In this large-scale retrospective analysis of the MIMIC IV clinical database, we employed both univariate and multivariate COX regression analyses to identify risk factors associated with the 28-day mortality in patients with sepsis and comorbid diabetes mellitus (DM). Ultimately, 12 clinical variables were confirmed and incorporated into the predictive model, including Norepinephrine maximum, Temperature, Antibiotic, Heart Rate, SAPS II, Age, BMI, Bilirubin total, Stroke, Troponin, Metastatic solid tumor, APTT. Subsequently, a predictive model in the form of a Nomogram was developed, representing the first instance of a visual model for predicting the 28-day mortality in sepsis patients with diabetes. Our study revealed a 28-day mortality rate of 24.1% in patients with sepsis and comorbid diabetes, which is comparable to previously reported rates[5] . Sepsis in combination with diabetes was more common in males, and common sources of infection included pulmonary, skin and soft tissue, urinary, abdominal, and intestinal infections.
Sepsis, characterized by life-threatening organ dysfunction due to a dysregulated host response to infection, has seen an increased occurrence in diabetes patients as the prevalence of diabetes mellitus (DM) rises globally. Approximately 20-35% of sepsis patients are reported to have diabetes[7,19,20] . However, there is ongoing debate about the prognostic impact of sepsis in diabetes, with three main perspectives: firstly, some studies suggest an increased mortality rate in sepsis patients with comorbid diabetes[21-24] ; secondly, other research indicates no association with mortality[5,6,25-29] ; thirdly, some studies have found a protective effect of diabetes during sepsis[30-32]. From a clinical perspective, diabetes can lead to immune dysfunction and metabolic disturbances, impairing the function of neutrophils, reducing antibacterial activity, and affecting the regulation of cytokines, resulting in compromised host defense, endothelial cell damage, mitochondrial damage, and inflammatory activation, leading to catastrophic consequences such as acute kidney injury, myocardial infarction, and increased risks of vascular narrowing, thereby lowering sepsis survival rates[33-35] . However, physiologically, glucose plays a crucial role in the metabolic requirements and maintenance of immune cell function, as well as the synthesis of immunomodulators[36,37]. Due to sustained high glucose concentrations, diabetes patients develop tolerance to hyperglycemia, converting detrimental high glucose levels into an energy storage reservoir[38]. Therefore, high or low blood glucose concentrations seem to have no significant impact on mortality. Regarding the protective effect of diabetes on sepsis patients, current research attributes it to the anti-inflammatory effects of exogenous insulin injection, preventing acute lung injury[6], adapting to previous oxidative stress, and improving the nutrition or substrate of obese diabetic patients[39]. Our study results indicate no statistical difference in blood glucose levels between the survival and death groups, suggesting that both high and low blood glucose are unrelated to the mortality of sepsis patients with comorbid diabetes. These findings contradict the widespread understanding of the destructive consequences of diabetes and hyperglycemia, aligning more with the first perspective mentioned earlier[5,6,28,29]
. However, some studies have shown that in non-diabetic patients, both high and low blood glucose significantly increase the in-hospital mortality of sepsis patients[20,40-43]。revealing a U-shaped relationship between blood glucose and mortality[44].
As is well known, SOFA score and SAPS II score have been proven to be useful tools for predicting mortality in sepsis patients[18,45] . However, there is limited evidence on whether they are suitable for assessing the prognostic value of sepsis in patients with diabetes, and they may not be sufficient on their own to predict the prognosis of such patients. Therefore, our predictive model includes not only the SAPS II score but also other indicators assessing predictive factors.
Firstly, we conducted multivariate COX regression analysis on patients with sepsis and comorbid diabetes from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. The results showed that variables closely related to patient prognosis included Norepinephrine maximum dose, SAPS II, Stroke, Bilirubin total, Age, Metastatic solid tumor, Temperature, BMI, SOFA, Liver disease, Troponin, Antibiotic, Malignant cancer, APTT, CRRT, Breath rate, Base excess. Among these variables, the hazard ratio (HR) values for Temperature, BMI, Antibiotic, and CRRT were less than 1, indicating a positive correlation with in-hospital survival in patients with sepsis and comorbid diabetes. Secondly, we constructed the current predictive model using data extracted from the MIMIC-IV database and compared the predictive model with SOFA and SAPS II scores. The results showed that our predictive model had higher predictive performance than SAPS II and SOFA scores for predicting the 28-day mortality of hospitalized patients. To verify the clinical accuracy of the predictive nomogram, we also conducted calibration curve analysis on the training set and validation set, showing good consistency between observed values and ideal reference values. Additionally, to validate the clinical effectiveness of the predictive model, we performed decision curve analysis on patients with sepsis and comorbid diabetes, revealing that interventions based on the nomogram could provide better outcomes than SOFA and SAPS II scores when the probability threshold was between 0.1 and 0.6, resulting in greater net benefits.
