Machine Learning for Renal Replacement Therapy in Patients with Severe Acute Kidney Injury Associated with Sepsis

Machine Learning for Renal Replacement Therapy in Patients with Severe Acute Kidney Injury Associated with Sepsis | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Machine Learning for Renal Replacement Therapy in Patients with Severe Acute Kidney Injury Associated with Sepsis Qiqiang Liang, Sumian Zhang, Haiyan Ye, Mei Yang, Xuebin Wang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6321504/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background : The high risk of renal replacement therapy (RRT) in patients with sepsis-associated acute kidney injury (SA-AKI) remains unclear. Methods : This two-center retrospective study analyzed data from the MIMIC database and the eICU database. SA-AKI patients were included, and demographic data and laboratory parameters from the 5 days before diagnosis were collected as variables. The MIMIC database served as the training and validation set for developing models using multivariate logistic regression, random forest, SVM and XGBoost. Moreover, the eICU database was used as an external test set. We visualized the model using nomograms and the SHAP method, and evaluated model performance through variety of methods. Results : A total of 22,220 patients with severe SA-AKI were included in the analysis, with 1,358 (6.11%) receiving RRT. The RRT group exhibited more severe metabolic acidosis, including lower PH (7.23 vs. 7.31, P<0.001) , lower buffer excess(-8.25 vs. -3.10, P<0.001), higher lactate levels (5.07 vs. 3.03, P<0.001), more disordered electrolytes and poorer coagulation function. The six most important parameters were the APS score, SOFA score, minimum calcium level, maximum APTT, maximum BUN, and minimum buffer excess in the XGBoost model with great performance. The Area Under the Receiver Operating Characteristic Curve for the internal and external test and sets were 0.906 and 0.816, respectively. In the meanwhile, XGBoost has the best Area Under the Precision-Recall Curve。 Conclusion : Our study constructed a predictive model for initiating RRT in critically ill patients with severe SA-AKI using twocenter databases and the machine learning algorithms. Sepsis-associated acute kidney injury Renal Replacement Therapy Two-center databases Machine learning algorithms Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Background Acute kidney injury (AKI) being one of the most common organ dysfunctions of sepsis 1 , 2 . Up to 60% of sepsis patients develop AKI, but the pathophysiology remains incompletely understood 1 . Sepsis involves a harmful inflammatory cascade that contributes to AKI. However, AKI also results from factors such as hypovolemia, nephrotoxic antibiotics, and complex urinary tract infections; hence, AKI cannot be solely attributed to sepsis. Sepsis-associated acute kidney injury (SA-AKI) has long been recognized, but a precise definition was only established in 2023. The Acute Disease Quality Initiative (ADQI) 28 Workgroup defined SA-AKI as AKI occurring within 7 days of sepsis diagnosis 3 . The rationale for the suggested 7-day is based on observations of sepsis cases with AKI occurring within a few days of the onset of sepsis, whereas AKI occurring after a week may not be directly related to the initial sepsis 3 . AKI is categorized into three stages according to the Kidney Disease: Improving Global Outcomes (KDIGO) criteria, which are based on the baseline and changes in serum creatinine levels, as well as urine output. KDIGO stage II/III is often defined as severe AKI 4 . In terms of disease progression, the majority of patients recover from mild AKI, which corresponds to KDIGO stage I; in contrast, only a small proportion of patients progress to severe AKI. Patients with severe AKI have a higher risk of requiring renal replacement therapy (RRT). RRT is a critical component in the management of SA-AKI patients, providing support in maintaining fluid balance, removing inflammatory mediators and toxins, and supporting renal function recovery. The emergency indications for initiating RRT for SA-AKI are similar to other types of AKI. However, the timing of initiating RRT remains a clinical challenge and a controversial topic in the absence of emergency indications 5 . The ELAIN study concluded that early initiation of RRT in patients with severe AKI reduced 90-day mortality and length of hospital stay 6 . However, some high-quality RCTs showed conflicting results, including the recent STARRT-AKI and IDEAL-ICU studies 7 , 8 . Despite these studies employing different AKI diagnostic criteria and varying definitions of RRT timing, none found a difference in 90-day survival rates between the early and delayed groups. The debate over the timing of RRT initiation in patients with severe AKI reflects the impracticality of a one-size-fits-all strategy. Consequently, scholars have proposed strategies such as dynamic renal function assessment, focusing on renal demand-capacity matching and dynamic evaluation of renal biomarkers, to closely monitor and decide on the timing of RRT 5 . With the advances in artificial intelligence, a growing number of scholars have employed machine learning to construct powerful models to assist in identifying patients with AKI who require RRT. Li G et al. analyzed data from 8289 patients from the MIMIC-III database, among whom 591 received RRT and 7698 did not. The team developed a logistic regression model and plotted a nomogram to predict whether SA-AKI patients could benefit from RRT, determining that the class transition model is more effective in predicting individual treatment outcomes 9 . Moreover, Palmowski L et al. conducted a prospective, multicenter study involving 99 patients, predicting whether SA-AKI patients would require RRT through the furosemide stress test (FST) and urinary biomarkers TIMP-2*IGFBP-7, achieving a prediction accuracy of 83% 10 . However, the data of the above models was limited to a single center and was not subjected to external validation, so the results cannot be extrapolated. The present study aims to use two databases combined with machine learning algorithms to construct a prediction model to screen patients needing RRT in severe SA-AKI cases. Methods Study Design This two-center retrospective study collected data from two large, publicly available ICU databases: the MIMIC database and the eICU-CRD database. The MIMIC database includes MIMIC-III and MIMIC-IV databases, which encompass clinical data from over 200,000 patients at the Beth Israel Deaconess Medical Center in Boston, Massachusetts. The institutional review boards of the Massachusetts Institute of Technology (No. 0403000206) and Beth Israel Deaconess Medical Center (2001-P-001699/14) approved the use of this database for research purposes. The eICU database is composed of health data from over 200,000 ICU admissions in the United States between 2014 and 2015. Both databases integrate comprehensive clinical data, including demographics, hourly vital signs, clinical measurements, laboratory results, and nursing records. All datasets were de-identified to comply with the Safe Harbor provisions of the Health Insurance Portability and Accountability Act (HIPAA). The study author (Qiqiang Liang) gained access to the datasets through an approved application (certification number 64964465). The inclusion criteria is these patients who met the criteria for sepsis 3.0 and the AKI KDIGO stage II/III diagnostic criteria. The exclusion criteria comprised patients who had severe AKI before the diagnosis of sepsis, patients diagnosed with AKI more than 7 days after the diagnosis of sepsis, all patients with chronic kidney disease (CKD) stages III to V, including basal creatinine values more than 256umol/L and kidney transplantation, children under 16 years of age, and pregnant women. Subsequently, patients with SA-AKI were stratified based on whether they received RRT, using the MIMIC database as the training and validation set. Models were created using multivariate logistic regression and three machine learning algorithms, and the eICU database was used as an external test set for model performance analysis. Study definition All patients underwent screening for suspected sites of infection and calculation of the Sequential Organ Failure Assessment (SOFA) score. A diagnosis of sepsis was confirmed in case of an infection and an increase of 2 points or more in the SOFA score 2 . Severe AKI was defined according to KDIGO criteria, with an increase in serum creatinine to 2.0 times the baseline value or a urine output of <0.5 mL/kg/hour for ≥12 hours, as detailed in our previous study protocol 11,12 . Referring to the large randomized controlled trial (RCT) STARRT-AKI, patients with KDIGO stage II/III AKI were defined as having severe AKI 7 . SA-AKI was defined according to the Acute Dialysis Quality Initiative (ADQI) 28 working group's definition as AKI occurring within 7 days of sepsis diagnosis 3 . Suspected sites of infection were determined based on the type of cultures taken before the diagnosis of sepsis, prioritizing culture-positive sites, followed by sterile site cultures. Variables collection The incorporated variables included demographic characteristics such as age, gender, race, weight, height, body mass index (BMI), discharge status, admission and discharge times, ICU admission and discharge times. In addition, laboratory parameters within 5 days before the diagnosis of severe AKI were analyzed, including complete blood count, liver and renal function, blood glucose, arterial blood gas analysis, and other relevant data, selecting maximum, minimum, or both values based on clinical significance. For example, hemoglobin would opt for the maximum value, whereas platelets would select the minimum value. White blood cells, due to their clinical relevance in both elevation and reduction, would adopt both the maximum and minimum values. Advanced life support records, including mechanical ventilation and RRT, were also collected. The severity scores included the SOFA score and the Acute Physiology Score (APS). In terms of comorbidities, the Charlson Comorbidity Index (CCI) and data such as diabetes, cerebral infarction, chronic renal failure, cirrhosis, malignancy, chronic heart failure, and structural lung disease were retrieved. Moreover, suspected sources of infection, such as blood, urine, sputum, and others, the time of infection, and the culture results were collected. Statistics Statistical Tools R (version 3.5.3, St. Louis, USA), RStudio (version 1.2.1335, Boston, USA), and related R packages for analysis, primarily including “dplyr”, “mice”, “tidymodels”, “XGBoost”, “SHAPforXGBoost”, “caret”, “MatchIt”, “survival”, “survminer”, “ggplot2”, “pROC“, “forestplot“, “table1“, “rattle“, etc. Data Cleaning Extreme outliers in the data distribution were assigned missing values. Variables with more than 40% missing values were excluded, and cases with variables missing over 50% were removed. We addressed missing values using multiple imputation, with the number of imputations times to 20 with linear regression model, ultimately leading to the construction of five imputed datasets. These datasets were then pooled to generate the final dataset for model analysis by Eestimating calibration slope using linearly predicted regression coefficients by Rubin rules. The extent of missingness prior to multiple imputation is presented in Supple Figure 1。 