In this study, we developed and validated a prognostic nomogram to predict 1-year mortality in patients with SAE using a comprehensive set of clinical variables. Our findings demonstrate that the nomogram exhibits robust discriminative ability and calibration performance, surpassing traditional scoring systems such as the GCS score and other common disease severity indices. By integrating easily obtainable clinical parameters, our model provides clinicians with a practical and reliable tool for risk stratification in SAE, facilitating more informed decision-making and personalized patient management. These results underscore the importance of incorporating multifaceted clinical data to enhance prognostic accuracy and improve outcomes in critically ill patients with sepsis.Our research has made substantial advancements compared to previous studies on prognostic models for SAE. Earlier studies primarily concentrated on short-term outcomes and, although useful for immediate clinical decision-making, often lacked the ability to predict long-term outcomes, particularly for SAE patients11,12,13,14,15. For instance, using the MIMIC III database, a user-friendly nomogram was created to predict 30-day mortality risk in SAE patients, with an AUC of 0.763 [0.736–0.791]11, indicating moderate predictive performance. Another study developed a nomogram based on clinical data to predict in-hospital mortality in SAE patients12. However, this model included the Simplified Acute Physiology Score II (SAPS II) score as a predictive variable, necessitating its prior completion, thereby increasing the model’s complexity and limiting its practical application.Moreover, sophisticated machine learning methods have been employed to construct mortality prediction models, effectively predicting the 30-day mortality rate or ICU mortality rate in SAE patients13,14. Among these models, the APS III score emerged as a significant predictor, yet its necessity for model use adds complexity, hindering clinical adoption. A notable effort involved a stacking ensemble model that achieved a high AUC (0.807) in the test set and 0.671 in external validation for predicting ICU mortality risk in SAE patients using common clinical variables15.In contrast, our nomogram, based on logistic regression, leverages common clinical features as predictive factors, making it both interpretable and straightforward. This simplicity enhances its clinical applicability and ease of use. By focusing on easily obtainable clinical features and ensuring transparency in how these factors influence mortality, our model offers a significant improvement in both predictive performance and practical implementation, facilitating broader clinical adoption.In our predictive model, key risk factors for 1-year mortality in SAE patients include a history of malignancy, higher CCI scores, elevated minimum lactate levels, lower mean body temperature, and decreased maximum lactate levels. Conversely, SAE patients receiving oxygen supplementation exhibited lower 1-year mortality rates. Cancer patients are more susceptible to sepsis than the general population, with sepsis being a leading cause of ICU admissions among these individuals16. Compared to non-cancer sepsis patients, those with cancer have a significantly higher risk of late mortality (OR = 2.46, 95% CI: 1.42–4.25, I²=99%)17. Cancer patients undergo complex immune alterations, with treatments often inducing local or systemic inflammation as a result of tissue damage and the death of cancer cells18. Both chronic host state dysregulation due to cancer and acute host response dysregulation due to sepsis mediate mortality in sepsis patients with pre-existing malignant cancer16.The CCI is a well-established predictor of outcomes in sepsis19, with comorbidities being a significant determinant of infection-related in-hospital mortality20. Accumulation of comorbid conditions is closely linked to increased severity of acute organ dysfunction, underscoring the critical role of comorbidities in the clinical course and prognosis of septic patients21. Consequently, the CCI serves as a valuable tool in stratifying risk and guiding clinical decision-making in patients with sepsis, underscoring the necessity of comprehensive comorbidity assessment in improving prognostication and individualized patient care.Body temperature is inversely correlated with the prognosis of patients with sepsis-associated encephalopathy (SAE), aligning with previous research findings. A systematic review of 42 studies reported mortality rates of 22.2% for septic patients with a fever > 38 °C, 31.2% for normothermic patients, and 47.3% for hypothermic patients (< 36.0 °C)22. Fever appears to enhance the innate immune response, and many bacteria exhibit reduced replication at higher temperatures23. Conversely, hypothermia is common in sepsis and is associated with increased mortality24. Therefore, interventions aimed at warming patients with hypothermic sepsis may improve prognosis.The measurement of serum lactate levels is a critical component in the clinical management of critically ill patients, particularly those with sepsis or septic shock1. Various metabolic changes in sepsis can elevate blood lactate levels, such as increased glycolysis, heightened Na-K pump activity stimulated by catecholamines, alterations in pyruvate dehydrogenase activity, and decreased lactate clearance due to impaired liver perfusion25. Elevated lactate levels are recognized as an independent risk factor for mortality in sepsis patients26, whereas lower lactate concentrations are associated with better outcomes25. Lactate clearance, defined by the change in lactate levels between two time points, is an efficient and cost-effective parameter that holds promise