https://orcid.org/0009-0007-3104-0362
Figures
Abstract
This study aimed to develop models for predicting the 30-day mortality of sepsis-associated delirium (SAD) by multiple machine learning (ML) algorithms. On the whole, a cohort of 3,197 SAD patients were collected from the Medical Information Mart for Intensive Care (MIMIC)-IV database. Among them, a total of 659 (20.61%) patients died following SAD. The patients who died were about 73.00 (62.00, 82.00) years old and mostly male (56.75%). Recursive feature elimination (RFE) was used to distinguish risk factors. Subsequently, six ML algorithms including artificial neural network (NNET), gradient boosting machine (GBM), adaptive boosting (Ada), random forest (RF), eXtreme Gradient Boosting (XGB) and logistic regression (LR) were employed to establish models to predict the 30-day mortality of SAD. The performance of models was assessed via both discrimination and calibration by cross-validation with 100 resamples. Overall, 10 independent predictors, including Glasgow Coma Scale (GCS), Sequential Organ Failure Assessment (SOFA), anion gap (AG), continuous renal replacement therapy (CRRT), temperature, mean corpuscular hemoglobin concentration (MCHC), vasopressor, blood urea nitrogen (BUN), base excess (BE), and bicarbonate were identified as independent predictors for the 30-day mortality of SAD. The validation cohort demonstrated that all these six models had relatively favorable differentiation, while among them, the GBM model had the highest area under the curve (AUC) of 0.845 (95% Confidence Interval (CI): 0.816, 0.874). Furthermore, the calibration curve of these six models was close to the diagonal line in the validation sets. As for decision curve analysis, the predictive models were clinically useful as well. Based on real-world research, we developed ML models to provide personalized predictions of delirium-related mortality in sepsis patients, potentially enabling clinicians to identify high-risk SAD patients more promptly.
Citation: Shen X, Shang D, Sun W, Ru S (2025) Machine learning approach for the prediction of 30-day mortality in patients with sepsis-associated delirium. PLoS ONE 20(4): e0319519. https://doi.org/10.1371/journal.pone.0319519
Editor: Chinh Quoc Luong,, Bach Mai Hospital, VIETNAM
Received: November 4, 2024; Accepted: February 3, 2025; Published: April 9, 2025
Copyright: © 2025 Shen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All data are available from the MIMIC-IV database https://github.com/pcstudy2019/sad.
Funding: This study was financially supported by the Wuxi Science and Technology Association project (2023) in the form of an award (KX-23-C165) received by SR. No additional external funding was received for this study. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Sepsis-associated delirium (SAD) is recognized as a widespread cerebral dysfunction resulting from the systemic inflammatory response to an infection, in the absence of any central nervous system (CNS) infection [1]. In terms of epidemiology, its reported incidence rates range from 17.7 to 48% among septic patients who are admitted to the intensive care unit (ICU) [2,3]. It is reported that SAD survivors are likely to develop persistent neurocognitive impairment, such as overt dementia [4]. Significantly, the presence of delirium and coma typically signals severe brain dysfunction [5]. Unfortunately, the pathophysiology of SAD remains not well understood, but it is known to involve systemic inflammation that can impair the blood-brain barrier, further resulting in the infiltration of peripheral leukocytes into the central nervous system [6]. The resultant neuroinflammation causes a state of cholinergic failure, predisposing patients to delirium [7].
