within 30 days for sepsis with malignant tumors. Internal validation has demonstrated its effectiveness. The research findings aid clinical physicians in making treatment decisions for these patients, thus reducing the risk of sepsis and death in malignancy patients. Declarations Conflict of Interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Ethics approval and consent to participate The data used in this study were acquired from an open source and do not require approval by any ethical committee. Competing interests The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interests. Funding No funding. Author Contribution Lin Qian and Weiting Sun: Conceptualization, Writing – original draft, Formal analysis, Writing – review & editing, Validation, Visualization, Investigation. Peng Ding: Investigation. Song Zhang: Conceptualization, Writing – review & editing, Project administration. Kun-lan Long: Supervision. Data Availability Statement The original contributions presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding authors. References Rhodes A, Evans LE, Alhazzani W, Levy MM, Antonelli M, Ferrer R, et al. Surviving Sepsis Campaign: International Guidelines for Management of Sepsis and Septic Shock: 2016. Intensive Care Med (2017) 43(3):304-377. doi: 10.1007/s00134-017-4683-6. Singer M, Deutschman CS, Seymour CW, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA. 2016, 315(8):801-10. doi: 10.1001/jama.2016.0287. Mayr FB, Yende S, Angus DC. Epidemiology of severe sepsis. Virulence. 2014,5(1):4-11. doi: 10.4161/viru.27372. Nathan N, Sculier JP, Ameye L, et al. Sepsis and Septic Shock Definitions in Patients With Cancer Admitted in ICU. J Intensive Care Med. 2021,36(3):255-261. doi: 10.1177/0885066619894933. Mirouse A, Vigneron C, Llitjos JF, et al. Sepsis and Cancer: An Interplay of Friends and Foes. Am J Respir Crit Care Med. 2020,202(12):1625-1635. doi: 10.1164/rccm.202004-1116TR. Annane D, Aegerter P, Jars-Guincestre MC, et al. Current epidemiology of septic shock: the CUB-Réa Network. Am J Respir Crit Care Med. 2003,168(2):165-72. doi: 10.1164/rccm.2201087. Vincent JL, Rello J, Marshall J, et al. International study of the prevalence and outcomes of infection in intensive care units. JAMA. 2009,302(21):2323-9. doi: 10.1001/jama.2009.1754. Hensley MK, Donnelly JP, Carlton EF, et al. Epidemiology and Outcomes of Cancer-Related Versus Non-Cancer-Related Sepsis Hospitalizations. Crit Care Med. 2019,47(10):1310-1316. doi: 10.1097/CCM.0000000000003896. Williams JC, Ford ML, Coopersmith CM. Cancer and sepsis. Clin Sci (Lond). 2023,137(11):881-893. doi: 10.1042/CS20220713. Rosolem MM, Rabello LS, Lisboa T, et al. Critically ill patients with cancer and sepsis: clinical course and prognostic factors. J Crit Care. 2012,27(3):301-7. doi: 10.1016/j.jcrc.2011.06.014. Knaus WA, Draper EA, Wagner DP, et al. APACHE II: a severity of disease classification system. Crit Care Med. 1985,13(10):818-29. Knaus WA, Wagner DP, Draper EA, et al. The APACHE III prognostic system. Risk prediction of hospital mortality for critically ill hospitalized adults. Chest. 1991,100(6):1619-36. doi: 10.1378/chest.100.6.1619. Kopterides P, Liberopoulos P, Ilias I, et al. General prognostic scores in outcome prediction for cancer patients admitted to the intensive care unit. Am J Crit Care. 2011,20(1):56-66. doi: 10.4037/ajcc2011763. Namendys-Silva SA, Texcocano-Becerra J, Herrera-Gómez A. Prognostic factors in critically ill patients with solid tumours admitted to an oncological intensive care unit. Anaesth Intensive Care. 2010,38(2):317-24. doi: 10.1177/0310057X1003800214. Aygencel G, Turkoglu M, Turkoz Sucak G, et al. Prognostic factors in critically ill cancer patients admitted to the intensive care unit. J Crit Care. 2014,29(4):618-26. doi: 10.1016/j.jcrc.2014.01.014. Chang YM, Chou YT, Kan WC, et al. Sepsis and Acute Kidney Injury: A Review Focusing on the Bidirectional Interplay. Int J Mol Sci. 2022,23(16):9159. doi: 10.3390/ijms23169159. Lai TS, Wang CY, Pan SC, et al. Risk of developing severe sepsis after acute kidney injury: a population-based cohort study. Crit Care. 2013,17(5):R231. doi: 10.1186/cc13054. Manrique-Caballero CL, Del Rio-Pertuz G, Gomez H. Sepsis-Associated Acute Kidney Injury. Crit Care Clin. 2021,37(2):279-301. doi: 10.1016/j.ccc.2020.11.010. Lu Z, Tao G, Sun X, et al. Association of Blood Glucose Level and Glycemic Variability With Mortality in Sepsis Patients During ICU Hospitalization. Front Public Health. 2022,10:857368. doi: 10.3389/fpubh.2022.857368. Wang J, Zhu CK, Yu JQ, Tet al. Hypoglycemia and mortality in sepsis patients: A systematic review and meta-analysis. Heart Lung. 2021,50(6):933-940. doi: 10.1016/j.hrtlng.2021.07.017. Kushimoto S, Abe T, Ogura H, et al. Impact of blood glucose abnormalities on outcomes and disease severity in patients with severe sepsis: An analysis from a multicenter, prospective survey of severe sepsis. PLoS One. 2020,15(3):e0229919. doi: 10.1371/journal.pone.0229919. NICE-SUGAR Study Investigators; Finfer S, Liu B, Chittock DR, et al. Hypoglycemia and risk of death in critically ill patients. N Engl J Med. 2012,367(12):1108-18. doi: 10.1056/NEJMoa1204942. Tables Table 1 and 2 are available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files Table1.xlsx Table 1. Basic characteristics of the training and validation cohorts. Table2.xlsx Table 2. The relationship of selected variables with the outcomes in three models. TableS1.xlsx Supplementary Table 1. Percentage of missing data in the variables of interest. TableS2.xlsx Supplementary Table 2. The relationship of all selected variables with the outcomes. FigureS1.jpg Supplementary Figure 1. Ranks of variable importance in the RF model (A) and GBT model (B). BUN, blood urea nitrogen; PO2, partial pressure of oxygen; MBP, mean blood pressure; PH, potential of hydrogen; AKI, acute kidney injury; APTT, activated partial thromboplastin time; SPO2, pulse oxygen saturation; PCO2, partial pressure of carbon dioxide; CPD, chronic pulmonary disease; CHF, congestive heart failure; MI, myocardial infarction; RRT, renal replacement therapy. FigureS2.jpg Supplementary Figure 2. Kaplan-Meier survival curve of 30-Day (A), 60-Day (B), 100-Day (C) cumulative survival rate for 4 groups. Develop a predictive model using Model 1 and divide the data evenly into four groups based on risk scores. The group 'predictors=4', which was assessed as high-risk by the model, had the highest number of deaths, while