Model Construction Patients were categorized into the RRT group and no RRT groups based on whether they received RRT. In univariate analysis, continuous variables were analyzed using independent samples t-tests and rank-sum tests based on their distribution, and binary variables were analyzed using chi-square tests. Data were randomly split into training, validation, and internal test sets at 70:15:15 ratio. RRT initiation prediction models were constructed using multivariate logistic regression, random forest, SVM, and XGBoost models. We screened the included variables, and conducted variable exploration and dimensionality reduction via principal component analysis. The multivariate logistic regression model employed a stepwise backward elimination method for variable selection, which was visualized by nomograms. Random forest is a comprehensive decision of hundreds of decision trees, and each decision tree is independent of each other. The accuracy is higher than the decision tree. Five hundred trees were constructed and the exhaustive method was used to adjust the parameters in the random forest. The XGBoost model was set with a decision tree depth of 10, a learning rate of 0.5, and 500 sampling iterations. The SHAP (SHapley Additive exPlanations) method was used for global and local interpretation and visualization of the XGBoost model. Considering the small proportion of RRT patients, ten-fold cross-validation was employed to control for overfitting. The eICU database was used as an external test set to assess model stability. The receiver operating characteristic (ROC) curve, Area Under the Precision-Recall Curve (AUPRC) , and Decision Curve Analysis (DCA) were plotted to evaluate predictive performance and assess accuracy, precision, sensitivity, specificity, recall, and F1 score. This machine learning modeling strategy adhered to the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement recommendations 13 . Results Based on the inclusion and exclusion criteria set for the study, a total of 22,220 SA-AKI patients were included for analysis. Among them, 18,873 SA-AKI patients were included from the MIMIC database, while 3,347 patients were retrieved from the eICU-CRD database. Specifically, 1,358 patients (6.11%) received RRT, comprising 905 patients from the MIMIC database and 453 from the eICU-CRD database. The detailed screening process for this study is shown in Figure 1. The univariate analysis of basic demographic data is detailed in Table 1 and in supple table of two database respectively. In this study cohort, a higher proportion of male patients with severe SA-AKI required RRT, and there were racial disparities observed, with Caucasians having a lower incidence of RRT utilization compared to African Americans and Hispanics. Among the suspected sources of infection, bloodstream infection and pulmonary infection were the most common, accounting for 36.7% and 30.3%, respectively, with an overall positive culture rate of 24.3%. The proportion of bloodstream infections in SA-AKI patients receiving RRT was higher (42.2% vs. 36.3%, P<0.001), whereas the proportions of pulmonary and urinary tract infections were lower. Correspondingly, the SOFA scores (6.92 vs. 4.42, P<0.001) and APS scores (78.7 vs. 52.9, P<0.001) were significantly higher. Compared to patients not receiving RRT, mechanical ventilation was significantly more prevalent in the RRT group (69.3% vs. 60.4%, P<0.001). Moreover, the RRT group exhibited more severe metabolic acidosis, including lower PH (7.23 vs. 7.31, P<0.001) , lower buffer excess(-8.25 vs. -3.10, P<0.001), higher lactate levels (5.07 vs. 3.03, P<0.001), more disordered electrolytes, such as serum sodium and serum calcium and poorer coagulation function, including higher INR (2.20 vs. 1.73, P<0.001) and APTT (70.9 vs. 53.4, P<0.001). it was showed by principal component analysis in Supple Figure 2. The univariate analysis of baseline distribution for the MIMIC and eICU databases can be referenced in Supple Table. Multivariate logistic regression models and three machine learning models were constructed and tested, employing ten-fold cross-validation to reduce overfitting. The multivariate logistic regression model showed significant differences for many variables based on principal component analysis. Younger, higher body weight, more severe scoring upon admission to ICU, worse homeostasis and electrolyte balance as well as coagulation function, and mechanical ventilation before the diagnosis, were the high-risk factors of RRT for severe SA-AKI patients as shown in Table 2 and the nomogram in Figure 2. Using the XGBoost algorithm for variable analysis, the six most important parameters were the APS score, SOFA score, minimum calcium level, maximum APTT, maximum BUN, and minimum buffer excess. These parameters all have clinical interpretability and overlaps with the risk factors of the multiple logistic regression model. The importance of the model is detailed in Figure 3. Overall, the XGBoost model outperformed the multivariate logistic regression model and random forest or SVM models. In the internal test of MIMIC database and external test sets from the eICUdatabase, the XGBoost model's AUROC curves were 0.906 and 0.816, respectively, demonstrating superior accuracy and specificity. Random Forest Model has excellent performance slightly above XGBoost. The AUROC curves for the multivariate logistic regression model were 0.874 and 0.740, respectively. The performance of SVM was similar to the multivariate logistic regression model. Given the relatively low prevalence of patients requiring RRT interventions, we opted to compute the AUPRC and DCA as a measure of model performance. Notwithstanding a discernible decline in AUPRC for both models during the external test phase conducted with the eICU database, the XGBoost model retained a commendably high degree of accuracy. Detailed diagnostic parameters are shown in Table 3, while the AUROC and AUPRC are illustrated in Figure 4. DCA shown in the supple Figure 4. In addition, the SHAP algorithm was used to visualize the XGBoost model variables, and the SHAP value distribution for these six variables is shown in Supple Figure 3. The distribution heat map for the top 15 important variables in the XGBoost model is shown in the bee-swarm plot in Figure 5. For any given patient, the input variable values can be used to calculate a score, achieving a visualized SHAP value similar to the nomogram, as detailed in the waterfall plot in Figure 6. Discussion The precision medicine approach for AKI has been a long-standing vision, with a particular emphasis on determining which AKI patients would benefit from RRT 14 . Despite the identification of numerous high-risk factors for RRT in AKI patients through multivariate logistic regression models 15 , 16 , the gap between these isolated risk factors and the precision required for accurate prediction remains significant. The emergence of machine learning has opened up new avenues for achieving precision in the treatment of AKI patients 17 , 18 . The present study demonstrates that machine learning can accurately predict the demand of RRT when patients was diagnosed as severe SA-AKI . A growing number of studies have investigated the application of artificial intelligence models for AKI diagnosis and early detection, as well as for RRT initiation and weaning 19 . Using local databases and public databases, our research team constructed models for the diagnosis and early detection of severe AKI, as well as RRT weaning models, achieving good performance with both Random Forest and XGBoost 11 , 20 . Moreover, with the development of SHAP, the explainability of machine learning has greatly improved. Zhiyan Fan et al. have developed a robust prognostic prediction model for SA-AKI patients, which exhibits consistent performance at 7 days, 14 days, and 28 days post-admission, was constructed by integrating data from the MIMIC-IV database with a local database. The model was built using an ensemble of 40 predictive variables, processed through the advanced XGBoost machine learning algorithm 21 . Gao et al. corroborated these findings by employing the random forest algorithm to analyze SA-AKI patients sourced from the MIMIC-IV database. Their analysis yielded analogous outcomes, further validating the efficacy of machine learning techniques in predicting the prognosis of SA-AKI patients within the critical care setting 22 . Nonetheless, studies on the prediction of RRT in SA-AKI are relatively scarce. Li Zhao et al. collected data from local sepsis patients over three years, including 300 sepsis patients for model construction and validation. The AUROCs for the training and validation cohorts were 0.873 (95% CI 0.825–0.921) and 0.826 (95% CI 0.727–0.924), respectively 23 . Chun-Fu Lai et al. conducted a prospective observational cohort study, collecting data over 10 years from a total of 1000 local SA-AKI patients requiring RRT. Three sub-phenotypes were constructed using unsupervised clustering and multivariable-adjusted Cox regression models and Fine-Gray sub-distribution hazard models were successfully established, identifying pre-dialysis hyperlactatemia ≥ 3.3 mmol/L as an independent outcome predictor 24 . The utilization rate of RRT for severe SA-AKI remains low, with only 6.3% of patients receiving RRT in the MIMIC and eICU databases. One of the reasons is the exclusion of CKD stages III to V and renal transplant recipients. The second reason is that, based on the findings of recent landmark randomized controlled trials (RCTs), the therapeutic approach to RRT in AKI patients has become more conservative. The IDEAL-ICU study, focusing on patients with septic shock, randomized patients diagnosed with acute kidney injury using RIFLE criteria to immediate initiation of RRT or delayed initiation within 48 hours if needed. All 488 enrolled patients had septic shock, but the study was prematurely halted due to futility, finding no difference in 90-day mortality between the early group (58%) and the delayed strategy group (54%). A meta-analysis by Pan H et al., which included 10 RCTs involving 4753 patients with severe AKI, reported no significant difference in mortality between the early RRT group and the standard group 25 . Additionally, 38% of the delayed group did not require RRT as their renal function spontaneously recovered 8 . White KA-O et al. conducted a large retrospective study in Australia, including 13,451 SA-AKI patients with 4,051 cases (30%) of KDIGO II/III stages, but only 1,135 patients (8.4%) received RRT 26 . The low rate of RRT initiation suggests that the majority of patients do not require RRT and can recover through conservative treatment. Identifying SA-AKI patients who do not require RRT and allowing for a longer observation period could potentially avoid the high risks and costs associated with RRT 3 . The CRTSAKI study is an ongoing large-scale, multicenter, randomized controlled trial related to sepsis-associated AKI, aiming to enroll 460 patients with sepsis-associated AKI (SA-AKI) at KDIGO II/III stages. The protocol stipulates immediate RRT initiation in the early group post-randomization, while the delayed group follows a standard approach similar to the STARRT-AKI study. The primary outcome is overall survival within a 90-day follow-up period (90-day all-cause mortality) (Trial registration: NCT03175328) 27 . This study is expected to provide further insights into the timing of RRT. The current body of evidence from RCTs suggests that the timing of initiating RRT in patients with severe AKI may not significantly influence clinical outcomes. This apparent lack of effect could be attributed to the inclusion of a heterogeneous patient population in these studies, where a significant proportion of patients may not actually require RRT. The dilution of the treatment effect due to the inclusion of unnecessary RRT recipients can lead to a larger negative control group, which in turn reduces the statistical power to detect a true positive effect. To address this issue, the application of AI models offers a promising solution. AI algorithms can be trained to predict which patients are most likely to benefit from RRT, thereby enabling a more targeted approach to clinical trial design. This targeted approach not only enhances the statistical power of the study but also ensures that resources are allocated efficiently, potentially leading to more personalized and effective treatment strategies for patients with severe AKI. Our study confronted the ubiquitous challenges prevalent in the field of artificial intelligence research, particularly concerning the explainability of models. Despite the advances in the model's decision-making process since the introduction of the SHAP algorithm, machine learning models remain inherently "black box" systems. Their inner workings can be opaque in certain scenarios, limiting our confidence in the model's predictive outcomes and the acceptance of its clinical applications. The eICU database was utilized as an external test set to validate the model's generalizability, yet the AUROC dropped from 91% in the training set to 81% in the test set. This discrepancy suggests variations in model performance across different datasets, potentially due to disparities in patient characteristics, healthcare settings, or data collection methods between the two databases, thereby constraining the model's adaptability. The reasons for patients to use RRT are unavailable, and doctors from different centers may use RRT based on different clinical experiences and evidence, resulting in the occurrence of offsets. Lastly, to ensure data quality and analytical accuracy, unmeasured or confounding variables were removed, potentially introducing bias in the exposure-outcome association. Hence, the model may not have fully captured all factors influencing patient outcomes, impacting the accuracy and reliability of the study results. The model's limitations highlight the ongoing need for refinement in balancing predictive power with interpretability, particularly in the context of clinical decision-making. Conclusion Our study constructed a predictive model for screen RRT patients with severe SA-AKI using two-center databases and the XGBoost algorithm. We enhanced model interpretability by employing the SHAP method and