Previous studies have shown that the development of SAD is linked to extended durations of mechanical ventilation and longer ICU stays [8]. As indicated by Tokuda R et al [6], delirium by itself not only worsens short-term prognosis, but it becomes a barrier to early mobilization and physical rehabilitation. Further, all these conditions may aggravate the outcome of post-intensive care syndrome (PICS) [9]. In short, as a disease manifested as prolonged or aggravated brain dysfunction, SAD can cause a great financial burden to families in individual and societies in general, respectively. However, there are still no specific predictive models, as well as no specific early warning systems for severe SAD patients. Recently, machine learning (ML) algorithms can deal with massive, redundant, non-linear clinical information [10]. Moreover, ML algorithms are good at recognizing patterns and learning from input data to classify and predict outcomes [11]. In the field of sepsis, ML algorithms are widely used, and they also show sufficient advantages, such as predicting the occurrence of sepsis, which could trigger direct clinical action including administration of antibiotics [12]. Specifically, ML-based prediction models are capable of determining optimal antibiotic dosing without the need for daily therapeutic drug monitoring (TDM) [13]. The aim of this study was to explore potential independent predictors, and then try to develop ML models that can quantitatively estimate the probability of 30-day mortality in patients with SAD.
Methods
Data source
This is a retrospective study conducted on the Medical Information Mart for Intensive Care (MIMIC)-IV version 2.2 [14]. Briefly, the MIMIC-IV database included comprehensive, de-identified data of patients admitted to the ICUs at the Beth Israel Deaconess Medical Center in Boston, Massachusetts, between 2008 and 2019, including data from 383,220 admissions. Requirement for individual patient consent was waived because the study did not impact clinical care. The study was written in accordance with the REporting of studies Conducted using Observational Routinely collected health Data (RECORD) statement [15].
Participant selection
Patients were extracted from the MIMIC-IV according to the following criteria. (1) first admitted to the ICU; (2) age ≥ 18; (3) diagnosed with sepsis; (4) ICU admission time ≥ 48 h; and (5) delirium diagnosis. The Confusion Assessment Method for ICU Patients (CAM-ICU) is used for identifying delirium in this paper [16]. More precisely, SAD in our study was defined as sepsis patients with a positive assessment of CAM-ICU and delirium assessment during their stay in ICUs. The diagnosis of sepsis was based on the Third International Consensus Definitions for Sepsis (Sepsis-3), which includes patients with documented or suspected infection and Sequential Organ Failure Assessment (SOFA) [17]. People with primary encephalopathy, chronic alcohol or drug abuse, pre-existing liver or kidney failure affecting consciousness were excluded from the study.
Predictors of 30-day mortality in SAD patients
All research data were extracted from the MIMIC-IV by Structured query language (SQL). And they included (1) patients’demographics; (2) coexisting disorders; (3) vital signs (temperature, mean artery pressure (MAP), heart rate and respiratory rate) on the day of ICU admission; (4) laboratory measurements (such as red blood cell (RBC), white blood cell (WBC) and so on) on the day of ICU admission; (5) details of therapy (vasopressor, continuous renal replacement therapy (CRRT) and mechanical ventilation (MV)); (6) severity score including Glasgow Coma Scale (GCS) and SOFA. Multiple imputation was used to deal with missing data. Details of missing data are demonstrated in S1Table.
Statistical analysis
Continuous variables were described using means and standard deviations for normally distributed data, or medians with interquartile ranges (IQR) for non-normally distributed data. And categorical variables were expressed as total numbers with percentages. Proportions were compared using χ² test or Fisher exact tests whereas continuous variables were compared by the t test or Wilcoxon rank sum test, as appropriate.
Initially, the population was randomly divided into training and validation sets in a 7:3 ratio. As a variable selection method, recursive feature elimination (RFE) was used to choose the most relevant variables related to 30-day mortality in SAD patients. In summary, RFE recursively fits into a model based on smaller resource set. In each loop, characteristics were divided by variable numbers of 5/10/15/20/25/30/35/ALL (ALL = 50 variables, as represented in Fig 1). 5-fold cross-validation was used as the resampling method to find the optimal hyperparameters. The training test process was repeated 100 times. In this study, six different ML algorithms were used to develop models, including artificial neural network (NNET), gradient boosting machine (GBM), adaptive boosting (Ada), random forest (RF), eXtreme Gradient Boosting (XGB) and logistic regression (LR). The NNET algorithm is a ML algorithm based on biological networks. Based on the backpropagation algorithm, this model relies on nodes and connections to make complex decisions [18]. GBM is a type of ML algorithm that sequentially combines weak models into a strong model. Model integration means constructing a prediction model by combining multiple base models, and the prediction performance of the final integrated model is better than that of any single base model [19]. Ada is a simple and general ensemble algorithm that creates a strong classifier by using many weak classifiers. The algorithm starts with weak learners and assigns weights to each training sample. Finally, the predictions are combined with an ensemble weight to obtain the final prediction [20]. RF is a common ML method based on decision trees and has been used in a variety of medical studies. It can be applied to various complex data sets. Through the bootstrap aggregation process, the variance of the model can be reduced, thereby improving the prediction accuracy [21]. XGB is an integrated gradient boosting method, which uses the negative gradient of the loss function as the residual value for model fitting to achieve accurate classification results [22].