In our predictive model, SAPS II, Troponin, Bilirubin total, and Norepinephrine maximum dose carried the highest weights, indicating that they are the most important predictive factors crucial for predicting the 28-day mortality in patients with sepsis and comorbid diabetes. Undoubtedly, SAPS II score is widely used for mortality risk analysis and prognosis assessment in sepsis patients[46]. Compared to the SOFA score, SAPS II score enhances the discriminatory, calibration, and predictive capabilities for the mortality risk of sepsis patients and has been recommended by Sepsis 3.0 for the identification and prognosis of sepsis patients[47] . Therefore, SAPS II score not only holds significant predictive value for the prognosis of sepsis patients but also serves as an important predictive factor in our study's predictive model.
Troponin is the most important marker of myocardial injury and is observed in severely septic patients[48] . It is considered one of the most serious complications and primary causes of death in sepsis, directly impacting the prognosis of septic patients. Studies have shown that the incidence of sepsis with concomitant myocardial injury is 20%, associated with a significantly higher short-term mortality rate. The mortality rate is as high as 70% in septic patients with accompanying heart dysfunction, while those without heart dysfunction have a mortality rate of only 20%[49,50]. In septic patients, there is a significant correlation between elevated troponin levels and increased mortality[51].
Our study consistently indicates that troponin is an independent risk factor and an effective predictor for the 28-day mortality in patients with sepsis and comorbid diabetes. Both univariate and multivariate COX regression analyses show a positive correlation between troponin levels and the 28-day mortality in these patients. Regarding the mechanisms of sepsis-induced myocardial injury, current research primarily suggests that elevated troponin levels and cardiac dysfunction result from mechanisms such as myocardial suppression, sympathetic nervous system activation, mitochondrial damage, and calcium homeostasis imbalance[52-56].
Total bilirubin is another crucial predictive factor. Bilirubin is the final product of heme metabolism in mammals and is typically considered a lipid-soluble metabolite that needs to be excreted. It serves as a common indicator reflecting liver metabolism. Severe septic patients often exhibit elevated bilirubin levels, and increased bilirubin levels can induce inflammation and cell apoptosis[57] . There is growing evidence suggesting that high concentrations of bilirubin can induce inflammation, cell apoptosis, and oxidative stress[58,59]. Studies have shown that elevated bilirubin levels upon admission are associated with an increased risk of mortality[57,60,61]. Compared to patients with bilirubin levels of 1 mg/dL, patients with levels between 1.1-2 mg/dL and >2 mg/dL had adjusted mortality rates 3.85 times and 9.85 times higher, respectively. Higher bilirubin levels above 1 mg/dL are associated with a higher proportion of individuals developing septic shock and longer hospital stays[62]. Our study results are consistent with these findings, showing a positive correlation between bilirubin levels and mortality[57] . The current understanding is that mechanisms leading to elevated bilirubin in sepsis involve hemolysis, liver dysfunction, and bile stasis, possibly mediated by cytokines and endotoxins regulating a reduction in bile transport in the basolateral membrane and tubules[63].
Another important predictive factor is the maximum dose of vasoactive drugs. Vasopressors are recommended for the treatment of septic shock because of their rapid and significant impact on blood pressure. Different doses and timing of vasopressor use may lead to different clinical outcomes. Current research and guidelines recommend norepinephrine as the first-line treatment for severe sepsis or septic shock[16,64,65] . Our study indicates that patients receiving vasopressor treatment generally have a higher 28-day mortality rate, and the higher the dose used, the higher the risk of death. This is consistent with current related research[66-69]. However, further research is needed to determine the appropriate dose and optimal timing for vasopressor use, especially in elderly critically ill patients[67,69,70]。
Our study has several limitations. Firstly, we were unable to obtain information on the course and severity of diabetes in patients, preventing us from assessing the impact of these factors on prognosis. Additionally, we couldn't determine whether diabetes was newly diagnosed. Secondly, our study is retrospective, and the inclusion of each variable introduces potential confounding factors that could influence the results. Thirdly, during the importation of data from the MIMIC-IV database, we observed missing variables, with some missing data exceeding 50%, which were directly excluded. These missing variables might have an impact on our study. Lastly, we conducted only internal validation using this database, and future research should involve external validation based on our own data to further confirm the robustness and performance of the nomogram.