In this retrospective study of the MIMIC-IV database, the incidence of SAE was 62.1%. We identified gender, age, BMI, mean arterial pressure, temperature, platelets , sodium, use of midazolam, and SOFA score as independent risk factors for SAE. These results were used to construct a diagnostic prediction nomogram for SAE. The validity of our nomogram model was tested using multiple indicators such as AUC, calibration curve, Hosmer-Lemesow test, IDI, NRI, and DCA, showing high validity, discrimination, and clinical utility.
Infections with pathogenic pathogens can result in disturbance of the immune response in the host, ultimately leading to severe dysfunction of organs[14]. The existing definition of sepsis, referred to as the "Sepsis-3" criterion, emphasizes the occurrence of organ failure in sepsis patients and necessitates the evaluation of sequential organ failure with a minimum score of two[1]. It is estimated that sepsis affected approximately 49 million individuals globally in 2017, resulting in the deaths of 11 million people[15]. We have also observed a gradual reduction in the mortality rate of sepsis adjusted for age, which could be attributed to advancements in clinical guidelines and care. These improvements have consequently increased the number of sepsis patients who survive the condition[16]. In terms of acute brain dysfunction resulting from sepsis, approximately 50% of sepsis patients admitted to the intensive care unit (ICU) exhibited symptoms such as delirium and coma. This neurological manifestation, unrelated to direct brain infection by the pathogen, is recognized as sepsis-associated encephalopathy (SAE)[17]. Various research studies have extensively linked SAE to higher short-term mortality among sepsis patients[18-20]. Furthermore, our investigation discovered a notably increased in-hospital mortality rate for patients in the SAE group, as well as an extended duration of ICU stay (P<0.001). Currently, the diagnostic criteria and potential risk factors for SAE remain inadequately understood, and there is an absence of reliable methods for the clinical assessment of sepsis-induced neurological dysfunction. Consequently, the development of an early diagnostic predictive model could aid in the diagnosis of SAE and facilitate treatment decision-making.
Although sepsis can develop in patients of any age, age is a powerful risk factor, with patients over 65 experiencing a more than tenfold increase in incidence compared to younger individuals (18-49 years). Consequently, most sepsis survivors (56%) are over 65 years of age, half of whom do not fully recover but instead develop new functional impairments[14, 21], consistent with our study findings. We also found that female patients were more likely to develop SAE, considering differences in the immune system, brain tissue structure, blood-brain barrier, and neuroendocrine system between males and females[22]. However, this contradicts the findings of Feng's study[23]. Many current studies on SAE have matched for age and gender, indicating an understanding of the significant impact of these two factors on disease progression. However, data elucidating the exact way gender influences SAE are lacking, indicating an urgent need for further research. Although research suggests that a higher BMI in middle age is associated with dementia, the same studies have also found that a higher BMI in old age may be a protective factor[24]. Furthermore, a low BMI is associated with more severe neurodegenerative diseases and a higher mortality rate[25]. We also found that patients with a low BMI were more likely to develop SAE, as low BMI often indicates malnutrition, which can affect brain function and cause gut-brain axis dysfunction, leading to SAE.
Additionally, our study found that mean arterial pressure (MAP) and body temperature among vital signs were associated with the occurrence of SAE. MAP is the pressure that most significantly affects autoregulation of blood flow within organs. A MAP of 65-70mmHg is the initial systemic circulation target to ensure organ perfusion pressure[26]. However, patients with sepsis often have impaired cerebrovascular autoregulation[27]. Schramm et al. confirmed through TCD that cerebrovascular autoregulation dysfunction is one of the triggers for SAE[28]. We know that low MAP can lead to insufficient cerebral blood flow and cerebral perfusion disorders; conversely, if MAP is too high and exceeds the range of cerebrovascular autoregulation, it can also lead to increased intracranial pressure and a decrease in cerebral perfusion pressure, which is consistent with our research results. Therefore, individualized MAP based on cerebrovascular autoregulation monitoring is needed to prevent SAE[29]. Sepsis itself is a systemic inflammatory state caused by severe infection. The inflammatory response can disrupt thermoregulatory mechanisms, leading to abnormalities such as fever or hypothermia, and prolonged high fever can exacerbate blood-brain barrier damage and the degree of neuronal necrosis during SAE[30, 31].