externally validated the model using the eICU database. Abbreviations RRT: Renal Replacement Therapy SA-AKI: Sepsis-associated Acute Kidney Injury SCR: Serum Creatinine LOS: Length of Stay KDIGO: Kidney Disease: Improving Global Outcomes AUROC: Area Under the Receiver Operating Characteristic AUPRC: Area Under the Precision-Recall Curve TRIPOD：Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis PSM: Propensity Score Matching IPW: Inverse Probability Weighting CCI: Charlson Comorbidity Index; SOFA: Sequential Organ Failure Assessment APS: Acute Physiology Score BMI: Body Mass Index DCA: Decision Curve Analysis Declarations Clinical trial number: not applicable Ethics approval and consent to participate: This study is a database retrospective study, no private data was used, and no ethical approval was required. The two public databases have passed the ethics of their respective research institutions. Consent for publication: Not applicable. Availability of data and materials: The datasets generated and/or analyzed during the current study are available in the MIMIC and eICU-CRD databases. Competing interests: None. Funding : This work was supported by grants from the peak supporting clinical discipline of Shanghai health bureau (2023ZDFC0104 to L.T) Author contributions: Qiqiang Liang designed the research scheme, analyzed the characteristics of the data, constructed the prediction model, and drafted the manuscript. Sumian Zhang extracted and collated clinical data through the database, built models, and collected prospective research data. Haiyan Ye and Mei yang determined clinical variables, analyzed the importance of model variables, and completed clinical data in prospective studies. Xuebin Wang supervised the research process, provided clinical reference, and revised the paper. All authors read and approved the final manuscript. Acknowledgements: We thank the Ascetic Practitioners in Critical Care (APCC) team, and the easy Data Science for Medicine (easyDSM) team for sharing their knowledge and codes in big data of critical care, along with the cross-platform Big Data Master of Critical Care (BDMCC) software (https://github.com/ningyile/BDMCC_APP). We especially appreciate the MIMIC and eICU-CRD official team's efforts to open-source the database and codes. References Poston JT, Koyner JL. Sepsis associated acute kidney injury. Bmj. 2019;364:k4891. Singer M, Deutschman CS, Seymour CW, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). Jama. 2016;315(8):801-810. Zarbock A, Nadim MK, Pickkers P, et al. Sepsis-associated acute kidney injury: consensus report of the 28th Acute Disease Quality Initiative workgroup. Nature Reviews Nephrology. 2023;19(6):401-417. Khwaja A. KDIGO clinical practice guidelines for acute kidney injury. Nephron Clin Pract. 2012;120(4):c179-184. Pais T, Jorge S, Lopes JA. Acute Kidney Injury in Sepsis. International journal of molecular sciences. 2024;25(11). Zarbock A, Kellum JA, Schmidt C, et al. Effect of Early vs Delayed Initiation of Renal Replacement Therapy on Mortality in Critically Ill Patients With Acute Kidney Injury: The ELAIN Randomized Clinical Trial. JAMA. (315(20):2190-9). Bagshaw SM, Wald R, Adhikari NKJ, et al. Timing of Initiation of Renal-Replacement Therapy in Acute Kidney Injury. N Engl J Med. 2020;383(3):240-251. Barbar SD, Clere-Jehl R, Bourredjem A, et al. Timing of Renal-Replacement Therapy in Patients with Acute Kidney Injury and Sepsis. N Engl J Med. 2018;379(15):1431-1442. Li G, Li B, Song B, et al. Uplift modeling to predict individual treatment effects of renal replacement therapy in sepsis-associated acute kidney injury patients. (2045-2322 (Electronic)). Palmowski L, Lindau S, Henk LC, et al. Predictive enrichment for the need of renal replacement in sepsis-associated acute kidney injury: combination of furosemide stress test and urinary biomarkers TIMP-2 and IGFBP-7. (2110-5820 (Print)). Liang Q, Xu Y, Zhou Y, Chen X, Chen J, Huang M. Severe acute kidney injury predicting model based on transcontinental databases: a single-centre prospective study. BMJ Open. 2022;12(3):e054092. Liang Q, Ding S, Chen J, et al. Prediction of carbapenem-resistant gram-negative bacterial bloodstream infection in intensive care unit based on machine learning. BMC medical informatics and decision making. 2024;24(1):123. Moons Kg Fau - Altman DG, Altman Dg Fau - Reitsma JB, Reitsma Jb Fau - Ioannidis JPA, et al. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration. (1539-3704 (Electronic)). Meersch M, Mayerhöfer T, Joannidis M. Acute kidney injury subphenotyping and personalized medicine. Curr Opin Crit Care. 2024;30(6):555-562. Chen JJ, Chang CH, Huang YT, Kuo G. Furosemide stress test as a predictive marker of acute kidney injury progression or renal replacement therapy: a systemic review and meta-analysis. Crit Care. 2020;24(1):202. Romagnoli S, Clark WR, Ricci Z, Ronco C. Renal replacement therapy for AKI: When? How much? When to stop? Best practice & research. Clinical anaesthesiology. 2017;31(3):371-385. Wilson FP, Greenberg JH. Acute Kidney Injury in Real Time: Prediction, Alerts, and Clinical Decision Support. Nephron. 2018;140(2):116-119. Gottlieb ER, Samuel M, Bonventre JV, Celi LA, Mattie H. Machine Learning for Acute Kidney Injury Prediction in the Intensive Care Unit. Advances in Chronic Kidney Disease. 2022;29(5):431-438. Cheungpasitporn W, Thongprayoon C, Kashani KB. Advances in critical care nephrology through artificial intelligence. (1531-7072 (Electronic)). Liang Q, Xu X, Ding S, Wu J, Huang M. Prediction of successful weaning from renal replacement therapy in critically ill patients based on machine learning. Ren Fail. 2024;46(1):2319329. Gao T, Nong Z, Luo Y, et al. Machine learning-based prediction of in-hospital mortality for critically ill patients with sepsis-associated acute kidney injury. (1525-6049 (Electronic)). Fan Z, Jiang J, Xiao C, et al. Construction and validation of prognostic models in critically Ill patients with sepsis-associated acute kidney injury: interpretable machine learning approach. (1479-5876 (Electronic)). Li X, Liu C, Mao Z, Li Q, Zhou FA-OX. Timing of renal replacement therapy initiation for acute kidney injury in critically ill patients: a systematic review of randomized clinical trials with meta-analysis and trial sequential analysis. (1466-609X (Electronic)). Lai CA-O, Liu JH, Tseng LJ, et al. Unsupervised clustering identifies sub-phenotypes and reveals novel outcome predictors in patients with dialysis-requiring sepsis-associated acute kidney injury. (1365-2060 (Electronic)). Pan HC, Chen YY, Tsai IJ, et al. Accelerated versus standard initiation of renal replacement therapy for critically ill patients with acute kidney injury: a systematic review and meta-analysis of RCT studies. (1466-609X (Electronic)). White KA-O, Serpa-Neto A, Hurford R, et al. Sepsis-associated acute kidney injury in the intensive care unit: incidence, patient characteristics, timing, trajectory, treatment, and associated outcomes. A multicenter, observational study. Intensive Care Med. (1432-1238 (Electronic)). Chen WA-O, Cai LH, Zhang ZH, et al. The timing of continuous renal replacement therapy initiation in sepsis-associated acute kidney injury in the intensive care unit: the CRTSAKI Study (Continuous RRT Timing in Sepsis-associated AKI in ICU): study protocol for a multicentre, randomised controlled trial. (2044-6055 (Electronic)). Tables Table 1: Basic demographic characteristics of severe acute kidney injury associated with sepsis No RRT, N=20862 RRT, N=1358 P Gender (Male) 11299 (54.2%) 796 (58.6%) 0.00155 Race White 14964 (71.7%) 885 (65.2%) <0.001 Black 1659 (8.0%) 126 (9.3%) 0.0909 Latin 668 (3.2%) 56 (4.1%) 0.0759 Others 3571 (17.1%) 291 (21.4%) <0.001 Suspicious infection site Blood 7571 (36.3%) 573 (42.2%) <0.001 Lung 6409 (30.7%) 318 (23.4%) <0.001 Urine 4571 (21.9%) 199 (14.7%) <0.001 Others 2042 (9.8%) 200 (14.7%) <0.001 Postive Culture 5149 (24.7%) 235 (17.3%) <0.001 Comorbidity Cirrhosis 1425 (6.8%) 249 (18.3%) <0.001 Chronic heart failure 5578 (26.7%) 285 (21.0%) <0.001 Myocardial infarction 3068 (14.7%) 178 (13.1%) 0.115 Chronic structural lung disease 5143 (24.7%) 249 (18.3%) <0.001 Malignant tumour 2626 (12.6%) 126 (9.3%) <0.001 Diabetes mellitus 5507 (26.4%) 383 (28.2%) 0.153 CCI (mean, SD) 4.76 (2.56) 4.28 (2.40) <0.001 SOFA score (mean, SD) 4.42 (2.82) 6.92 (3.87) <0.001 APS (mean, SD) 52.9 (22.1) 78.7 (26.7) <0.001 Time of admission ICU to AKI (minutes, mean, SD) 4220 (10900) 5440 (11100) <0.001 Age (Y, mean, SD) 64.1 (16.0) 58.4 (15.2) <0.001 Height (cm, mean, SD) 169 (10.7) 170 (10.5) <0.001 Weight (Kg, mean, SD) 84.6 (25.8) 92.1 (28.0) <0.001 BMI (mean, SD) 29.5 (8.29) 31.7 (8.95) <0.001 WBCmin (*10^9, mean, SD) 11.3 (7.69) 14.0 (11.0) <0.001 WBCmax (*10^9, mean, SD) 15.7 (9.60) 16.8 (14.0) 0.00373 Hbmin (g/l, mean, SD) 10.0 (2.33) 9.55 (2.52) <0.001 Pltmax (*10^9, mean, SD) 193 (116) 156 (114) <0.001 HCTmin (%, mean, SD) 30.1 (7.01) 29.0 (7.79) <0.001 Nemax (%, mean, SD) 80.5 (13.5) 79.8 (15.5) 0.116 PHmax (mean, SD) 7.41 (0.0814) 7.36 (0.116) <0.001 PHmin (mean, SD) 7.31 (0.109) 7.23 (0.135) <0.001 PaO2min (mmHg, mean, SD) 88.3 (59.6) 79.0 (52.2) <0.001 PaCO2max (mmHg, mean, SD) 48.6 (14.9) 48.1 (16.4) 0.215 BEmax (mmol/L, mean, SD) 0.678 (5.31) -3.05 (6.89) <0.001 BEmin (mmol/L, mean, SD) -3.10 (5.78) -8.25 (7.12) <0.001 HCO3max (mmol/L, mean, SD) 25.3 (5.43) 21.6 (6.28) <0.001 HCO3min (mmol/L, mean, SD) 21.7 (5.13) 17.8 (5.79) <0.001 Lactatemax (mmol/L, mean, SD) 3.03 (2.39) 5.07 (4.25) <0.001 Kmax (mmol/L, mean, SD) 4.44 (0.950) 4.85 (1.13) <0.001 CLmax (mmol/L, mean, SD) 107 (7.05) 105 (8.31) <0.001 CLmin (mmol/L, mean, SD) 101 (6.81) 99.7 (7.80) <0.001 NAmax (mmol/L, mean, SD) 140 (6.01) 139 (7.07) <0.001 NAmin (mmol/L, mean, SD) 136 (5.85) 135 (6.48) <0.001 Camin (mmol/L, mean, SD) 1.04 (0.119) 0.89 (0.140) <0.001 Glu (mg/dL, mean, SD) 187 (114) 206 (130) <0.001 ALP (U/L, mean, SD) 128 (146) 153 (174) <0.001 Tbil (mean, SD) 1.91 (4.23) 5.00 (9.10) <0.001 ALTmax (U/L, mean, SD) 145 (615) 401 (1180) <0.001 ASTmax (U/L, mean, SD) 215 (866) 745 (1960) <0.001 BUNmax (mg/dL, mean, SD) 31.7 (23.1) 53.6 (36.1) <0.001 INRmax (mean, SD) 1.73 (1.42) 2.20 (1.64) <0.001 APTTmax (s, mean, SD) 53.4 (37.5) 70.9 (44.1) <0.001 Mechanical ventilation 12598 (60.4%) 941 (69.3%) <0.001 ICU death 3589 (17.2%) 589 (43.4%) <0.001 All death 8902 (42.7%) 749 (55.2%) <0.001 Abbreviation: CCI: Charlson Comorbidity Index; Hb: Hemoglobin; SOFA: Sequential Organ Failure Assessment; APS: Acute Physiology Score; BMI: Body Mass Index; Plt: Platelet; HCT: Hematocrit; Ne: Neutrophil Proportion; BE: Buffuer excess; K: Potassium; Cl: Chlorine; NA: Natrium; Ca: Calcium; Glu: Glucose; ALP: Alkaline Phosphatase; Tbil: Total Bilirubin; ALT: Alanine Aminotransferase; AST: Aspertate Aminotransferase; BUN: Blood Urea Nitrogen; INR: International Normalized Ratio for Prothrombin Time; APTT: Activated Partial Prothrombin Time. Table 2: Parameters in the multivariable logistic regression model Value P OR 2.5% CI 97.5% CI Age (Y, mean, SD) <0.001 0.67 0.60 0.75 Weight (Kg, mean, SD) <0.001 1.30 1.20 1.41 SOFA score (mean, SD) <0.001 1.49 1.36 1.63 APS (mean, SD) <0.001 2.14 1.95 2.35 Hb (g/l, mean, SD) <0.001 0.73 0.66 0.81 HCO3max (mmol/L, mean, SD) <0.001 0.40 0.35 0.44 Kmax (mmol/L, mean, SD) <0.001 1.25 1.16 1.36 CLmax (mmol/L, mean, SD) <0.001 0.64 0.56 0.74 CLmin (mmol/L, mean, SD) <0.001 0.74 0.66 0.83 NAmax (mmol/L, mean, SD) <0.001 1.52 1.38 1.68 APTT (s, mean, SD) <0.001 1.53 1.44 1.62 Mechanical ventilation <0.001 1.52 1.37 1.63 Abbreviation: SOFA: Sequential Organ Failure Assessment; APS: Acute Physiology Score; Hb: Hemoglobin; K: Potassium; Cl: Chlorine; NA: Natrium; Ca: Calcium;Glu: Glucose; APTT: Activated Partial Prothrombin Time. Table 3: Model prediction performance evaluation for RRT in severe SA-AKI patients AUROC 95%CI Accuracy Sensitivity Specificity PPV NPV Precision Recall F1 Youden Glm Internal test 0.874 83.84%-90.91% 75.9 87.0 75.4 14.6 99.2 14.6 87.0 0.3 62.4 XGBoost Internal test 0.906 87.72%-93.4% 79.0 90.6 78.4 17.8 99.4 17.8 90.6 0.3 69.0 Random Forest Internal test 0.905 86.54%-94.1% 82.4 67.7 83.1 15.8 98.2 15.8 67.7 0.25 50.8 SVM Internal test 0.862 82.53%-92.47% 79.2 83.4 79.0 15.7 99.0 15.7 83.4 0.3 62.4 Glm External test 0.740 71.45%-76.64% 77.9 56.3 81.3 32.0 92.2 32.0 56.3 0.4 37.6 XGBoost External test 0.816 79.63%-83.65% 69.6 78.8 68.2 27.7 95.4 27.7 78.8 0.4 47.0 Random Forest External test 0.814 78.54%-83.93% 87.1 72.6 67.4 26.8 87.4 26.8 72.6 0.4 40 SVM External test 0.775 74.18%-82.39% 78.6 57.1 83.4 28.7 95.6 28.7 57.1 0.4 40.5 Additional Declarations No competing interests reported. Supplementary Files Supplefile.pdf Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6321504","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":451486028,"identity":"1b8ebfb9-04fc-4fcd-8285-b53222d92cd4","order_by":0,"name":"Qiqiang