the group assessed as low-risk by the Model 1 had relatively fewer deaths. FigureS3.jpg Supplementary Figure 3. Receiver operating characteristic (ROC) curves for predictors related to in-hospital mortality index. The solid red line indicates the ROC curve for APACHE III, and the AUC was better for APACHE III. FigureS4.jpg Supplementary Figure 4. Decision curve analysis for prediction model. Cite Share Download PDF Status: Posted Version 1 posted 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-6007779","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":416197085,"identity":"2a65f938-d5cd-477a-8b58-4a553ae61a8b","order_by":0,"name":"Lin Qian","email":"","orcid":"","institution":"Hospital of Chengdu University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Lin","middleName":"","lastName":"Qian","suffix":""},{"id":416197086,"identity":"3e7178a9-0ebb-4a99-a8e7-dca8e6caa00d","order_by":1,"name":"Weiting Sun","email":"","orcid":"","institution":"Hospital of Chengdu University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Weiting","middleName":"","lastName":"Sun","suffix":""},{"id":416197087,"identity":"2047af99-7554-4c6f-800a-437bd607f21e","order_by":2,"name":"Peng Ding","email":"","orcid":"","institution":"Hospital of Chengdu University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Peng","middleName":"","lastName":"Ding","suffix":""},{"id":416197088,"identity":"595dcb5d-45e0-42b8-bce0-76be41299e57","order_by":3,"name":"Song Zhang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8UlEQVRIiWNgGAWjYDACCRiDvbHhQEKFhJw88Vp4Dh988OGMhbFhA9FaJNKSDWe2VSQyHCCgQ35287OHX9sOy5k35JhJ886TSGBsYH746AYeLYxzjpkby7YdNpY5cAaoZZtEHjsDm7FxDh4tzBIJZtKSbYcTZzD2gLUUMzbwsEnj08Imkf4NpKV+BjMPUMscicSGAwS08EjkmEl+bDucIMHGBvR+AxFaJCRyyqQZzqUbzuBhBgbyMQljw2YCfpGfkb5N8keZtbyE/ENgVNbUycmzNz98jE8LCDDzsqFwCSgHAcYff4hQNQpGwSgYBSMXAAA+Ykk0r4GF6QAAAABJRU5ErkJggg==","orcid":"","institution":"Hospital of Chengdu University of Traditional Chinese Medicine","correspondingAuthor":true,"prefix":"","firstName":"Song","middleName":"","lastName":"Zhang","suffix":""},{"id":416197089,"identity":"d524511e-79b1-4557-8864-7151c8f6fc65","order_by":4,"name":"Kunlan Long","email":"","orcid":"","institution":"Hospital of Chengdu University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Kunlan","middleName":"","lastName":"Long","suffix":""}],"badges":[],"createdAt":"2025-02-11 13:23:29","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6007779/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6007779/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":76576020,"identity":"2e831d9d-b9c2-4967-8f03-43c6384c6c1a","added_by":"auto","created_at":"2025-02-18 14:12:38","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":268971,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of the study.\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/fdbcaee521390eb2f44253fe.jpg"},{"id":76574198,"identity":"35a93254-f290-4a8b-b2ee-750b8eb9e62b","added_by":"auto","created_at":"2025-02-18 14:04:38","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":433104,"visible":true,"origin":"","legend":"\u003cp\u003eVariables selection. (A, B) BSR method. (C, D) LASSO logistic regression method. (E, F) Logistics stepwise regression.\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/81bb83565b9c3e4e4f068013.jpg"},{"id":76574196,"identity":"f86c6780-e422-4039-8a6a-7124e7dcec5b","added_by":"auto","created_at":"2025-02-18 14:04:38","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":415732,"visible":true,"origin":"","legend":"\u003cp\u003eCalibration curves of the predicted nomogram in the trainingt cohort (A), validation cohort (B). The calibration of the three models in the training cohort (C), validation cohort (D). Decision analysis curves of training cohort (E) and validation cohort (F).\u003c/p\u003e","description":"","filename":"Figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/946df21d0e0406987f4272cc.jpg"},{"id":76576691,"identity":"6855ad51-8cb3-46cf-8efa-c244f168de37","added_by":"auto","created_at":"2025-02-18 14:20:38","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":774832,"visible":true,"origin":"","legend":"\u003cp\u003eNomogram for prediciting 30-Day mortality in sepsis patients coexisting with malignant tumors. AKI, acute kidney injury; BUN, blood urea nitrogen.\u003c/p\u003e\n\u003cp\u003eSupplementary Figure 1. Ranks of variable importance in the RF model (A) and GBT model (B). BUN, blood urea nitrogen; PO2, partial pressure of oxygen; MBP, mean blood pressure; PH, potential of hydrogen; AKI, acute kidney injury; APTT, activated partial thromboplastin time; SPO2, pulse oxygen saturation; PCO2, partial pressure of carbon dioxide; CPD, chronic pulmonary disease; CHF, congestive heart failure; MI, myocardial infarction; RRT, renal replacement therapy.\u003c/p\u003e","description":"","filename":"Figure4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/950d0745d4fd2acc2358b66b.jpg"},{"id":76923679,"identity":"914d6a75-25e1-47c8-ab9e-c154da4d8b36","added_by":"auto","created_at":"2025-02-22 12:16:38","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2510631,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/32b42822-c531-4c94-96ba-d51cde288ccd.pdf"},{"id":76576688,"identity":"6a6f2f23-60af-4133-84f4-06ec65bb348e","added_by":"auto","created_at":"2025-02-18 14:20:38","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":14489,"visible":true,"origin":"","legend":"\u003cp\u003eTable 1. Basic characteristics of the training and validation cohorts.\u003c/p\u003e","description":"","filename":"Table1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/6c754581c3adee13ee2f80fc.xlsx"},{"id":76576024,"identity":"145470a9-2205-469d-9dcd-e25daf8d6439","added_by":"auto","created_at":"2025-02-18 14:12:38","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":13306,"visible":true,"origin":"","legend":"\u003cp\u003eTable 2. The relationship of selected variables with the outcomes in three models.\u003c/p\u003e","description":"","filename":"Table2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/d3aebf9b20bbcf2f3be46d4a.xlsx"},{"id":76574193,"identity":"e7bdf562-4ab4-4e87-99ab-f73f7fe42717","added_by":"auto","created_at":"2025-02-18 14:04:38","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":12217,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Table 1. Percentage of missing data in the variables of interest.