Model performance was evaluated by area under the curve (AUC), accuracy, sensitivity, specificity, negative predictive value (NPV) and positive predictive value (PPV). Additionally, the calibration curve was used to evaluate how well the observed outcomes matched the predicted outcomes, while the decision curve analysis assessed the net clinical benefit. The SHapley Additive exPlanations (SHAP) method was applied to interpret the final predictive model. Higher SHAP values indicated a greater predicted probability of mortality among SAD patients [23]. In order to facilitate clinical use, we also launched a web page by “Shiny” application. For missing values, the method of multiple imputation was used.
All analyses were conducted using R version 4.0.2 (http://www.R-project.org, The R Foundation). For our study, we utilized the “Caret” and “Shiny” R packages to facilitate the process. Statistical significance was defined as P values of less than 0.05 (two-sided test).
Results
Participants and baseline characteristics
Based on the inclusion and exclusion criteria, the final dataset comprised 3,197 patients. Fig 1 demonstrated the process of data extraction, training preparation, data testing by multiple ML algorithms. Baseline characteristics of all included patients are shown in Table 1. Dead individuals were generally older and had more comorbidities, including myocardial infarction, mild and severe liver disease, renal disease, malignant cancer, and metastatic solid tumors. They also exhibited higher values for the Charlson Comorbidity Index (CCI), as well as elevated heart rate, respiratory rate, white blood cell count (WBC), red blood cell volume distribution width (RDW), activated partial thromboplastin time (APTT), prothrombin time (PT), international normalized ratio (INR), lactate, base excess (BE), anion gap (AG), potassium, glucose, creatinine (CRE), and blood urea nitrogen (BUN). As for therapy, SAD people who died were more likely to receive vasopressor, CRRT, and MV. Disease severity score indicated that lower GCS and higher SOFA occurred in SAD patients who died. Additionally, these individuals typically had lower temperature, mean arterial pressure (MAP), RBC, mean corpuscular hemoglobin concentration (MCHC), platelet count (PLT), hematocrit (HCT), pH, bicarbonate, partial pressure of oxygen (PaO2), chloride, and calcium levels. In summary, patients who died had a poorer baseline health status.
Variable selection
The RFE algorithm selected a total of 10 important predictors (S1 Fig). Specifically, predictors included GCS, SOFA, AG, CRRT, temperature, MCHC, vasopressor, BUN, BE, and bicarbonate. Subsequently, these 10 variables were used in all analysis for all models in both training and testing sets.
Model performance
Table 2 and Fig 2 describe the predictive performance of these models on the validation set. By comparison, GBM had the highest predictive performance among these six models (AUC 0.845, 95% confidence interval (CI) 0.816 to 0.874). As for the calibration performance, the NNET, GBM, Ada, RF and LR models had a good calibration—namely, being close to the ideal diagonal line, with P-values of 0.052, 0.345, 0.632, 0.356 and 0.346, respectively (Hosmer-Lemeshow test) (Fig 3). Furthermore, decision curve analysis (DCA) was performed on these three models, and the results are shown in Fig 4. The analysis indicated that these four prediction models provided favorable net benefit for predicting 30-day mortality of SAD. To be specific, the X-axis indicated the threshold probability for critical care outcome and Y-axis indicated the net benefit. (red line = Ada, blue line = GBM, green line = LR, purple line = NNET, orange line = RF, yellow line = XGB).