Liang","email":"","orcid":"","institution":"Tongji University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Qiqiang","middleName":"","lastName":"Liang","suffix":""},{"id":451486029,"identity":"0335fc52-45ed-4a61-b704-f8aa33989504","order_by":1,"name":"Sumian Zhang","email":"","orcid":"","institution":"Tongji University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Sumian","middleName":"","lastName":"Zhang","suffix":""},{"id":451486030,"identity":"368e7230-87b7-4f95-be51-3c84d7b5094d","order_by":2,"name":"Haiyan Ye","email":"","orcid":"","institution":"Tongji University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Haiyan","middleName":"","lastName":"Ye","suffix":""},{"id":451486031,"identity":"7c277e15-4921-4028-8759-28e51e7d5dfd","order_by":3,"name":"Mei Yang","email":"","orcid":"","institution":"Tongji University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Mei","middleName":"","lastName":"Yang","suffix":""},{"id":451486032,"identity":"b31b1127-49c8-474e-8c5b-f8389da9e22c","order_by":4,"name":"Xuebin Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA60lEQVRIiWNgGAWjYBACNoaDDQcSKtjkwDwJAyBxgIAWPsbDBx98OMNnzEO0FjnmY8mGM9vkEnvgQoS0sLGdMZPmbTNL38/ee/iFRQGDHN+NBMbPBfi08AC18JxLy+0BEhZAhxlL3khglp6BT4sESEvZsdweiRwzA6CWxA03EtiYefBpkX8D1ML2P50HqqWesBYGoPdntLElALUYPwBqSTAgrAUcyGyGPWfOmAEDWcJw5pmHzdL4tMg3QKJSnr29x/izxB8beb7jyQc/49OCYqO0BIMEkGZsIFIDAwPzxw9Eqx0Fo2AUjIKRBABjUkmc/c9nagAAAABJRU5ErkJggg==","orcid":"","institution":"Tongji University School of Medicine","correspondingAuthor":true,"prefix":"","firstName":"Xuebin","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2025-03-27 14:38:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6321504/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6321504/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":82273456,"identity":"5158f24d-a0c9-4149-926a-c981b6decd6b","added_by":"auto","created_at":"2025-05-08 14:29:04","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":249078,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of the study process in this research.\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6321504/v1/60a901dc3ebb79e6576f0229.jpeg"},{"id":82275063,"identity":"ebf49d49-2748-4be5-bf8d-e4e2c09d40b0","added_by":"auto","created_at":"2025-05-08 14:37:04","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":161794,"visible":true,"origin":"","legend":"\u003cp\u003eForest plot and nomogram of risk factors from the multivariate logistic regression model for SA-AKI patients undergoing RRT, including laboratory indicators and severity scores.\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6321504/v1/70a27f6b331be277bb072f0f.jpeg"},{"id":82275065,"identity":"ccbefe0d-759f-4de1-ad41-1155f6ada425","added_by":"auto","created_at":"2025-05-08 14:37:04","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":71849,"visible":true,"origin":"","legend":"\u003cp\u003eThe importance of the XGBoost model of RRT in SA-AKI patients.\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6321504/v1/809010a7382e5ffd620ba076.jpeg"},{"id":82275882,"identity":"517fcadb-9e25-46a1-829b-64a01c6c808e","added_by":"auto","created_at":"2025-05-08 14:45:05","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":236533,"visible":true,"origin":"","legend":"\u003cp\u003eBee-swarm plot of the XGBoost model predicting RRT in SA-AKI patients, displaying the top 15 variables with high importance.\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6321504/v1/379a2e597a40ff9fbc20ae21.jpeg"},{"id":82273458,"identity":"cccf4041-377d-479e-93a5-92fc91c84c0a","added_by":"auto","created_at":"2025-05-08 14:29:04","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":95276,"visible":true,"origin":"","legend":"\u003cp\u003eWaterfall plot of the XGBoost model predicting RRT in severe SA-AKI patients. The SHAP values shown are the parameter results for a randomly selected patient in the study.\u003c/p\u003e","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6321504/v1/71585ae87c7e4b44447c1ea2.jpeg"},{"id":82273469,"identity":"c2176c52-63e9-4052-a8b2-e779d0e0c605","added_by":"auto","created_at":"2025-05-08 14:29:05","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":393291,"visible":true,"origin":"","legend":"\u003cp\u003eAUROC and AUPRC of the XGBoost and multivariate logistic regression models in predicting severe SA-AKI patients undergoing RRT in the internal and external test set.\u003c/p\u003e","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6321504/v1/64504261e0dc3689b108dfca.jpeg"},{"id":85920407,"identity":"02963408-5692-4e8b-943a-8491e0499895","added_by":"auto","created_at":"2025-07-03 07:47:20","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2662151,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6321504/v1/580cebcc-b9c9-4212-a2ea-52715df8e980.pdf"},{"id":82273457,"identity":"3646b1db-4d1f-46e6-a920-d45e2b922cba","added_by":"auto","created_at":"2025-05-08 14:29:04","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":955361,"visible":true,"origin":"","legend":"","description":"","filename":"Supplefile.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6321504/v1/c15f37bc3564896e2ef425d2.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine Learning for Renal Replacement Therapy in Patients with Severe Acute Kidney Injury Associated with Sepsis","fulltext":[{"header":"Background","content":"\u003cp\u003eAcute kidney injury (AKI) being one of the most common organ dysfunctions of sepsis\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Up to 60% of sepsis patients develop AKI, but the pathophysiology remains incompletely understood\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Sepsis involves a harmful inflammatory cascade that contributes to AKI. However, AKI also results from factors such as hypovolemia, nephrotoxic antibiotics, and complex urinary tract infections; hence, AKI cannot be solely attributed to sepsis. Sepsis-associated acute kidney injury (SA-AKI) has long been recognized, but a precise definition was only established in 2023. The Acute Disease Quality Initiative (ADQI) 28 Workgroup defined SA-AKI as AKI occurring within 7 days of sepsis diagnosis\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. The rationale for the suggested 7-day is based on observations of sepsis cases with AKI occurring within a few days of the onset of sepsis, whereas AKI occurring after a week may not be directly related to the initial sepsis\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eAKI is categorized into three stages according to the Kidney Disease: Improving Global Outcomes (KDIGO) criteria, which are based on the baseline and changes in serum creatinine levels, as well as urine output. KDIGO stage II/III is often defined as severe AKI\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. In terms of disease progression, the majority of patients recover from mild AKI, which corresponds to KDIGO stage I; in contrast, only a small proportion of patients progress to severe AKI. Patients with severe AKI have a higher risk of requiring renal replacement therapy (RRT). RRT is a critical component in the management of SA-AKI patients, providing support in maintaining fluid balance, removing inflammatory mediators and toxins, and supporting renal function recovery. The emergency indications for initiating RRT for SA-AKI are similar to other types of AKI. However, the timing of initiating RRT remains a clinical challenge and a controversial topic in the absence of emergency indications\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. The ELAIN study concluded that early initiation of RRT in patients with severe AKI reduced 90-day mortality and length of hospital stay\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. However, some high-quality RCTs showed conflicting results, including the recent STARRT-AKI and IDEAL-ICU studies\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. Despite these studies employing different AKI diagnostic criteria and varying definitions of RRT timing, none found a difference in 90-day survival rates between the early and delayed groups. The debate over the timing of RRT initiation in patients with severe AKI reflects the impracticality of a one-size-fits-all strategy. Consequently, scholars have proposed strategies such as dynamic renal function assessment, focusing on renal demand-capacity matching and dynamic evaluation of renal biomarkers, to closely monitor and decide on the timing of RRT\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eWith the advances in artificial intelligence, a growing number of scholars have employed machine learning to construct powerful models to assist in identifying patients with AKI who require RRT. Li G et al. analyzed data from 8289 patients from the MIMIC-III database, among whom 591 received RRT and 7698 did not. The team developed a logistic regression model and plotted a nomogram to predict whether SA-AKI patients could benefit from RRT, determining that the class transition model is more effective in predicting individual treatment outcomes\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. Moreover, Palmowski L et al. conducted a prospective, multicenter study involving 99 patients, predicting whether SA-AKI patients would require RRT through the furosemide stress test (FST) and urinary biomarkers TIMP-2*IGFBP-7, achieving a prediction accuracy of 83%\u003csup\u003e10\u003c/sup\u003e. However, the data of the above models was limited to a single center and was not subjected to external validation, so the results cannot be extrapolated. The present study aims to use two databases combined with machine learning algorithms to construct a prediction model to screen patients needing RRT in severe SA-AKI cases.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003eStudy Design\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis two-center retrospective study collected data from two large, publicly available ICU databases: the MIMIC database and the eICU-CRD database. The MIMIC database includes MIMIC-III and MIMIC-IV databases, which encompass clinical data from over 200,000 patients at the Beth Israel Deaconess Medical Center in Boston, Massachusetts. The institutional review boards of the Massachusetts Institute of Technology (No. 0403000206) and Beth Israel Deaconess Medical Center (2001-P-001699/14) approved the use of this database for research purposes. The eICU database is composed of health data from over 200,000 ICU admissions in the United States between 2014 and 2015. Both databases integrate comprehensive clinical data, including demographics, hourly vital signs, clinical measurements, laboratory results, and nursing records. All datasets were de-identified to comply with the Safe Harbor provisions of the Health Insurance Portability and Accountability Act (HIPAA). The study author (Qiqiang Liang) gained access to the datasets through an approved application (certification number 64964465). The inclusion criteria is these patients who met the criteria for sepsis 3.0 and the AKI KDIGO stage II/III diagnostic criteria. The exclusion criteria comprised patients who had severe AKI before the diagnosis of sepsis, patients diagnosed with AKI more than 7 days after the diagnosis of sepsis, all patients with chronic kidney disease (CKD) stages III to V, including basal creatinine values more than 256umol/L and kidney transplantation, children under 16 years of age, and pregnant women. Subsequently, patients with SA-AKI were stratified based on whether they received RRT, using the MIMIC database as the training and validation set. Models were created using multivariate logistic regression and three machine learning algorithms, and the eICU database was used as an external test set for model performance analysis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStudy definition\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll patients underwent screening for suspected sites of infection and calculation of the Sequential Organ Failure Assessment (SOFA) score. A diagnosis of sepsis was confirmed in case of an infection and an increase of 2 points or more in the SOFA score\u003csup\u003e2\u003c/sup\u003e. Severe AKI was defined according to KDIGO criteria, with an increase in serum creatinine to 2.0 times the baseline value or a urine output of \u0026lt;0.5 mL/kg/hour for \u0026ge;12 hours, as detailed in our previous study protocol\u003csup\u003e11,12\u003c/sup\u003e. Referring to the large randomized controlled trial (RCT) STARRT-AKI, patients with KDIGO stage II/III AKI were defined as having severe AKI\u003csup\u003e7\u003c/sup\u003e. SA-AKI was defined according to the Acute Dialysis Quality Initiative (ADQI) 28 working group\u0026apos;s definition as AKI occurring within 7 days of sepsis diagnosis\u003csup\u003e3\u003c/sup\u003e. Suspected sites of infection were determined based on the type of cultures taken before the diagnosis of sepsis, prioritizing culture-positive sites, followed by sterile site cultures.