as a target for quantitative recovery27. Early lactate clearance-guided therapy has been linked to reduced mortality in sepsis27. Our study indicates that lower maximum lactate levels are associated with an increased risk of long-term mortality. This finding may initially seem counterintuitive, as elevated lactate levels are typically associated with worse outcomes. However, the interpretation of lactate levels must consider the overall clinical context and trends rather than isolated values. In this study, the association between lower maximum lactate levels and higher minimum lactate levels as indicators of poor prognosis in SAE patients indeed suggests a complex relationship. Specifically, when both lower maximum and higher minimum lactate levels are observed, it implies that the peak lactate level on the first day of ICU admission is relatively close to the trough, which may indicate suboptimal lactate clearance. This pattern could be indicative of inadequate metabolic recovery or persistent underlying issues, even if initial improvements are apparent. Thus, in the presence of other indicators of poor clinical progression, lower maximum lactate levels might reflect more complex clinical conditions that contribute to increased mortality risk. Unfortunately, due to the variability in timing for lactate remeasurement among patients and the absence of standardized protocols, precise lactate clearance rate data are not available in the MIMIC database. This limitation prevents us from obtaining standard lactate clearance rates and further elucidating their direct relationship with SAE mortality. We acknowledge this gap and agree that future research is needed to investigate the exact relationship between standard lactate clearance rates and SAE mortality. Such studies could provide valuable insights into the role of lactate dynamics in predicting outcomes and guide more effective management strategies for SAE patients.The clinical implications of this study are substantial. The developed nomogram, based on readily available clinical variables, provides a practical and reliable tool for predicting 1-year mortality in patients with SAE. By incorporating this predictive model into clinical practice, healthcare providers can more accurately stratify patients based on their risk, facilitating personalized treatment strategies and informed decision-making. This model aids in identifying high-risk patients who may benefit from intensified monitoring and therapeutic interventions, thereby potentially improving clinical outcomes. Furthermore, the nomogram’s superior performance compared to traditional scoring systems underscores its value in enhancing prognostic accuracy. The use of decision curve analysis further emphasizes the model’s clinical utility, demonstrating its ability to offer significant net benefits across a range of threshold probabilities. Ultimately, the adoption of this predictive tool in clinical settings could lead to improved resource allocation, better patient management, and enhanced communication between clinicians and patients regarding prognosis and care plans.This study offers several advantages. Firstly, the use of the MIMIC IV database, which includes a large and diverse patient cohort, enhances the generalizability and robustness of our research findings. The combination of LASSO regression and multivariate logistic regression ensures rigorous selection of predictive factors, thereby ensuring the correlation and accuracy of identified risk factors. Moreover, the development of nomograms incorporates a range of readily available clinical variables, making them practical tools for use across various clinical settings. The validation of the nomograms in both the training and validation cohorts underscores their reliability and robustness. Additionally, the inclusion of decision curve analysis to assess clinical utility provides valuable insights into the practical benefits of nomograms, highlighting their superior performance in predicting 1-year mortality rates in patients with sepsis-associated encephalopathy compared to traditional scoring systems. Therefore, this study not only introduces a novel predictive model but also establishes a benchmark for future research aimed at enhancing prognostic tools for critically ill patients.Several limitations should be noted in this study. Firstly, the retrospective nature of our analysis may introduce inherent biases, including selection and information biases, potentially affecting the generalizability of our findings. Prospective studies are warranted to further validate the established nomogram. Additionally, while the comprehensive use of the MIMIC IV database provides robust data, its restriction to a single healthcare system may limit the applicability of our results to broader populations and settings. Therefore, we plan to conduct external validation of our prediction model in diverse settings. Furthermore, although the nomogram demonstrates strong predictive performance, it relies on variables available within the database. During variable selection, we carefully considered factors such as the feasibility of data collection, accessibility, and economic considerations. Finally, while our study included a wide array of clinical variables, the MIMIC-IV database has inherent limitations, including the absence of certain potential predictive factors such as some biomarkers, genetic data, electroencephalography, and cranial imaging examinations. These limitations underscore the need for further research to explore these additional factors and enhance our understanding of SAE. Future prospective studies should address these limitations and validate our findings across various medical contexts and patient populations.
