In this study, iron metabolism markers are analyzed to see if they are effective in forecasting SAE. In the present study, the occurrence rate of SAE was found to be 69.3%. To identify independent risk factors associated with SAE, Our LASSO regression analysis identified significant predictors as mean arterial pressure, respiratory rate, serum iron, elective surgery, microorganism, SASPIII and Oasis score. These findings were subsequently utilized to develop a diagnostic prediction nomogram for SAE. The performance of our nomogram model was evaluated using various metrics including AUC, calibration curve, Hosmer-Lemeshow test, IDI, NRI, and DCA. The results of these assessments demonstrated the robustness, discriminative ability, and clinical utility of our nomogram model. Among the biochemical parameters, serum iron is the only biochemical marker. (OR = 0.997, 95% CI (0.993-1.000)). As a biomarker for SAE, serum iron may be quite effective.
Recent research has revealed a significant association between sepsis and iron metabolic disorders. In normal physiological circumstances, the regulation of iron homeostasis is meticulously controlled by hepcidin. However, in sepsis, intracellular iron transport and absorption increase, and iron output is decreased. This adaptive response by the body serves as a conservative measure to combat invading pathogens. While intracellular iron overload has been demonstrated to possess a certain degree of protective efficacy, as the disease progresses, it may cause oxidative damage and cellular demise, as typified by pyroptosis and ferroptosis. [16, 17]. Research on iron metabolism and SAE is limited, however. This study revealed a significant association between serum iron concentration and the occurrence of SAE. The SAE group exhibited significantly lower serum iron levels compared to the non-SAE group (P < 0.05), and a decrease in serum iron was found to be correlated with an increased incidence of SAE (OR = 0.997, 95%CI = 0.993-1.000). Zhang et al[18] also found that for patients with aneurysmal subarachnoid hemorrhage(aSAH), lower serum iron levels at admission were significantly correlated with WFNS grading and Fisher grading, and could predict postoperative delayed cerebral ischemia (DCI) and poor 90-day prognosis in aSAH patients. Wei et al[19] found that CLP mice exhibited impaired integrity of the blood-brain barrier, resulting in heightened permeability. Furthermore, the levels of reactive oxygen species (ROS) and iron ions in the cerebral cortex homogenate were significantly elevated compared to the control group. In addition, HE staining revealed changes in the morphology and quantity of neurons in the model mice. This can be attributed to the deposition of iron in microglia, leading to their activation and subsequent increase in blood-brain barrier permeability, which aligns with the pathophysiological characteristic of SAE.
Furthermore, our study has also identified a significant association between the MAP and RR in vital signs and the occurrence of SAE. MAP plays a crucial role in regulating blood flow in organs and maintaining the homeostasis mechanism of systemic hemodynamics, particularly through pressure receptors[20]. Additionally, cerebral blood flow (CBF) possesses the ability to autoregulate, thereby maintaining a stable blood flow within a specific range of blood pressure. However, sepsis patients often have impaired cerebral vascular autoregulation[21]. Pierrakos et al[22] found that the increase in PI is related to the occurrence of SAE in 40 sepsis patients through TCD examination, mainly due to cerebral perfusion disorder. If the MAP surpasses the range of cerebral vascular autoregulation, it can lead to elevated intracranial pressure and a reduction in cerebral perfusion pressure, aligning with our research findings.
Lu et al[23] devised a machine learning model to forecast the likelihood of SAE, with the XGBoost classification model highlighting the respiratory rate as a significant risk factor. The SHAP value analysis indicates an inverse relationship between the respiratory rate and the probability of SAE. Our predictive model further substantiates the association between the respiratory rate and the incidence of SAE, as a diminished respiratory rate may signify compromised temperature regulation, oxygenation capacity, and circulatory function, as well as heightened susceptibility to severe infection, all of which contribute to an augmented risk of SAE.
Delirium is the most common postoperative complication. According to Leslie et al[24],in the United States, 3 million elderly people are hospitalized for elective surgery each year, with delirium occurring in 25% of these cases. Elective surgery is identified as a risk factor for SAE in our predictive model, primarily due to the potential induction of delirium by specific drugs, notably benzodiazepines, administered during and after surgery. Furthermore, certain analgesics may contribute to an increased risk of delirium[25].
Our study revealed that patients infected with Gram-negative bacteria have a higher susceptibility to SAE. Li et al 's research[26] demonstrated that the gut microbiota can influence SAE through the vagus nerve, as pathogenic microbes can impact nerve function by releasing specific neurotransmitters. Notably, Escherichia coli predominantly secretes norepinephrine and serotonin, whereas Candida and Streptococcus primarily secrete serotonin.
Research has indicated that the SOFA score exhibits favorable diagnostic and prognostic efficacy in sepsis patients[27]. Nevertheless, the findings from the ROC curve analysis conducted in this study demonstrate that, the predictive capacity of the SOFA score for SAE is comparatively inferior to that of the OASIS and the SAPS III. The OASIS comprises ten readily accessible parameters[28]. Chen et al[29] conducted a retrospective study involving 10,305 patients and found that OASIS could potentially serve as an initial predictive indicator of clinical outcomes in sepsis patients. The SAPS III score, which was introduced in 2005 is the largest multinational prospective study conducted thus far. Zhu et al[30] examined a cohort of 12,691 sepsis patients and determined that the SAPS III exhibited superior discriminatory ability in predicting 28-day mortality compared to the SOFA. Nevertheless, there is a paucity of literature regarding the predictive capabilities of the OASIS score or the SAPS III score in relation to SAE. The findings of our study demonstrate that the results obtained from the LASSO regression analysis, as well as the subsequent multivariate logistics regression analysis, consistently indicate that the OASIS and SAPS III scores possess significant predictive capabilities for SAE (OR = 1.076 and OR = 1.034), suggesting that an increase in these scores corresponds to an increased likelihood of SAE occurrence.
Strengths and Limitations
To the best of our knowledge, this study serves as the first clinical examination of the prediction of SAE using serum iron as a biomarker. The utilization of the publicly accessible MIMIC-IV database, which contains a comprehensive set of data related to critically ill patients, has provided us with substantial evidence to substantiate our discoveries. The experimental parameters established within our SAE prediction model can be easily obtained in routine clinical practice, making the model easily understandable. As a result, this research bears significance in facilitating the timely detection of SAE in clinical environments and establishing a theoretical basis for subsequent explorations into biomarkers associated with SAE.
Our study is subject to several limitations. Firstly, the lack of definitive diagnostic criteria for SAE presents challenges. Despite our diligent establishment of inclusion and exclusion criteria, the possibility of misdiagnosis and missed diagnosis persists. Secondly, our study exclusively relies on the MIMIC-IV database, which exhibits a certain degree of homogeneity. Furthermore, we solely conducted internal validation utilizing this database. Future research should encompass external databases to augment the model's reliability and performance. Lastly, certain vital laboratory indicators, such as procalcitonin, C-reactive protein, and others, were excluded due to a high prevalence of missing values.