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eVariables collection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe incorporated variables included demographic characteristics such as age, gender, race, weight, height, body mass index (BMI), discharge status, admission and discharge times, ICU admission and discharge times. In addition, laboratory parameters within 5 days before the diagnosis of severe AKI were analyzed, including complete blood count, liver and renal function, blood glucose, arterial blood gas analysis, and other relevant data, selecting maximum, minimum, or both values based on clinical significance.\u0026nbsp;For example, hemoglobin would opt for the maximum value, whereas platelets would select the minimum value. White blood cells, due to their clinical relevance in both elevation and reduction, would adopt both the maximum and minimum values. Advanced life support records, including mechanical ventilation and RRT, were also collected. The severity scores included the SOFA score and the Acute Physiology Score (APS). In terms of comorbidities, the Charlson Comorbidity Index (CCI) and data such as diabetes, cerebral infarction, chronic renal failure, cirrhosis, malignancy, chronic heart failure, and structural lung disease were retrieved. Moreover, suspected sources of infection, such as blood, urine, sputum, and others, the time of infection, and the culture results were collected.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical Tools\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eR (version 3.5.3, St. Louis, USA), RStudio (version 1.2.1335, Boston, USA), and related R packages for analysis, primarily including \u0026ldquo;dplyr\u0026rdquo;, \u0026ldquo;mice\u0026rdquo;, \u0026ldquo;tidymodels\u0026rdquo;, \u0026ldquo;XGBoost\u0026rdquo;, \u0026ldquo;SHAPforXGBoost\u0026rdquo;, \u0026ldquo;caret\u0026rdquo;, \u0026ldquo;MatchIt\u0026rdquo;, \u0026ldquo;survival\u0026rdquo;, \u0026ldquo;survminer\u0026rdquo;, \u0026ldquo;ggplot2\u0026rdquo;, \u0026ldquo;pROC\u0026ldquo;, \u0026ldquo;forestplot\u0026ldquo;, \u0026ldquo;table1\u0026ldquo;, \u0026ldquo;rattle\u0026ldquo;, etc.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Cleaning \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eExtreme outliers in the data distribution were assigned missing values. Variables with more than 40% missing values were excluded, and cases with variables missing over 50% were removed. We addressed missing values using multiple imputation, with the number of imputations times to 20 with linear regression model, ultimately leading to the construction of five imputed datasets. These datasets were then pooled to generate the final dataset for model analysis by Eestimating calibration slope using linearly predicted regression coefficients by Rubin rules. The extent of missingness prior to multiple imputation is presented in Supple Figure 1。\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel Construction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePatients were categorized into the RRT group and no RRT groups based on whether they received RRT. In univariate analysis, continuous variables were analyzed using independent samples t-tests and rank-sum tests based on their distribution, and binary variables were analyzed using chi-square tests. Data were randomly split into training, validation, and internal test sets at 70:15:15 ratio. RRT initiation prediction models were constructed using multivariate logistic regression, random forest, SVM, and XGBoost models. We screened the included variables, and conducted variable exploration and dimensionality reduction via principal component analysis. The multivariate logistic regression model employed a stepwise backward elimination method for variable selection, which was visualized by nomograms.\u0026nbsp;Random forest is a comprehensive decision of hundreds of decision trees, and each decision tree is independent of each other. The accuracy is higher than the decision tree. Five hundred trees were constructed and the exhaustive method was used to adjust the parameters in the random forest.\u0026nbsp;The XGBoost model was set with a decision tree depth of 10, a learning rate of 0.5, and 500 sampling iterations. The SHAP (SHapley Additive exPlanations) method was used for global and local interpretation and visualization of the XGBoost model. Considering the small proportion of RRT patients, ten-fold cross-validation was employed to control for overfitting. The eICU database was used as an external test set to assess model stability. The receiver operating characteristic (ROC) curve, Area Under the Precision-Recall Curve (AUPRC) , and Decision Curve Analysis (DCA) were plotted to evaluate predictive performance and assess accuracy, precision, sensitivity, specificity, recall, and F1 score. This machine learning modeling strategy adhered to the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement recommendations\u003csup\u003e13\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eBased on the inclusion and exclusion criteria set for the study, a total of 22,220 SA-AKI patients were included for analysis. Among them, 18,873 SA-AKI patients were included from the MIMIC database, while 3,347 patients were retrieved from the eICU-CRD database. Specifically, 1,358 patients (6.11%) received RRT, comprising 905 patients from the MIMIC database and 453 from the eICU-CRD database. The detailed screening process for this study is shown in Figure 1.\u003c/p\u003e\n\u003cp\u003eThe univariate analysis of basic demographic data is detailed in Table 1 and in supple table of two database respectively. In this study cohort, a higher proportion of male patients with severe SA-AKI required RRT, and there were racial disparities observed, with Caucasians having a lower incidence of RRT utilization compared to African Americans and Hispanics. Among the suspected sources of infection, bloodstream infection and pulmonary infection were the most common, accounting for 36.7% and 30.3%, respectively, with an overall positive culture rate of 24.3%. The proportion of bloodstream infections in SA-AKI patients receiving RRT was higher (42.2% vs. 36.3%, P\u0026lt;0.001), whereas the proportions of pulmonary and urinary tract infections were lower. Correspondingly, the SOFA scores (6.92 vs. 4.42, P\u0026lt;0.001) and APS scores (78.7 vs. 52.9, P\u0026lt;0.001) were significantly higher. Compared to patients not receiving RRT, mechanical ventilation was significantly more prevalent in the RRT group (69.3% vs. 60.4%, P\u0026lt;0.001). Moreover, the RRT group exhibited more severe metabolic acidosis, including lower PH (7.23 vs. 7.31, P\u0026lt;0.001) , lower buffer excess(-8.25 vs. -3.10, P\u0026lt;0.001), higher lactate levels (5.07 vs. 3.03, P\u0026lt;0.001), more disordered electrolytes, such as serum sodium and serum calcium and poorer coagulation function, including higher INR (2.20 vs. 1.73, P\u0026lt;0.001) and APTT (70.9 vs. 53.4, P\u0026lt;0.001). it was showed by principal component analysis in Supple Figure 2. The univariate analysis of baseline distribution for the MIMIC and eICU databases can be referenced in Supple Table.\u003c/p\u003e\n\u003cp\u003eMultivariate logistic regression models and three machine learning models were constructed and tested, employing ten-fold cross-validation to reduce overfitting. The multivariate logistic regression model showed significant differences for many variables based on principal component analysis. Younger, higher body weight, more severe scoring upon admission to ICU, worse homeostasis and electrolyte balance as well as coagulation function, and mechanical ventilation before the diagnosis, were the high-risk factors of RRT for severe SA-AKI patients as shown in Table 2 and the nomogram in Figure 2. Using the XGBoost algorithm for variable analysis, the six most important parameters were the APS score, SOFA score, minimum calcium level, maximum APTT, maximum BUN, and minimum buffer excess. These parameters all have clinical interpretability and overlaps with the risk factors of the multiple logistic regression model. The importance of the model is detailed in Figure 3. Overall, the XGBoost model outperformed the multivariate logistic regression model and random forest or SVM models. In the internal test of MIMIC database and external test sets from the eICUdatabase, the XGBoost model\u0026apos;s AUROC curves were 0.906 and 0.816, respectively, demonstrating superior accuracy and specificity. Random Forest Model has excellent performance slightly above XGBoost. The AUROC curves for the multivariate logistic regression model were 0.874 and 0.740, respectively. The performance of SVM was similar to the multivariate logistic regression model. Given the relatively low prevalence of patients requiring RRT interventions, we opted to compute the AUPRC and DCA as a measure of model performance. Notwithstanding a discernible decline in AUPRC for both models during the external test phase conducted with the eICU database, the XGBoost model retained a commendably high degree of accuracy. Detailed diagnostic parameters are shown in Table 3, while the AUROC and AUPRC are illustrated in Figure 4. DCA shown in the supple Figure 4.\u003c/p\u003e\n\u003cp\u003eIn addition, the SHAP algorithm was used to visualize the XGBoost model variables, and the SHAP value distribution for these six variables is shown in Supple Figure 3. The distribution heat map for the top 15 important variables in the XGBoost model is shown in the bee-swarm plot in Figure 5. For any given patient, the input variable values can be used to calculate a score, achieving a visualized SHAP value similar to the nomogram, as detailed in the waterfall plot in Figure 6.\u0026nbsp;\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe precision medicine approach for AKI has been a long-standing vision, with a particular emphasis on determining which AKI patients would benefit from RRT\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. Despite the identification of numerous high-risk factors for RRT in AKI patients through multivariate logistic regression models\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e, the gap between these isolated risk factors and the precision required for accurate prediction remains significant. The emergence of machine learning has opened up new avenues for achieving precision in the treatment of AKI patients\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e,\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. The present study demonstrates that machine learning can accurately predict the demand of RRT when patients was diagnosed as severe SA-AKI .\u003c/p\u003e \u003cp\u003eA growing number of studies have investigated the application of artificial intelligence models for AKI diagnosis and early detection, as well as for RRT initiation and weaning \u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. Using local databases and public databases, our research team constructed models for the diagnosis and early detection of severe AKI, as well as RRT weaning models, achieving good performance with both Random Forest and XGBoost\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. Moreover, with the development of SHAP, the explainability of machine learning has greatly improved. Zhiyan Fan et al. have developed a robust prognostic prediction model for SA-AKI patients, which exhibits consistent performance at 7 days, 14 days, and 28 days post-admission, was constructed by integrating data from the MIMIC-IV database with a local database. The model was built using an ensemble of 40 predictive variables, processed through the advanced XGBoost machine learning algorithm\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. Gao et al. corroborated these findings by employing the random forest algorithm to analyze SA-AKI patients sourced from the MIMIC-IV database. Their analysis yielded analogous outcomes, further validating the efficacy of machine learning techniques in predicting the prognosis of SA-AKI patients within the critical care setting\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. Nonetheless, studies on the prediction of RRT in SA-AKI are relatively scarce. Li Zhao et al. collected data from local sepsis patients over three years, including 300 sepsis patients for model construction and validation. The AUROCs for the training and validation cohorts were 0.873 (95% CI 0.825\u0026ndash;0.921) and 0.826 (95% CI 0.727\u0026ndash;0.924), respectively\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. Chun-Fu Lai et al. conducted a prospective observational cohort study, collecting data over 10 years from a total of 1000 local SA-AKI patients requiring RRT. Three sub-phenotypes were constructed using unsupervised clustering and multivariable-adjusted Cox regression models