\u003c/p\u003e","description":"","filename":"TableS1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/dbbea8e4b0243ca963f18bfd.xlsx"},{"id":76574202,"identity":"0650709e-acc7-49d5-b433-7ce630578e76","added_by":"auto","created_at":"2025-02-18 14:04:38","extension":"xlsx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":12265,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Table 2. The relationship of all selected variables with the outcomes.\u003c/p\u003e","description":"","filename":"TableS2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/4f321ea75254194000ee5576.xlsx"},{"id":76574204,"identity":"d0468181-0cd6-4574-83df-313650c9e77c","added_by":"auto","created_at":"2025-02-18 14:04:38","extension":"jpg","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":475702,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Figure 1. Ranks of variable importance in the RF model (A) and GBT model (B). BUN, blood urea nitrogen; PO2, partial pressure of oxygen; MBP, mean blood pressure; PH, potential of hydrogen; AKI, acute kidney injury; APTT, activated partial thromboplastin time; SPO2, pulse oxygen saturation; PCO2, partial pressure of carbon dioxide; CPD, chronic pulmonary disease; CHF, congestive heart failure; MI, myocardial infarction; RRT, renal replacement therapy.\u003c/p\u003e","description":"","filename":"FigureS1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/5188256783efe6a4945af6da.jpg"},{"id":76574205,"identity":"0dbca449-5d1b-4976-a786-e3c5bb00215b","added_by":"auto","created_at":"2025-02-18 14:04:38","extension":"jpg","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":369442,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Figure 2. Kaplan-Meier survival curve of 30-Day (A), 60-Day (B), 100-Day (C) cumulative survival rate for 4 groups. Develop a predictive model using Model 1 and divide the data evenly into four groups based on risk scores. The group 'predictors=4', which was assessed as high-risk by the model, had the highest number of deaths, while the group assessed as low-risk by the Model 1 had relatively fewer deaths.\u003c/p\u003e","description":"","filename":"FigureS2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/f0059867aeb33ec3a17bf505.jpg"},{"id":76576027,"identity":"9f2cf95e-663b-42b0-8971-db50e27ff279","added_by":"auto","created_at":"2025-02-18 14:12:38","extension":"jpg","order_by":7,"title":"","display":"","copyAsset":false,"role":"supplement","size":1537326,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Figure 3. Receiver operating characteristic (ROC) curves for predictors related to in-hospital mortality index. The solid red line indicates the ROC curve for APACHE III, and the AUC was better for APACHE III.\u003c/p\u003e","description":"","filename":"FigureS3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/9880b5e83d4363bbbc3d320d.jpg"},{"id":76574228,"identity":"a0935296-4540-48fb-b90f-39b8cbcc43f2","added_by":"auto","created_at":"2025-02-18 14:04:39","extension":"jpg","order_by":8,"title":"","display":"","copyAsset":false,"role":"supplement","size":4693957,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Figure 4. Decision curve analysis for prediction model.\u003c/p\u003e","description":"","filename":"FigureS4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6007779/v1/a9c557823eca8ef4287f7895.jpg"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development and Validation of a Predictive Nomogram for 30-Day Mortality in Sepsis Patients Coexisting with Malignant Tumors ： a Retrospective Cohort Study Using the MIMIC-IV Database","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eSepsis and septic shock affect millions of people worldwide each year, with a mortality rate of 1/4 [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. The mortality rate is even higher, exceeding 40%, for septic shock [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Important risk factors for sepsis include advanced age, low socioeconomic status, obesity, presence of chronic diseases (including chronic obstructive pulmonary disease, cancer, kidney or liver disease, and diabetes), as well as impaired immune function due to human immunodeficiency virus (HIV) infection or use of immunosuppressive drugs [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe risk of sepsis in cancer patients is tenfold higher than in non-cancer patients, and sepsis is the most common reason for their admission to the intensive care unit (ICU), with higher in-hospital mortality rates [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Several large retrospective analyses have shown that nearly 20% of patients admitted to the ICU for sepsis have underlying malignant tumors [\u003cspan additionalcitationids=\"CR7\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Malignant tumors are also independently associated with an increased risk of in-hospital mortality for sepsis patients [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Despite significant improvements in the survival rates of sepsis patients with coexisting malignant tumors in the past two decades, sepsis remains a major cause of death for cancer patients [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe predictive value of existing critical illness scoring systems for cancer patients is still unclear. Our aim is to establish a predictive model by combining multiple prognostic factors to evaluate the outcomes of these patients.\u003c/p\u003e"},{"header":"2 Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Database and study population\u003c/h2\u003e \u003cp\u003e The data employed in this research were procured from the Medical Information Mart for Intensive Care (MIMIC)-IV, version 2.2. MIMIC-IV, an accessible online clinical database, is renowned in the field of critical care research, incorporating information from in excess of 50,000 intensive care unit (ICU) admissions for adult patients at the Beth Israel Deaconess Medical Center, Boston, Massachusetts, spanning the period from 2008 to 2019. Patient informed consent was not necessary for this research as the database received approval from the Institutional Review Committee at MIT and the Beth Israel Deaconess Medical Center.