Model interpretations
To determine the key factors in the model, we visualized the feature importance ranking using the GBM algorithm. As shown in Fig 5, each variable included in the study showed different levels of significance for SAD mortality based on the GBM approach. Overall, GCS emerged as the most crucial factor, followed by MCHC, temperature, and so forth.
Model application
In order to further visualize the results, and facilitate the use of GBM models by clinicians, we further uploaded an online calculator (https://pcstudy.shinyapps.io/Mortality_of_sad/). As shown in Fig 6, the Shiny package further visualizes the results, explaining the weight of each variable in predicting 30-day mortality for SAD. Take a patient for example: GCS (7), SOFA (16), AG (20), CRRT, temperature (33.1 °C), MCHC (31.9), vasopressor, BUN (91), BE (12), bicarbonate (37). Through model analysis, the 30-day mortality risk of SAD patients was 84.90%, indicating that the 30-day mortality rate of SAD patients was high.
Entering the input value determined the mortality and displayed how each value contributed to the prediction.
Discussion
Timely and swift recognition of SAD patients who have a poor prognosis continues to be a significant challenge. In this research, our findings indicated that around 20.61% of SAD patients in the ICU succumbed to their condition, with SAD patients in the mortality group having older ages, more comorbidities, worse vital signs compared to those in the non-mortality group. We then used RFE to screen variables most associated with death. Then, six models including NNET, GBM, Ada, RF, XGB and LR, based on ML algorithms were established, Comparably, GBM-based ML predictive model demonstrated commendable predictive performance in the validation cohort, enabling early prediction of the 30-day mortality of SAD on ICU admission. For ease of use by clinicians, we then developed a web-based calculator.
Existing research on SAD has focused on the diagnosis of SAD or examined incidence and risk factors. For instance, Gu Q et al [24] found that SOFA score, requiring mechanical ventilation, first lactate level, and first serum phosphate on ICU admission are indicators that can forecast the onset of delirium in sepsis patients. Another study suggested that emergency surgery, higher doses of midazolam, and fentanyl could serve as independent risk factors for SAD in mechanically ventilated patients suffering from sepsis [3]. In addition, in Zhang Y’s study [25], the XGBoost model outperformed other models in predicting SAD. It is worth mentioning that these studies have focused on the diagnosis of SAD, and few studies have predicted the prognosis of SAD. Previous study conducted by Zhang L et al [26] identified five independent risk factors for developing a prediction model for mortality in SAD: total length of stay in the ICU, hemoglobin level, albumin level, activated partial thromboplastin time, and total bilirubin level. But simple logistic regression may not fit the relationship between variables well. Actually, predictions about mortality can help doctors identify patients with critical SAD. In this study, ten variables were identified as predictors, involving GCS, SOFA, AG, CRRT, temperature, MCHC, vasopressor, BUN, BE, and bicarbonate. Research involving 16,343 sepsis patients revealed that the initial 24-hour GCS score is a significant predictor of SAD, aligning with the findings of two earlier models for diagnosing SAD [25,27,28]. Apart from that, Peng L [29] reported that systematic scores including GCS and SOFA can be used as predictors for the prognosis of SAD patients. Additionally, a series of previous studies also have found that higher values of the GCS and SOFA were related to the mortality rate of SAD patients [6,30,31]. As an accepted method of assessing the severity of a patient’s condition upon admission to the ICU, GCS and SOFA are tools to forecast the results for patients in the ICU [32]. In our study, we found that GCS and SOFA were prominent variables in the importance plot, indicating their significant predictive capability for the 30-day mortality of SAD patients. This highlights the utilization of the GCS and SOFA in evaluating the severity and prognosis of individuals suffering from SAD. Patients with SAD who experience multiple organ dysfunction have a higher likelihood of mortality. Cascade immune response, circulatory abnormalities, coupled with metabolic acidosis increase may be associated with such a complicated pathophysiological process[33]. Thus, the management of SAD mainly concerns the treatment of sepsis itself and the correction of potential neurotoxic dysfunction. Moreover, our study found that body temperature among vital signs was associated with the prognosis of SAD. The potential mechanism may be that inflammatory responses caused by temperature can disrupt thermoregulatory mechanisms, resulting in fever or hypothermia. Sustained abnormal body temperature can disrupt blood-brain barrier and neural function during SAD [30,34]. Jin Z et al [35] found that temperature control has some association with the risk of postoperative delirium. Similarly, in Ju JW et al’s study [36], intraoperative hypothermia was significantly associated with an increased risk of postoperative delirium.