and Fine-Gray sub-distribution hazard models were successfully established, identifying pre-dialysis hyperlactatemia\u0026thinsp;\u0026ge;\u0026thinsp;3.3 mmol/L as an independent outcome predictor\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe utilization rate of RRT for severe SA-AKI remains low, with only 6.3% of patients receiving RRT in the MIMIC and eICU databases. One of the reasons is the exclusion of CKD stages III to V and renal transplant recipients. The second reason is that, based on the findings of recent landmark randomized controlled trials (RCTs), the therapeutic approach to RRT in AKI patients has become more conservative. The IDEAL-ICU study, focusing on patients with septic shock, randomized patients diagnosed with acute kidney injury using RIFLE criteria to immediate initiation of RRT or delayed initiation within 48 hours if needed. All 488 enrolled patients had septic shock, but the study was prematurely halted due to futility, finding no difference in 90-day mortality between the early group (58%) and the delayed strategy group (54%). A meta-analysis by Pan H et al., which included 10 RCTs involving 4753 patients with severe AKI, reported no significant difference in mortality between the early RRT group and the standard group\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. Additionally, 38% of the delayed group did not require RRT as their renal function spontaneously recovered\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. White KA-O et al. conducted a large retrospective study in Australia, including 13,451 SA-AKI patients with 4,051 cases (30%) of KDIGO II/III stages, but only 1,135 patients (8.4%) received RRT\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e. The low rate of RRT initiation suggests that the majority of patients do not require RRT and can recover through conservative treatment. Identifying SA-AKI patients who do not require RRT and allowing for a longer observation period could potentially avoid the high risks and costs associated with RRT\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. The CRTSAKI study is an ongoing large-scale, multicenter, randomized controlled trial related to sepsis-associated AKI, aiming to enroll 460 patients with sepsis-associated AKI (SA-AKI) at KDIGO II/III stages. The protocol stipulates immediate RRT initiation in the early group post-randomization, while the delayed group follows a standard approach similar to the STARRT-AKI study. The primary outcome is overall survival within a 90-day follow-up period (90-day all-cause mortality) (Trial registration: NCT03175328) \u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. This study is expected to provide further insights into the timing of RRT. The current body of evidence from RCTs suggests that the timing of initiating RRT in patients with severe AKI may not significantly influence clinical outcomes. This apparent lack of effect could be attributed to the inclusion of a heterogeneous patient population in these studies, where a significant proportion of patients may not actually require RRT. The dilution of the treatment effect due to the inclusion of unnecessary RRT recipients can lead to a larger negative control group, which in turn reduces the statistical power to detect a true positive effect. To address this issue, the application of AI models offers a promising solution. AI algorithms can be trained to predict which patients are most likely to benefit from RRT, thereby enabling a more targeted approach to clinical trial design. This targeted approach not only enhances the statistical power of the study but also ensures that resources are allocated efficiently, potentially leading to more personalized and effective treatment strategies for patients with severe AKI.\u003c/p\u003e \u003cp\u003eOur study confronted the ubiquitous challenges prevalent in the field of artificial intelligence research, particularly concerning the explainability of models. Despite the advances in the model's decision-making process since the introduction of the SHAP algorithm, machine learning models remain inherently \"black box\" systems. Their inner workings can be opaque in certain scenarios, limiting our confidence in the model's predictive outcomes and the acceptance of its clinical applications. The eICU database was utilized as an external test set to validate the model's generalizability, yet the AUROC dropped from 91% in the training set to 81% in the test set. This discrepancy suggests variations in model performance across different datasets, potentially due to disparities in patient characteristics, healthcare settings, or data collection methods between the two databases, thereby constraining the model's adaptability. The reasons for patients to use RRT are unavailable, and doctors from different centers may use RRT based on different clinical experiences and evidence, resulting in the occurrence of offsets. Lastly, to ensure data quality and analytical accuracy, unmeasured or confounding variables were removed, potentially introducing bias in the exposure-outcome association. Hence, the model may not have fully captured all factors influencing patient outcomes, impacting the accuracy and reliability of the study results. The model's limitations highlight the ongoing need for refinement in balancing predictive power with interpretability, particularly in the context of clinical decision-making.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eOur study constructed a predictive model for screen RRT patients with severe SA-AKI using two-center databases and the XGBoost algorithm. We enhanced model interpretability by employing the SHAP method and externally validated the model using the eICU database.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eRRT: Renal Replacement Therapy\u003c/p\u003e\n\u003cp\u003eSA-AKI: Sepsis-associated Acute Kidney Injury\u003c/p\u003e\n\u003cp\u003eSCR: Serum Creatinine\u003c/p\u003e\n\u003cp\u003eLOS: Length of Stay \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eKDIGO: Kidney Disease: Improving Global Outcomes\u003c/p\u003e\n\u003cp\u003eAUROC: Area Under the Receiver Operating Characteristic\u003c/p\u003e\n\u003cp\u003eAUPRC: Area Under the Precision-Recall Curve\u003c/p\u003e\n\u003cp\u003eTRIPOD：Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis\u003c/p\u003e\n\u003cp\u003ePSM: Propensity Score Matching\u003c/p\u003e\n\u003cp\u003eIPW: Inverse Probability Weighting\u003c/p\u003e\n\u003cp\u003eCCI: Charlson Comorbidity Index;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSOFA: Sequential Organ Failure Assessment\u003c/p\u003e\n\u003cp\u003eAPS: Acute Physiology Score\u003c/p\u003e\n\u003cp\u003eBMI: Body Mass Index\u003c/p\u003e\n\u003cp\u003eDCA: Decision Curve Analysis\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eClinical trial number:\u003c/strong\u003e not applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate:\u003c/strong\u003e This study is a database retrospective study, no private data was used, and no ethical approval was required. The two public databases have passed the ethics of their respective research institutions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication:\u003c/strong\u003e Not applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials:\u003c/strong\u003e The datasets generated and/or analyzed during the current study are available in the MIMIC and eICU-CRD databases.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests:\u003c/strong\u003e None.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e: This work was supported by grants from the peak supporting clinical discipline of Shanghai health bureau (2023ZDFC0104 to L.T)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions: \u003c/strong\u003eQiqiang Liang designed the research scheme, analyzed the characteristics of the data, constructed the prediction model, and drafted the manuscript. Sumian Zhang extracted and collated clinical data through the database, built models, and collected prospective research data. Haiyan Ye and Mei yang determined clinical variables, analyzed the importance of model variables, and completed clinical data in prospective studies. Xuebin Wang supervised the research process, provided clinical reference, and revised the paper. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements:\u003c/strong\u003e We thank the Ascetic Practitioners in Critical Care (APCC) team, and the easy Data Science for Medicine (easyDSM) team for sharing their knowledge and codes in big data of critical care, along with the cross-platform Big Data Master of Critical Care (BDMCC) software (https://github.com/ningyile/BDMCC_APP). We especially appreciate the MIMIC and eICU-CRD official team\u0026apos;s efforts to open-source the database and codes.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003ePoston JT, Koyner JL. Sepsis associated acute kidney injury. \u003cem\u003eBmj. \u003c/em\u003e2019;364:k4891.\u003c/li\u003e\n\u003cli\u003eSinger M, Deutschman CS, Seymour CW, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). \u003cem\u003eJama. \u003c/em\u003e2016;315(8):801-810.\u003c/li\u003e\n\u003cli\u003eZarbock A, Nadim MK, Pickkers P, et al. Sepsis-associated acute kidney injury: consensus report of the 28th Acute Disease Quality Initiative workgroup. \u003cem\u003eNature Reviews Nephrology. \u003c/em\u003e2023;19(6):401-417.\u003c/li\u003e\n\u003cli\u003eKhwaja A. KDIGO clinical practice guidelines for acute kidney injury. \u003cem\u003eNephron Clin Pract. \u003c/em\u003e2012;120(4):c179-184.\u003c/li\u003e\n\u003cli\u003ePais T, Jorge S, Lopes JA. Acute Kidney Injury in Sepsis. \u003cem\u003eInternational journal of molecular sciences. \u003c/em\u003e2024;25(11).\u003c/li\u003e\n\u003cli\u003eZarbock A, Kellum JA, Schmidt C, et al. Effect of Early vs Delayed Initiation of Renal Replacement Therapy on Mortality in Critically Ill Patients With Acute Kidney Injury: The ELAIN Randomized Clinical Trial. \u003cem\u003eJAMA. \u003c/em\u003e(315(20):2190-9).\u003c/li\u003e\n\u003cli\u003eBagshaw SM, Wald R, Adhikari NKJ, et al. Timing of Initiation of Renal-Replacement Therapy in Acute Kidney Injury. \u003cem\u003eN Engl J Med. \u003c/em\u003e2020;383(3):240-251.\u003c/li\u003e\n\u003cli\u003eBarbar SD, Clere-Jehl R, Bourredjem A, et al. Timing of Renal-Replacement Therapy in Patients with Acute Kidney Injury and Sepsis. \u003cem\u003eN Engl J Med. \u003c/em\u003e2018;379(15):1431-1442.\u003c/li\u003e\n\u003cli\u003eLi G, Li B, Song B, et al. Uplift modeling to predict individual treatment effects of renal replacement therapy in sepsis-associated acute kidney injury patients. (2045-2322 (Electronic)).\u003c/li\u003e\n\u003cli\u003ePalmowski L, Lindau S, Henk LC, et al. Predictive enrichment for the need of renal replacement in sepsis-associated acute kidney injury: combination of furosemide stress test and urinary biomarkers TIMP-2 and IGFBP-7. (2110-5820 (Print)).\u003c/li\u003e\n\u003cli\u003eLiang Q, Xu Y, Zhou Y, Chen X, Chen J, Huang M. Severe acute kidney injury predicting model based on transcontinental databases: a single-centre prospective study. \u003cem\u003eBMJ Open. \u003c/em\u003e2022;12(3):e054092.\u003c/li\u003e\n\u003cli\u003eLiang Q, Ding S, Chen J, et al. Prediction of carbapenem-resistant gram-negative bacterial bloodstream infection in intensive care unit based on machine learning. \u003cem\u003eBMC medical informatics and decision making. \u003c/em\u003e2024;24(1):123.\u003c/li\u003e\n\u003cli\u003eMoons Kg Fau - Altman DG, Altman Dg Fau - Reitsma JB, Reitsma Jb Fau - Ioannidis JPA, et al. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration. (1539-3704 (Electronic)).\u003c/li\u003e\n\u003cli\u003eMeersch M, Mayerh\u0026ouml;fer T, Joannidis M. Acute kidney injury subphenotyping and personalized medicine. \u003cem\u003eCurr Opin Crit Care. \u003c/em\u003e2024;30(6):555-562.\u003c/li\u003e\n\u003cli\u003eChen JJ, Chang CH, Huang YT, Kuo G. Furosemide stress test as a predictive marker of acute kidney injury progression or renal replacement therapy: a systemic review and meta-analysis. \u003cem\u003eCrit Care. \u003c/em\u003e2020;24(1):202.\u003c/li\u003e\n\u003cli\u003eRomagnoli S, Clark WR, Ricci Z, Ronco C. Renal replacement therapy for AKI: When? How much? When to stop? \u003cem\u003eBest practice \u0026amp; research. Clinical anaesthesiology. \u003c/em\u003e2017;31(3):371-385.\u003c/li\u003e\n\u003cli\u003eWilson FP, Greenberg JH. Acute Kidney Injury in Real Time: Prediction, Alerts, and Clinical Decision Support. \u003cem\u003eNephron. \u003c/em\u003e2018;140(2):116-119.\u003c/li\u003e\n\u003cli\u003eGottlieb ER, Samuel M, Bonventre JV, Celi LA, Mattie H. Machine Learning for Acute Kidney Injury Prediction in the Intensive Care Unit. \u003cem\u003eAdvances in Chronic Kidney Disease. \u003c/em\u003e2022;29(5):431-438.