\u003c/p\u003e \u003cp\u003eTo safeguard patient confidentiality, the MIMIC-IV database has undergonethorough de-identification of all personal information. In the MIMIC-IV v2.2 dataset, a total of 299712 patients generated 431231 hospitalization records, of which 73181 were ICU admissions. According to the Sepsis 3.0 criteria, 33177 patients within the ICU were diagnosed with sepsis, and among them, 5312 patients had concurrent malignant tumors. After excluding patients younger than 18 years of age, those with non-initial ICU admissions, and those who passed away within 24 hours of ICU admission, a cohort of 4196 septic patients with malignant tumors was established for this study. These patients were randomly divided into training and validation sets at a 7:3 ratio, comprising 2937 and 1259 patients, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Data collection\u003c/h2\u003e \u003cp\u003eTo enhance the simplicity of the model, we selected variables that are readily available in clinical settings. The collected data encompasses patient demographic information (age, gender, weight), complications (myocardial infarction, congestive heart failure, chronic pulmonary disease, liver disease, diabetes, renal disease, malignant cancer, metastatic solid tumor, acute kidney injury), vital signs (heart rate, respiratory rate, temperature, SpO\u003csub\u003e2\u003c/sub\u003e), severity of illness (Sequential Organ Failure Assessment, Acute Physiology and Chronic Health Evaluation III), extracorporeal life support measures (cardiovascular, renal replacement therapy, ventilation), and laboratory indicators (white blood cells, hemoglobin, platelets, anion gap, blood urea nitrogen, calcium, creatinine, glucose, sodium, potassium, international normalized ratio, prothrombin time, activated partial thromboplastin time, bilirubin, lactate, pH, PO\u003csub\u003e2\u003c/sub\u003e, PCO\u003csub\u003e2\u003c/sub\u003e). All clinical-related covariates are based on the patient's indicators on the first day of ICU admission. If the same indicator has multiple measurements on the first day of ICU admission, the worst result from that day is chosen. All of the aforementioned data were extracted from the MIMIC-IV database using Structured Query Language (SQL) and facilitated by Navicat Premium 16.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Variables Selection\u003c/h2\u003e \u003cp\u003eTo ensure that our model neither overfits nor underfits, we employed five advanced statistical strategies for variable selection in the training cohort: the best subsets regression (BSR), the least absolute shrinkage and selection operator (LASSO), and the forward stepwise regression (FSR). For BSR, we used the Bayesian Information Criterion (BIC) as the basis for variable selection, while for FSR, we utilized the Akaike Information Criterion (AIC). Both random forests and gradient boosting trees are ensemble learning methods, which enhance prediction accuracy and mitigate overfitting by building and combining multiple decision trees. In the variable selection process, we also employed the feature importance derived from Random Forest (RF) and Gradient Boosting Trees (GBT).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Statistical analysis\u003c/h2\u003e \u003cp\u003eIn the data preprocessing stage, we handled missing values to ensure the robustness of subsequent statistical analyses. Given the potential nonlinear relationships among variables, we adopted a random forest-based imputation method to deal with the missing data. We partitioned the entire dataset into two subsets by randomly splitting it in a 7:3 ratio. To ensure the accuracy of the model construction, 70% of the data was designated as the training cohort for model selection and development, while the remaining 30% served as the validation cohot. Baseline characteristics are presented separately for the original cohort, training set, and validation set. Categorical variables are described as percentages (%). Continuous variables with non-normal distributions are represented by the median and interquartile ranges (IQRs), while those with normal distributions are expressed as the mean and standard deviation (Mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD). Differences between categorical variables were assessed using the chi-square test, while differences between continuous variables were evaluated using the t-test or non-parametric tests. The performance of the model was evaluated in terms of calibration and discrimination in both the training and validation sets. The calibration capability of the prediction plot was assessed using calibration curves, while the Harrell's Consistency Index (C-index) and the Area Under the Receiver Operating Characteristic (ROC) Curve (AUC) were both employed to assess discrimination. Additionally, the clinical utility of the model for septic patients with malignant tumors was assessed using Decision Curve Analysis (DCA). Furthermore, the three models selected based on BSR, LASSO, and FSR were compared in terms of discrimination, calibration, and clinical decision-making. The variables that were ultimately determined as independent risk factors were incorporated into the final logistic regression model, and the corresponding nomogram was plotted. All statistical analyses were performed using the Python software (version 3.9, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.python.org\u003c/span\u003e\u003cspan address=\"https://www.python.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), R software (version 4.3.0, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.R-project.org\u003c/span\u003e\u003cspan address=\"http://www.R-project.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), and Free Statistics software version 1.9. This study was conceptualized and conducted in alignment with the guidelines provided by the TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis) statement.\u003c/p\u003e \u003c/div\u003e"},{"header":"3 Results","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Clinical Characteristics\u003c/h2\u003e \u003cp\u003eThe baseline clinical and demographic characteristics of the training and validation cohorts can be found in Table\u0026nbsp;1. The primary cohort comprised 4196 sepsis patients with concurrent malignant tumors, of which 2937 were assigned to the training set and 1,259 to the validation set. Within this primary cohort, the general age distribution was approximately 66.6\u0026thinsp;\u0026plusmn;\u0026thinsp;13 years, with males accounting for 61.7% of the population. A review of the baseline characteristics revealed commendable consistency in demographic characteristics, comorbidities, vital signs, severity of illness scores, extracorporeal life support measures, and laboratory results between the training and validation sets. Among the comparative analyses of the variables between the two groups, only the history of renal disease presented a significant difference (p\u0026thinsp;=\u0026thinsp;0.035), whereas the p-values for the remaining variables exceeded 0.05. Before imputation using the random forest method, various baseline variables exhibited certain levels of missingness. Specific proportions and counts of missing data for each variable can be found in Supplementary Table\u0026nbsp;1.