In clinical practice, vasopressin is increasingly utilized as a vasoactive agent. A prior study by Maheshwari et al discovered a significant association between changes in blood pressure and decreased survival in patients with septic shock [37]. Early-stage resuscitation is generally recognized as a promising approach for treating sepsis, and administering vasopressin can reduce the severity of the condition, potentially aiding in cognitive improvement for patients with SAD [38, 39]. Additionally, patients with septic shock need vasoactive agents to ensure sufficient tissue perfusion. Following norepinephrine, vasopressin is recommended as a second-line adjunctive treatment for those with ongoing insufficient mean arterial pressure [40].
Additionally, Chen J et al [41] found that low MCHC was risk factors of delirium, which was similar to our findings. During sepsis, systemic inflammation and oxidative stress can disrupt erythropoiesis by lowering erythropoietin (EPO) levels and reducing iron availability, which subsequently leads to decreased hemoglobin levels [42–45]. Worse all, inflammation can interfere with the maturation of RBCs in the bone marrow, causing the premature release of immature RBCs into the bloodstream, further disturbing the hemoglobin level [46, 47]. Moreover, decreased MCHC leads to reduced oxygen transport capacity and hypoxia of tissue cells, and MCHC is also a marker of malnutrition [48]. Therefore, tracking hemoglobin levels is essential for managing sepsis and assessing long-term survival prospects in patients with SAD.
The study’s strengths are found in its use of ML. Based on parameters collected within 24 hours of admission in ICU, this ML model is able to predict the mortality of SAD much earlier. In addition, cross-validation can decrease potential overfitting.
There were limitations in this study. First, as a retrospective observational study, it has inherent limitations, such as the potential presence of unmeasured confounders. Second, since the study was conducted at a single center, further validation is needed to ensure its findings are applicable to other populations. Third, ML algorithms are susceptible to data biases, such as the selection of specific patients, and potential biases inherent in retrospective datasets [49,50]. Fourth, the prediction process of ML algorithm is mostly “black box”, and its complex process makes many clinicians remain wary of the model. Thus, there has been a recent explosion of research in the field of explainable ML aimed at addressing these concerns [51]. Lastly, the ML algorithm is dependent on the data set used for training, and its universal application may be limited by different clinical environments.
Conclusion
In summary, SAD is prevalent among ICU patients with sepsis and is associated with higher mortality rates compared to sepsis alone. By employing a ML-based early prediction model, we can more effectively predict 30-day mortality in SAD patients. Moreover, developing this model aids in early risk detection and allows for the implementation of preventive strategies, which could help lower the incidence and mortality of SAD to some degree.