\u003c/li\u003e\n\u003cli\u003eCheungpasitporn W, Thongprayoon C, Kashani KB. Advances in critical care nephrology through artificial intelligence. (1531-7072 (Electronic)).\u003c/li\u003e\n\u003cli\u003eLiang Q, Xu X, Ding S, Wu J, Huang M. Prediction of successful weaning from renal replacement therapy in critically ill patients based on machine learning. \u003cem\u003eRen Fail. \u003c/em\u003e2024;46(1):2319329.\u003c/li\u003e\n\u003cli\u003eGao T, Nong Z, Luo Y, et al. Machine learning-based prediction of in-hospital mortality for critically ill patients with sepsis-associated acute kidney injury. (1525-6049 (Electronic)).\u003c/li\u003e\n\u003cli\u003eFan Z, Jiang J, Xiao C, et al. Construction and validation of prognostic models in critically Ill patients with sepsis-associated acute kidney injury: interpretable machine learning approach. (1479-5876 (Electronic)).\u003c/li\u003e\n\u003cli\u003eLi X, Liu C, Mao Z, Li Q, Zhou FA-OX. Timing of renal replacement therapy initiation for acute kidney injury in critically ill patients: a systematic review of randomized clinical trials with meta-analysis and trial sequential analysis. (1466-609X (Electronic)).\u003c/li\u003e\n\u003cli\u003eLai CA-O, Liu JH, Tseng LJ, et al. Unsupervised clustering identifies sub-phenotypes and reveals novel outcome predictors in patients with dialysis-requiring sepsis-associated acute kidney injury. (1365-2060 (Electronic)).\u003c/li\u003e\n\u003cli\u003ePan HC, Chen YY, Tsai IJ, et al. Accelerated versus standard initiation of renal replacement therapy for critically ill patients with acute kidney injury: a systematic review and meta-analysis of RCT studies. (1466-609X (Electronic)).\u003c/li\u003e\n\u003cli\u003eWhite KA-O, Serpa-Neto A, Hurford R, et al. Sepsis-associated acute kidney injury in the intensive care unit: incidence, patient characteristics, timing, trajectory, treatment, and associated outcomes. A multicenter, observational study. \u003cem\u003eIntensive Care Med. \u003c/em\u003e(1432-1238 (Electronic)).\u003c/li\u003e\n\u003cli\u003eChen WA-O, Cai LH, Zhang ZH, et al. The timing of continuous renal replacement therapy initiation in sepsis-associated acute kidney injury in the intensive care unit: the CRTSAKI Study (Continuous RRT Timing in Sepsis-associated AKI in ICU): study protocol for a multicentre, randomised controlled trial. (2044-6055 (Electronic)).\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1: Basic demographic characteristics of severe acute kidney injury associated with sepsis\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"573\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo RRT, N=20862\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRRT, N=1358\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eP\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGender (Male)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e11299 (54.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e796 (58.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.00155\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 573px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRace\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eWhite\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e14964 (71.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e885 (65.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBlack\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e1659 (8.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e126 (9.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.0909\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLatin\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e668 (3.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e56 (4.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.0759\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOthers\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e3571 (17.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e291 (21.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 573px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSuspicious infection site\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBlood\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e7571 (36.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e573 (42.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLung\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e6409 (30.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e318 (23.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eUrine\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e4571 (21.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e199 (14.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOthers\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e2042 (9.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e200 (14.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePostive Culture\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e5149 (24.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e235 (17.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 573px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eComorbidity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCirrhosis\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e1425 (6.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e249 (18.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eChronic heart failure\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e5578 (26.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e285 (21.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMyocardial infarction\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e3068 (14.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e178 (13.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.115\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eChronic structural lung disease\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e5143 (24.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e249 (18.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMalignant tumour\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e2626 (12.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e126 (9.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDiabetes mellitus\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e5507 (26.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e383 (28.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.153\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCCI (mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e4.76 (2.56)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e4.28 (2.40)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSOFA score (mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e4.42 (2.82)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e6.92 (3.87)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAPS (mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e52.9 (22.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e78.7 (26.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTime of admission ICU to AKI (minutes, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e4220 (10900)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e5440 (11100)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAge (Y, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e64.1 (16.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e58.4 (15.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHeight (cm, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e169 (10.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e170 (10.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eWeight (Kg, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e84.6 (25.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e92.1 (28.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBMI (mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e29.5 (8.29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e31.7 (8.95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eWBCmin (*10^9, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e11.3 (7.69)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e14.0 (11.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eWBCmax (*10^9, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e15.7 (9.60)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e16.8 (14.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.00373\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHbmin (g/l, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e10.0 (2.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e9.55 (2.52)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePltmax (*10^9, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e193 (116)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e156 (114)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHCTmin (%, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e30.1 (7.01)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e29.0 (7.79)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNemax (%, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e80.5 (13.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e79.8 (15.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.116\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePHmax (mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e7.41 (0.0814)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e7.36 (0.116)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePHmin (mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e7.31 (0.109)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e7.23 (0.135)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePaO2min (mmHg, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e88.3 (59.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e79.0 (52.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePaCO2max (mmHg, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e48.6 (14.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e48.1 (16.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.215\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBEmax (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e0.678 (5.31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e-3.05 (6.89)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBEmin (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e-3.10 (5.78)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e-8.25 (7.12)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHCO3max (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e25.3 (5.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e21.6 (6.28)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHCO3min (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e21.7 (5.13)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e17.8 (5.79)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLactatemax (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e3.03 (2.39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e5.07 (4.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eKmax (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e4.44 (0.950)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e4.85 (1.13)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCLmax (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e107 (7.05)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e105 (8.31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCLmin (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e101 (6.81)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e99.7 (7.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNAmax (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e140 (6.01)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e139 (7.07)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNAmin (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e136 (5.85)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e135 (6.48)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCamin (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e1.04 (0.119)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e0.89 (0.140)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGlu (mg/dL, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e187 (114)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e206 (130)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eALP (U/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e128 (146)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e153 (174)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTbil (mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e1.91 (4.23)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e5.00 (9.10)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eALTmax (U/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e145 (615)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e401 (1180)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eASTmax (U/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e215 (866)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e745 (1960)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBUNmax (mg/dL, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e31.7 (23.