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Variable Selection via BSR\u003c/h2\u003e \u003cp\u003eThe BSR method excels in variable selection due to its comprehensive computation of all possible variable combinations. The ultimate selection criterion is based on the minimum Bayesian Information Criterion (BIC). As illustrated in Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB, the report encompasses an evaluation of all 38 parameters, resulting in a BIC value of -662. The final number of selected variables is 8, which is based on the minimum BIC value in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA. The variables in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB with a BIC value less than \u0026minus;\u0026thinsp;660 represent the coefficients that constitute the final variable combination determined through BSR. This combination includes 8 variables from the training cohort: weight, acute kidney injury, heart rate, temperature, Apache II score, BUN, creatinine, and glucose. The relationship of all selected variables with the outcomes is detailed in Table\u0026nbsp;2 and Supplementary Table\u0026nbsp;2. Subsequently, Model 1 was established based on the variables screened by BSR.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Variable Selection via LASSO\u003c/h2\u003e \u003cp\u003eTo identify the pivotal variables within our predictive model, we further employed the LASSO logistic regression method. LASSO achieves variable selection and complexity adjustment by penalizing coefficients, enhancing both predictive accuracy and interpretability of the model. This approach is particularly suited for addressing multicollinearity and high-dimensional data. We initiated the process by standardizing all predictive variables, and through cross-validation, we ascertained two critical regularization parameters (λ values): lambda.min and lambda.1se. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC illustrates the coefficient paths of individual variables, demonstrating which variables enter the model first and which ones persist as λ values change. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eD portrays the relationship between model mean squared error and log(λ), with 23 variables included in the model at lambda.1se\u0026thinsp;=\u0026thinsp;0.01: Age, Gender, Weight, CPD, Diabetes, AKI, Heart Rate, Temperature, SpO\u003csub\u003e2\u003c/sub\u003e, APACHE III, Cardiovascular, RRT, Ventilation, Anion gap, Bun, Calcium, Creatinine, Glucose, Sodium, Bilirubin, pH, PO\u003csub\u003e2\u003c/sub\u003e, PCO\u003csub\u003e2\u003c/sub\u003e. The relationships of all selected variables with the outcomes are detailed in Table\u0026nbsp;2 and Supplementary Table\u0026nbsp;2. Subsequently, Model 2 was established based on the variables screened by LASSO.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Variable Selection via FSR\u003c/h2\u003e \u003cp\u003eStepwise regression automates the selection of independent variables most relevant to the dependent variable, thereby enhancing the predictive performance of the model. By iteratively adding or removing independent variables, it aids in constructing a more concise and high-performance model. First, we executed a forward stepwise regression based on the Akaike Information Criterion (AIC), commencing with zero variables and assessing the impact of introducing a new variable into the model at each step. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eE displays the AIC values at each step of the stepwise regression process. AIC values exhibited a decreasing trend with the addition of each new predictive variable, reflecting the incremental model fit. In the 10th step, following the inclusion of the 10th variable 'Creatinine,' the model's AIC reached its minimum at 2983.10. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eF illustrates the variation in P-values for each variable throughout the stepwise regression process. The red dashed line indicates the significance threshold of 0.05. Notably, each variable in the final model, at the step of its inclusion, displayed P-values significantly below the 0.05 threshold, implying their significant predictive power within the model. The stepwise regression analysis identified the following 10 variables as predictive factors for the final logistic regression model: Weight, AKI, Heart Rate, Temperature, SpO\u003csub\u003e2\u003c/sub\u003e, APACHE III, Ventilation, Bun, Creatinine, Glucose. These variables consistently exhibited significant p-values in their respective steps and substantially reduced the AIC upon inclusion in the model, thereby confirming their inclusion in Model 3 (Table\u0026nbsp;2 and Supplementary Table\u0026nbsp;2).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Variable Selection via RF and GBT\u003c/h2\u003e \u003cp\u003eRandom Forest and Gradient Boosting Trees are both ensemble learning methods that improve prediction accuracy and control overfitting by constructing and combining multiple decision trees. We utilized the Random Forest Classifier and Gradient Boosting Classifier classes from the Scikit-learn library to train the Random Forest and Gradient Boosting Tree models, respectively. Feature importance is computed based on the frequency and depth of splits within the trees, offering a quantitative measure of each feature's contribution to the model's predictive capabilities. From Supplementary Fig.\u0026nbsp;1A and Supplementary Fig.\u0026nbsp;1B, it is evident that the top three most important variables are APACHE III, Temperature, and Heart Rate. These variables were consistently included in all three aforementioned models.