Supporting information
S1 Table. Missing number (%) for included variables in the dataset.
https://doi.org/10.1371/journal.pone.0319519.s001
(DOCX)
References
- 1. Iacobone E, Bailly-Salin J, Polito A, Friedman D, Stevens RD, Sharshar T. Sepsis-associated encephalopathy and its differential diagnosis. Crit Care Med. 2009;37(10):S331-6. pmid:20046118
- 2. Kim Y, Jin Y, Jin T, Lee S-M. Risk factors and outcomes of sepsis-associated delirium in intensive care unit patients: a secondary data analysis. Intensive Crit Care Nurs. 2020;59:102844. pmid:32253122
- 3. Yamamoto T, Mizobata Y, Kawazoe Y, Miyamoto K, Ohta Y, Morimoto T, et al. Incidence, risk factors, and outcomes for sepsis-associated delirium in patients with mechanical ventilation: a sub-analysis of a multicenter randomized controlled trial. J Crit Care. 2020;56:140–4. pmid:31901649
- 4. Tauber SC, Djukic M, Gossner J, Eiffert H, Brück W, Nau R. Sepsis-associated encephalopathy and septic encephalitis: an update. Expert Rev Anti Infect Ther. 2021;19(2):215–31. pmid:32808580
- 5. Pandharipande PP, Pun BT, Herr DL, Maze M, Girard TD, Miller RR, et al. Effect of sedation with dexmedetomidine vs lorazepam on acute brain dysfunction in mechanically ventilated patients: the MENDS randomized controlled trial. JAMA. 2007;298(22):2644–53. pmid:18073360
- 6. Tokuda R, Nakamura K, Takatani Y, Tanaka C, Kondo Y, Ohbe H, et al. Sepsis-associated delirium: a narrative review. J Clin Med. 2023;12(4):1273. pmid:36835809
- 7. Gao Q, Hernandes MS. Sepsis-associated encephalopathy and blood-brain barrier dysfunction. Inflammation. 2021;44(6):2143–50. pmid:34291398
- 8. Salluh JIF, Wang H, Schneider EB, Nagaraja N, Yenokyan G, Damluji A, et al. Outcome of delirium in critically ill patients: systematic review and meta-analysis. BMJ. 2015;350h2538. pmid:26041151
- 9. Inoue S, Hatakeyama J, Kondo Y, Hifumi T, Sakuramoto H, Kawasaki T, et al. Post-intensive care syndrome: its pathophysiology, prevention, and future directions. Acute Med Surg. 2019;6(3):233–46. pmid:31304024
- 10. Haug C, Drazen J. Artificial intelligence and machine learning in clinical medicine. N Engl J Med. 2023;388(13):1201–8.
- 11. Ramkumar PN, Kunze KN, Haeberle HS, Karnuta JM, Luu BC, Nwachukwu BU, et al. Clinical and research medical applications of artificial intelligence. Arthroscopy. 2021;37(5):1694–7. pmid:32828936
- 12. Fleuren LM, Klausch TLT, Zwager CL, Schoonmade LJ, Guo T, Roggeveen LF, et al. Machine learning for the prediction of sepsis: a systematic review and meta-analysis of diagnostic test accuracy. Intensive Care Med. 2020;46(3):383–400. pmid:31965266
- 13. Marko B, Palmowski L, Nowak H, Witowski A, Koos B, Rump K, et al. Employing artificial intelligence for optimising antibiotic dosages in sepsis on intensive care unit: a study protocol for a prospective observational study (KI.SEP). BMJ Open. 2024;14(12):e086094. pmid:39672586
- 14. Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci Data. 2023;10(1):1. pmid:36596836
- 15. Benchimol EI, Smeeth L, Guttmann A, Harron K, Moher D, Petersen I, et al. The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statement. PLoS Med. 2015;12(10):e1001885. pmid:26440803
- 16. Miranda F, Gonzalez F, Plana MN, Zamora J, Quinn TJ, Seron P. Confusion Assessment Method for the Intensive Care Unit (CAM-ICU) for the diagnosis of delirium in adults in critical care settings. Cochrane Database System Rev 2023;11(11):Cd013126.
- 17. Singer M, Deutschman CS, Seymour CW, Shankar-Hari M, Annane D, Bauer M, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA. 2016;315(8):801–10. pmid:26903338
- 18. Ojha VK, Abraham A, Snášel V. Metaheuristic design of feedforward neural networks: a review of two decades of research. Eng Appl Artif Intell. 2017;60:97–116.