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e53.6 (36.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eINRmax (mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e1.73 (1.42)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e2.20 (1.64)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAPTTmax (s, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e53.4 (37.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e70.9 (44.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMechanical ventilation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e12598 (60.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e941 (69.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eICU death\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e3589 (17.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e589 (43.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 198px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAll death\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e8902 (42.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e749 (55.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eAbbreviation:\u0026nbsp;\u003c/strong\u003eCCI: Charlson Comorbidity Index; Hb: Hemoglobin; SOFA: Sequential Organ Failure Assessment; APS: Acute Physiology Score; BMI: Body Mass Index; Plt: Platelet; HCT: Hematocrit; Ne: Neutrophil Proportion; BE: Buffuer excess; K: Potassium; Cl: Chlorine; NA: Natrium; Ca: Calcium; Glu: Glucose; ALP: Alkaline Phosphatase; Tbil: Total Bilirubin; ALT: Alanine Aminotransferase; AST: Aspertate Aminotransferase; BUN: Blood Urea Nitrogen; INR: International Normalized Ratio for Prothrombin Time; APTT: Activated Partial Prothrombin Time.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2: Parameters in the multivariable logistic regression model\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"587\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eValue\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eP\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e2.5% CI\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e97.5% CI\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAge (Y, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eWeight (Kg, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.41\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSOFA score (mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.63\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAPS \u0026nbsp;(mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e2.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e2.35\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHb (g/l, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHCO3max (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.44\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eKmax (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.36\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCLmax (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.74\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCLmin (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNAmax (mmol/L, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.68\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAPTT (s, mean, SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.62\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMechanical ventilation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e1.63\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eAbbreviation:\u003c/strong\u003e SOFA: Sequential Organ Failure Assessment; APS: Acute Physiology Score; Hb: Hemoglobin; K: Potassium; Cl: Chlorine; NA: Natrium; Ca: Calcium;Glu: Glucose; APTT: Activated Partial Prothrombin Time.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3: Model prediction performance evaluation for RRT in severe SA-AKI patients\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" align=\"\" width=\"709\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAUROC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e95%CI\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSensitivity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSpecificity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePPV\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNPV\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eYouden\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eGlm\u003c/p\u003e\n \u003cp\u003eInternal test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.874\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e83.84%-90.91%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e75.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e87.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e75.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e14.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e99.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e14.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e87.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e62.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003cp\u003eInternal test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.906\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e87.72%-93.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e79.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e90.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e78.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e17.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e99.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e17.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e90.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e69.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003cp\u003eInternal test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.905\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e86.54%-94.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e82.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e67.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e83.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e15.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e98.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e15.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e67.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e0.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e50.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003cp\u003eInternal test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.862\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e82.53%-92.47%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e79.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e83.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e79.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e15.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e99.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e15.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e83.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e62.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eGlm\u003c/p\u003e\n \u003cp\u003eExternal test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.740\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e71.45%-76.64%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e77.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e56.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e81.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e32.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e92.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e32.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e56.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e37.6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003cp\u003eExternal test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.816\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e79.63%-83.65%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e69.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e78.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e68.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e27.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e95.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e27.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e78.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e47.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003cp\u003eExternal test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.814\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e78.54%-83.93%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e87.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e72.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e67.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e26.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e87.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e26.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e72.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e40\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003cp\u003eExternal test\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.775\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e74.18%-82.39%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e78.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e57.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e83.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e28.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e95.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e28.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 41px;\"\u003e\n \u003cp\u003e57.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 50px;\"\u003e\n \u003cp\u003e40.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Sepsis-associated acute kidney injury, Renal Replacement Therapy, Two-center databases, Machine learning algorithms","lastPublishedDoi":"10.21203/rs.3.rs-6321504/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6321504/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground\u003c/strong\u003e: The high risk of renal replacement therapy (RRT) in patients with sepsis-associated acute kidney injury (SA-AKI) remains unclear.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e: This two-center retrospective study analyzed data from the MIMIC database and the eICU database. SA-AKI patients were included, and demographic data and laboratory parameters from the 5 days before diagnosis were collected as variables. The MIMIC database served as the training and validation set for developing models using multivariate logistic regression, random forest, SVM and XGBoost. Moreover, the eICU database was used as an external test set. We visualized the model using nomograms and the SHAP method, and evaluated model performance through variety of methods.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e: A total of 22,220 patients with severe SA-AKI were included in the analysis, with 1,358 (6.11%) receiving RRT. The RRT group exhibited more severe metabolic acidosis, including lower PH (7.23 vs. 7.31, P\u0026lt;0.001) , lower buffer excess(-8.25 vs. -3.10, P\u0026lt;0.001), higher lactate levels (5.07 vs. 3.03, P\u0026lt;0.001), more disordered electrolytes and poorer coagulation function. The six most important parameters were the APS score, SOFA score, minimum calcium level, maximum APTT, maximum BUN, and minimum buffer excess in the XGBoost model with great performance. The Area Under the Receiver Operating Characteristic Curve for the internal and external test and sets were 0.906 and 0.816, respectively. In the meanwhile, XGBoost has the best Area Under the Precision-Recall Curve。\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion\u003c/strong\u003e: Our study constructed a predictive model for initiating RRT in critically ill patients with severe SA-AKI using twocenter databases and the machine learning algorithms.\u003c/p\u003e","manuscriptTitle":"Machine Learning for Renal Replacement Therapy in Patients with Severe Acute Kidney Injury Associated with Sepsis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-08 14:29:00","doi":"10.21203/rs.3.rs-6321504/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bff457d5-9786-46a2-9c13-eec6640b36eb","owner":[],"postedDate":"May 8th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-07-03T07:38:58+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-08 14:29:00","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6321504","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6321504","identity":"rs-6321504","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}