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.6 Comparison of Model Performance and Establishment of the Final Model\u003c/h2\u003e \u003cp\u003eIn the training cohort, each significant variable underwent an initial assessment through univariate logistic regression analysis (Supplementary Table\u0026nbsp;2). Subsequently, multivariable logistic regression analysis revealed that several variables, comprising Age, Gender, CPD, Diabetes, AKI, Heart Rate, APACHE III, Cardiovascular, Anion gap, BUN, Calcium, Creatinine, Bilirubin, pH, PCO\u003csub\u003e2\u003c/sub\u003e, emerged as independent risk factors, whereas Weight, Temperature, SpO\u003csub\u003e2\u003c/sub\u003e, RRT, Ventilation, Glucose, Sodium, PO\u003csub\u003e2\u003c/sub\u003e were identified as independent protective factors (Table\u0026nbsp;2).\u003c/p\u003e \u003cp\u003eAcross the three regression models examined, Model 2 consistently exhibited the highest C-index in both the training and validation cohorts. However, the inclusion of 23 variables in Model 2 resulted in a more cumbersome model. Model 1, on the other hand, displayed the lowest C-index in both cohorts, but it also comprised the fewest variables, totaling eight. Model 3's C-index and number of variables placed it between Models 1 and 2 (Table\u0026nbsp;2). In the training cohort, Model 2 achieved the highest area under the curve (AUC), followed by Model 3, and then Model 1, with specific values being AUC1: 0.785 (0.768, 0.803), AUC2: 0.802 (0.785, 0.819), and AUC3: 0.790 (0.773, 0.808). The comparison among the three yielded a p-value of less than 0.01 (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). Similarly, in the validation cohort, Model 2 had the highest AUC, with Model 3 and Model 1 trailing, respectively. The exact values were AUC1: 0.789 (0.762, 0.815), AUC2: 0.793 (0.767, 0.819), and AUC3: 0.790 (0.764, 0.816), with a comparative p-value of less than 0.01 (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWe subsequently compared the calibration of the three models in both the training and validation cohorts (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC-D). In Figs.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD, the blue dashed line represents Model 1, the green solid line denotes Model 2, and the red dashed line signifies Model 3. The ideal calibration curve coincides with the 45\u0026deg; black dashed line in the figures, which represents a scenario where the model's predicted probabilities perfectly match the observed event rates. In practice, the closer a model's curve aligns with the ideal line, the more its predicted probabilities correspond to the actual event occurrences. As evident from the figures, all three models demonstrate high consistency in the training cohort. Although the calibration curves deviate somewhat from the ideal 45\u0026deg; line in the validation cohort, the variations among the three models are minimal. In the training cohort's Hosmer-Lemeshow (H-L) test, Model 1 yielded an X1-squared value of 11.84 with an H-L P-value of 0.159. For Model 2, the X2-squared value was 5.61 with an H-L P-value of 0.691, and for Model 3, the X3-squared value was 5.44 with an H-L P-value of 0.709. In the validation cohort's H-L test, Model 1 had an X1-squared value of 13.18 with an H-L P-value of 0.106, Model 2 presented an X2-squared value of 11.44 with an H-L P-value of 0.178, and Model 3 recorded an X3-squared value of 9.32 with an H-L P-value of 0.316. All P-values exceeded 0.05, suggesting that the observed event rates for the three models closely aligned with their predicted probabilities, indicating satisfactory calibration (Table\u0026nbsp;2).\u003c/p\u003e \u003cp\u003eAs depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE-F, decision curve analysis demonstrates that for the training cohort with a threshold probability ranging from 8\u0026ndash;82%, and the validation cohort ranging from 7\u0026ndash;86%, the use of the line chart to predict the 30-day mortality rate offers a net benefit surpassing both the \"treat all\" and the \"treat none\" strategies. This underscores the clinical utility of the nomogram established by Model 1. The clinical relevance of the nomogram derived from Model 2 is evident in the training cohort with a threshold probability between 4% and 88%, and in the validation cohort between 4% and 98%. For Model 3, its nomogram displays clinical applicability within the training cohort for threshold probabilities of 5\u0026ndash;88% and in the validation cohort from 5\u0026ndash;92%. The research findings suggest that nomograms developed from all three models can benefit sepsis patients with concomitant malignant tumors.\u003c/p\u003e \u003cp\u003eModel 2 exhibited the best performance in terms of discrimination, calibration, and clinical utility. However, its inclusion of 23 variables significantly hinders its future clinical application, as complex models carry a higher risk of overfitting. Model 1, incorporating only 8 variables, achieved model performance comparable to both Model 3 and Model 2 (Table\u0026nbsp;2). A more concise model is favorable for future clinical use, which is why we constructed the line chart based on Model 2 (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Additionally, the variables selected in Model 1 had minimal missing data in the database. Only the body temperature had a missing rate of 1.35%, while the missing rates for the other variables were all below 0.67% (Supplementary Table\u0026nbsp;1). After developing the prediction model with Model 1 and categorizing the data based on risk scores, we observed that the high-risk group consistently exhibited the highest mortality risk at 30 days, 60 days, and 100 days (Supplementary Fig.\u0026nbsp;2). Concurrently, we presented a consolidated view of the relationships between the variables in Model 1 and the outcomes (Supplementary Fig.\u0026nbsp;3). Through decision curve analysis, we identified that at most threshold probabilities accepted by patients or clinicians, our prediction model offers a superior net benefit over other strategies. This suggests that, within these thresholds, employing our model for clinical decision-making can yield significant benefits for sepsis patients with concurrent malignant tumors (Supplementary Fig.\u0026nbsp;4).