Thrombocytopenia is a common complication of sepsis, and previous studies have confirmed that thrombocytopenia is associated with poor prognosis in patients with SAE[31]. However, our study suggests that thrombocytosis upon admission is associated with the occurrence of SAE. This is primarily because platelets can release cytokines and neurotransmitters, such as serotonin, IL-1β, and platelet activating factor, which can promote inflammatory responses and leukocyte migration. Additionally, platelet surface receptors, such as GPIb, can bind to leukocyte surface receptors, such as Mac-1, promoting leukocyte adhesion to the vascular endothelium and entry into brain tissue, thereby triggering neuroinflammation[33]. Sodium, an essential electrolyte for nerve cells and a significant component of plasma osmotic pressure, can cause an increase in plasma osmotic pressure in hypernatremia, leading to a transfer of water from the brain to nerve cells. This can result in symptoms such as somnolence, epilepsy, and delirium[34, 35]. Our study found that hypernatremia is also an independent risk factor for SAE, consistent with the findings of Romain Sonneville's study[31].
Midazolam, a benzodiazepine drug, is commonly used for sedation in ICU patients. However, it may increase the risk of neurocognitive impairment by affecting β-amyloid protein clearance, increasing tau protein levels, exacerbating the inflammatory response, and affecting synaptic plasticity. Previous studies have shown that the use of midazolam is an independent risk factor for predicting the development of neurocognitive impairment in ICU patients, increasing their risk[36, 37]. Our study also confirms that the use of midazolam is one of the risk factors for SAE.
In terms of sepsis diagnosis, the most commonly used tool is the SOFA score, which is also one of the diagnostic criteria for sepsis 3.0. Previous studies have reported that the SOFA score has good diagnostic and prognostic predictive value in sepsis patients[38]. However, whether it is applicable for the diagnosis of SAE is currently unclear[39]. Therefore, we extracted data from the MIMIC-IV dataset to develop the current prediction model. The results show that our prediction model is superior to the current SOFA combined with delirium diagnostic system and displays acceptable discrimination and calibration. In addition, guided by the nomogram, we performed a clinical decision analysis for the diagnosis of SAE patients and found that the current prediction model has a higher net benefit. Note that a GCS score of less than 15 is used to diagnose SAE patients; however, the GCS score is part of the SOFA score. Patients with a higher SOFA score are more likely to develop SAE, which may bias this conclusion.
To the best of our current understanding, there is a scarcity of research on the predictive model to diagnose SAE. In our study, we utilized a publicly accessible database known as MIMIC-IV, which comprises a vast collection of data from critically ill patients. This database served as a robust source of evidence for our findings. The parameters incorporated into the SAE prediction model established in this investigation can be easily acquired through clinical practices. Moreover, the model itself is interpretable, rendering it valuable for early prognosis of SAE in clinical settings
Nonetheless, our study possesses certain limitations. Firstly, the absence of clearly defined diagnostic criteria for SAE poses a challenge. Although we established some inclusion and exclusion criteria, misdiagnoses and missed diagnoses are bound to occur. Secondly, our study extensively relied on the MIMIC-IV database, which is known for its homogeneity. Consequently, we only conducted internal validation with this specific database. To enhance the model's robustness and performance, the inclusion of external databases in future investigations is imperative. Moreover, it is important to acknowledge that our study is retrospective by nature, which inherently introduces biases. Lastly, due to limitations in data availability, several laboratory tests, including PCT, CRP, and IL-6, were not feasible to obtain. Consequently, these potential risk factors could not be included in the prediction model.