- 19. Friedman JH. Greedy function approximation: a gradient boosting machine. J Am Stat Assoc. 2001;291189–232.
- 20. Freund Y, Schapire RE. A decision-theoretic generalization of on-line learning and an application to boosting. J Comput Syst Sci. 1997;55(1):119–39.
- 21. Fernández-Delgado M, Cernadas E, Barro S, Amorim D. Do we need hundreds of classifiers to solve real world classification problems? J Mach Learn Res. 2014;15(1):3133–3181.
- 22. Fan Z, Song W, Ke Y, Jia L, Li S, Li JJ, et al. XGBoost-SHAP-based interpretable diagnostic framework for knee osteoarthritis: a population-based retrospective cohort study. Arthritis Res Ther. 2024;26(1):213.
- 23.
Lundberg S, Lee S-I. A unified approach to interpreting model predictions. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, CA, USA: Curran Associates Inc.; 2017: 4768–77.
- 24. Gu Q, Yang S, Fei D, Lu Y, Yu H. A nomogram for predicting sepsis-associated delirium: a retrospective study in MIMIC III. BMC Med Inform Decis Mak. 2023;23(1):184. pmid:37715189
- 25. Zhang Y, Hu J, Hua T, Zhang J, Zhang Z, Yang M. Development of a machine learning-based prediction model for sepsis-associated delirium in the intensive care unit. Sci Rep. 2023;13(1):12697. pmid:37542106
- 26. Zhang L, Li X, Huang J, Yang Y, Peng H, Yang L, et al. Predictive model of risk factors for 28-day mortality in patients with sepsis or sepsis-associated delirium based on the MIMIC-IV database. Sci Rep. 2024;14(1):18751. pmid:39138233
- 27. van den Boogaard M, Pickkers P, Slooter AJC, Kuiper MA, Spronk PE, van der Voort PHJ, et al. Development and validation of PRE-DELIRIC (PREdiction of DELIRium in ICu patients) delirium prediction model for intensive care patients: observational multicentre study. BMJ. 2012;344:e420. pmid:22323509
- 28. Gong KD, Lu R, Bergamaschi TS, Sanyal A, Guo J, Kim HB, et al. Predicting intensive care delirium with machine learning: model development and external validation. Anesthesiology. 2023;138(3):299–311. pmid:36538354
- 29. Peng L, Peng C, Yang F, Wang J, Zuo W, Cheng C, et al. Machine learning approach for the prediction of 30-day mortality in patients with sepsis-associated encephalopathy. BMC Med Res Methodol. 2022;22(1):183. pmid:35787248
- 30. Sonneville R, de Montmollin E, Poujade J, Garrouste-Orgeas M, Souweine B, Darmon M, et al. Potentially modifiable factors contributing to sepsis-associated encephalopathy. Intensive Care Med. 2017;43(8):1075–84. pmid:28466149
- 31. Liu X, Niu H, Peng J. Enhancing predictions with a stacking ensemble model for ICU mortality risk in patients with sepsis-associated encephalopathy. J Int Med Res. 2024;52(3):3000605241239013. pmid:38530021
- 32. Kari KA, Wan Muhd Shukeri WF, Yaacob NM, Li AY, Zaini RH, Mazlan MZ. Prevalence and outcome of sepsis: mortality and prolonged intensive care unit stay among sepsis patients admitted to a tertiary Centre in Malaysia. Malays J Med Sci. 2023;30(6):120–32. pmid:38239259
- 33. Nolt B, Tu F, Wang X, Ha T, Winter R, Williams DL, et al. Lactate and Immunosuppression in Sepsis. Shock. 2018;49(2):120–5. pmid:28767543
- 34. Widmann CN, Heneka MT. Long-term cerebral consequences of sepsis. Lancet Neurol. 2014;13(6):630–6. pmid:24849863
- 35. Jin Z, Hu J, Ma D. Postoperative delirium: perioperative assessment, risk reduction, and management. Br J Anaesth. 2020;125(4):492–504. pmid:32798069