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eCurrently, there are almost no models available for predicting the risk of death within 30 days in septic patients with concomitant malignant tumors. In this study, we used multivariable logistic regression analysis to identify independent risk factors for death in septic patients with concomitant malignant tumors in MIMIC-IV and developed a simplified model consisting of 8 predictive factors: weight, AKI (acute kidney injury), heart rate, temperature, APACHE III (Acute Physiology and Chronic Health Evaluation III), blood urea nitrogen, creatinine, and blood glucose. These variables are easily obtainable clinical indicators. The model showed good predictive ability and discriminatory power with favorable results in both the training and validation datasets. However, it is important to note that our study primarily analyzed data from MIMIC-IV and is only applicable to ICU patients, and the potential for selection bias in the data cannot be ignored.\u003c/p\u003e \u003cp\u003eThe APACHE (Acute Physiology and Chronic Health Evaluation) scoring system is a tool used to assess and predict the severity and prognosis of patients. APACHE II and APACHE III are two commonly used versions. The APACHE II scoring system is mainly based on physiological parameters, age, and disease type, among other factors. It includes 12 physiological parameters such as heart rate, temperature, blood pressure, as well as disease type and age. The APACHE III scoring system builds upon APACHE II by including additional physiological parameters and disease diagnostic categories, including indicators related to hematology, liver function, and the nervous system [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Due to the consideration of a wider range of physiological parameters and disease categories, APACHE III may be more accurate in predicting patient prognosis compared to APACHE II. Several studies have confirmed that APACHE II score is an independent predictive factor for in-hospital and ICU survival rates in septic patients with malignant tumors [\u003cspan additionalcitationids=\"CR14\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. However, there is limited research on the relationship between APACHE III score and sepsis in malignant tumor patients. Our study found that APACHE III score is an independent risk factor.\u003c/p\u003e \u003cp\u003eSepsis is a systemic inflammatory response caused by the introduction of bacteria or other infective microorganisms into the body. If sepsis is not treated promptly or inappropriately, the inflammatory response may lead to hemodynamic instability, tissue ischemia, cell damage, and other factors that can negatively affect the kidneys [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Increasing evidence suggests a higher risk of infection or sepsis following acute kidney injury (AKI), and some experts suggest considering AKI as an early sign of sepsis [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. For septic patients with concomitant malignant tumors, it is important to prevent infection as much as possible, enhance immunity, and identify and treat complications early. If AKI occurs, measures such as renal replacement therapy should be promptly taken to maintain the patient's life, and monitoring and management of treatment should be intensified.\u003c/p\u003e \u003cp\u003eMetabolic disturbances in glucose metabolism are common in septic patients. A large cohort study showed that hyperglycemia is associated with increased ICU mortality in septic patients, and the impact of blood glucose on mortality increases with the severity of sepsis [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Another multicenter prospective cohort study found that the mortality rate in septic patients with hypoglycemia is 2.5 times higher than that in septic patients with normal blood glucose levels [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. The NICE-SUGAR study also reported a strong association between hypoglycemia and mortality [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. It is evident that the mortality rate is higher in septic patients with hyperglycemia or hypoglycemia compared to patients with normal blood glucose levels, highlighting the importance of maintaining stable blood glucose concentrations.\u003c/p\u003e"},{"header":"5 Conclusion","content":"\u003cp\u003eThe predictive model includes 8 predictive factors, which are weight, AKI, heart rate, temperature, APACHE III, BUN, creatinine, and glucose. The model effectively predicts the risk of death within 30 days for sepsis with malignant tumors. Internal validation has demonstrated its effectiveness. The research findings aid clinical physicians in making treatment decisions for these patients, thus reducing the risk of sepsis and death in malignancy patients.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eConflict of Interest\u003c/h2\u003e\n\u003cp\u003eThe authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.\u003c/p\u003e\n\u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e\n\u003cp\u003eThe data used in this study were acquired from an open source and do not require approval by any ethical committee.\u003c/p\u003e\n\u003ch2\u003eCompeting interests\u003c/h2\u003e\n\u003cp\u003eThe authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interests.\u003c/p\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eNo funding.\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eLin Qian and Weiting Sun: Conceptualization, Writing \u0026ndash; original draft, Formal analysis, Writing \u0026ndash; review \u0026amp; editing, Validation, Visualization, Investigation. Peng Ding: Investigation. Song Zhang: Conceptualization, Writing \u0026ndash; review \u0026amp; editing, Project administration. Kun-lan Long: Supervision.\u003c/p\u003e\n\u003ch2\u003eData Availability Statement\u003c/h2\u003e\n\u003cp\u003eThe original contributions presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding authors.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eRhodes A, Evans LE, Alhazzani W, Levy MM, Antonelli M, Ferrer R, et al. Surviving Sepsis Campaign: International Guidelines for Management of Sepsis and Septic Shock: 2016. \u003cem\u003eIntensive Care Med\u003c/em\u003e (2017) 43(3):304-377. doi: 10.1007/s00134-017-4683-6. \u003c/li\u003e\n\u003cli\u003eSinger M, Deutschman CS, Seymour CW, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA. 2016, 315(8):801-10. doi: 10.1001/jama.2016.0287. \u003c/li\u003e\n\u003cli\u003eMayr FB, Yende S, Angus DC. Epidemiology of severe sepsis. Virulence. 2014,5(1):4-11. doi: 10.4161/viru.27372.