- 36. Ju J-W, Nam K, Sohn JY, Joo S, Lee J, Lee S, et al. Association between intraoperative body temperature and postoperative delirium: a retrospective observational study. J Clin Anesth. 2023;87:111107. pmid:36924749
- 37. Maheshwari K, Nathanson BH, Munson SH, Khangulov V, Stevens M, Badani H, et al. The relationship between ICU hypotension and in-hospital mortality and morbidity in septic patients. Intensive Care Med. 2018;44(6):857–67. pmid:29872882
- 38. Pan S, Lv Z, Wang R, Shu H, Yuan S, Yu Y, et al. Sepsis-induced brain dysfunction: pathogenesis, diagnosis, and treatment. Oxid Med Cell Longev. 2022;2022:1328729. pmid:36062193
- 39. Comim CM, Freiberger V, Ventura L, Mina F, Ferreira GK, Michels M, et al. Inhibition of indoleamine 2,3-dioxygenase 1/2 prevented cognitive impairment and energetic metabolism changes in the hippocampus of adult rats subjected to polymicrobial sepsis. J Neuroimmunol. 2017;305:167–71. pmid:28284339
- 40. Sacha GL, Bauer SR. Optimizing vasopressin use and initiation timing in septic shock: a narrative review. Chest. 2023;164(5):1216–27. pmid:37479058
- 41. Chen J, Shi X, Diao M, Jin G, Zhu Y, Hu W, et al. A retrospective study of sepsis-associated encephalopathy: epidemiology, clinical features and adverse outcomes. BMC Emerg Med. 2020;20(1):77. pmid:33023479
- 42. Förhécz Z, Gombos T, Borgulya G, Pozsonyi Z, Prohászka Z, Jánoskuti L. Red cell distribution width in heart failure: prediction of clinical events and relationship with markers of ineffective erythropoiesis, inflammation, renal function, and nutritional state. Am Heart J. 2009;158(4):659–66. pmid:19781428
- 43. Pierce CN, Larson DF. Inflammatory cytokine inhibition of erythropoiesis in patients implanted with a mechanical circulatory assist device. Perfusion. 2005;20(2):83–90. pmid:15918445
- 44. Scharte M, Fink MP. Red blood cell physiology in critical illness. Crit Care Med. 2003;31(12 Suppl):S651-7. pmid:14724462
- 45. Lippi G, Targher G, Montagnana M, Salvagno GL, Zoppini G, Guidi GC. Relationship between red blood cell distribution width and kidney function tests in a large cohort of unselected outpatients. Scand J Clin Lab Invest. 2008;68(8):745–8. pmid:18618369
- 46. Perlstein TS, Weuve J, Pfeffer MA, Beckman JA. Red blood cell distribution width and mortality risk in a community-based prospective cohort. Arch Intern Med. 2009;169(6):588–94. pmid:19307522
- 47. Piagnerelli M, Boudjeltia KZ, Vanhaeverbeek M, Vincent J-L. Red blood cell rheology in sepsis. Intensive Care Med. 2003;29(7):1052–61. pmid:12802488
- 48. Chen R, Zhou X, Rui Q, Wang X. Combined predictive value of the risk factors influencing the short-term prognosis of sepsis. Zhonghua Wei Zhong Bing Ji Jiu Yi Xue. 2020;32(3):307–12. pmid:32385994
- 49. Hashimoto DA, Witkowski E, Gao L, Meireles O, Rosman G. Artificial intelligence in anesthesiology: current techniques, clinical applications, and limitations. Anesthesiology 2020, 132(2):379–394.
- 50. Murthy VH, Krumholz HM, Gross CP. Participation in cancer clinical trials: race-, sex-, and age-based disparities. JAMA. 2004;291(22):2720–6. pmid:15187053
- 51. Petch J, Di S, Nelson W. Opening the black box: the promise and limitations of explainable machine learning in cardiology. Can J Cardiol. 2022;38(2):204–13. pmid:34534619