\u003c/li\u003e\n\u003cli\u003eNathan N, Sculier JP, Ameye L, et al. Sepsis and Septic Shock Definitions in Patients With Cancer Admitted in ICU. J Intensive Care Med. 2021,36(3):255-261. doi: 10.1177/0885066619894933. \u003c/li\u003e\n\u003cli\u003eMirouse A, Vigneron C, Llitjos JF, et al. Sepsis and Cancer: An Interplay of Friends and Foes. Am J Respir Crit Care Med. 2020,202(12):1625-1635. doi: 10.1164/rccm.202004-1116TR. \u003c/li\u003e\n\u003cli\u003eAnnane D, Aegerter P, Jars-Guincestre MC, et al. Current epidemiology of septic shock: the CUB-R\u0026eacute;a Network. Am J Respir Crit Care Med. 2003,168(2):165-72. doi: 10.1164/rccm.2201087. \u003c/li\u003e\n\u003cli\u003eVincent JL, Rello J, Marshall J, et al. International study of the prevalence and outcomes of infection in intensive care units. JAMA. 2009,302(21):2323-9. doi: 10.1001/jama.2009.1754. \u003c/li\u003e\n\u003cli\u003eHensley MK, Donnelly JP, Carlton EF, et al. Epidemiology and Outcomes of Cancer-Related Versus Non-Cancer-Related Sepsis Hospitalizations. Crit Care Med. 2019,47(10):1310-1316. doi: 10.1097/CCM.0000000000003896. \u003c/li\u003e\n\u003cli\u003eWilliams JC, Ford ML, Coopersmith CM. Cancer and sepsis. Clin Sci (Lond). 2023,137(11):881-893. doi: 10.1042/CS20220713. \u003c/li\u003e\n\u003cli\u003eRosolem MM, Rabello LS, Lisboa T, et al. Critically ill patients with cancer and sepsis: clinical course and prognostic factors. J Crit Care. 2012,27(3):301-7. doi: 10.1016/j.jcrc.2011.06.014.\u003c/li\u003e\n\u003cli\u003eKnaus WA, Draper EA, Wagner DP, et al. APACHE II: a severity of disease classification system. Crit Care Med. 1985,13(10):818-29. \u003c/li\u003e\n\u003cli\u003eKnaus WA, Wagner DP, Draper EA, et al. The APACHE III prognostic system. Risk prediction of hospital mortality for critically ill hospitalized adults. Chest. 1991,100(6):1619-36. doi: 10.1378/chest.100.6.1619.\u003c/li\u003e\n\u003cli\u003eKopterides P, Liberopoulos P, Ilias I, et al. General prognostic scores in outcome prediction for cancer patients admitted to the intensive care unit. Am J Crit Care. 2011,20(1):56-66. doi: 10.4037/ajcc2011763. \u003c/li\u003e\n\u003cli\u003eNamendys-Silva SA, Texcocano-Becerra J, Herrera-G\u0026oacute;mez A. Prognostic factors in critically ill patients with solid tumours admitted to an oncological intensive care unit. Anaesth Intensive Care. 2010,38(2):317-24. doi: 10.1177/0310057X1003800214. \u003c/li\u003e\n\u003cli\u003eAygencel G, Turkoglu M, Turkoz Sucak G, et al. Prognostic factors in critically ill cancer patients admitted to the intensive care unit. J Crit Care. 2014,29(4):618-26. doi: 10.1016/j.jcrc.2014.01.014. \u003c/li\u003e\n\u003cli\u003eChang YM, Chou YT, Kan WC, et al. Sepsis and Acute Kidney Injury: A Review Focusing on the Bidirectional Interplay. Int J Mol Sci. 2022,23(16):9159. doi: 10.3390/ijms23169159.\u003c/li\u003e\n\u003cli\u003eLai TS, Wang CY, Pan SC, et al. Risk of developing severe sepsis after acute kidney injury: a population-based cohort study. Crit Care. 2013,17(5):R231. doi: 10.1186/cc13054. \u003c/li\u003e\n\u003cli\u003eManrique-Caballero CL, Del Rio-Pertuz G, Gomez H. Sepsis-Associated Acute Kidney Injury. Crit Care Clin. 2021,37(2):279-301. doi: 10.1016/j.ccc.2020.11.010. \u003c/li\u003e\n\u003cli\u003eLu Z, Tao G, Sun X, et al. Association of Blood Glucose Level and Glycemic Variability With Mortality in Sepsis Patients During ICU Hospitalization. Front Public Health. 2022,10:857368. doi: 10.3389/fpubh.2022.857368. \u003c/li\u003e\n\u003cli\u003eWang J, Zhu CK, Yu JQ, Tet al. Hypoglycemia and mortality in sepsis patients: A systematic review and meta-analysis. Heart Lung. 2021,50(6):933-940. doi: 10.1016/j.hrtlng.2021.07.017. \u003c/li\u003e\n\u003cli\u003eKushimoto S, Abe T, Ogura H, et al. Impact of blood glucose abnormalities on outcomes and disease severity in patients with severe sepsis: An analysis from a multicenter, prospective survey of severe sepsis. PLoS One. 2020,15(3):e0229919. doi: 10.1371/journal.pone.0229919. \u003c/li\u003e\n\u003cli\u003eNICE-SUGAR Study Investigators; Finfer S, Liu B, Chittock DR, et al. Hypoglycemia and risk of death in critically ill patients. N Engl J Med. 2012,367(12):1108-18. doi: 10.1056/NEJMoa1204942.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1 and 2 are available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"malignant tumors, sepsis, nomogram, 30-Day mortality, MIMIC-IV","lastPublishedDoi":"10.21203/rs.3.rs-6007779/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6007779/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eSepsis is the main cause of death for cancer patients, and our study aims to evaluate risk factors and develop a model to predict the 30-Day mortality in sepsis patients coexisting with malignant tumors.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003e We obtained 4196 sepsis patients coexisting with malignant tumors from the MIMIC-IV database and randomly split them into a training set (2937 patients) and a validation set (1259 patients) at a ratio of 7:3. A multivariable logistic regression model was used to identify independent risk factors for predicting mortality, and the model's performance was evaluated.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eMultivariable logistic regression analysis showed that age, gender, CPD, diabetes, AKI, heart rate, APACHE III, cardiovascular system, anion gap, BUN, calcium, creatinine, bilirubin, pH, and PCO2 were independent risk factors. The nomogram achieved optimal performance in discrimination, calibration, and clinical application.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThe nomogram effectively predicts the 30-Day mortality in sepsis patients coexisting with malignant tumors, and internal validation confirms its effectiveness. The study results can help clinical doctors make decisions on the treatment of these patients, thus reducing the risk of sepsis and death for cancer patients.\u003c/p\u003e","manuscriptTitle":"Development and Validation of a Predictive Nomogram for 30-Day Mortality in Sepsis Patients Coexisting with Malignant Tumors ： a Retrospective Cohort Study Using the MIMIC-IV Database","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-02-18 14:04:33","doi":"10.21203/rs.3.rs-6007779/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c14955a3-a1c8-48ec-ad79-c77003202796","owner":[],"postedDate":"February 18th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-02-22T12:08:30+00:00","versionOfRecord":[],"versionCreatedAt":"2025-02-18 14:04:33","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6007779","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6007779","identity":"rs-6007779","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}